rickub
bots-garden/mini-mepublic Fork
⑂ 86f68f8
Branches
feature/acp-client-tools feature/acp-create feature/agents-file feature/ask-agent-hooks feature/ask-cli feature/decider-no-tools feature/declared-tools feature/display-called-hooks feature/dockerinit-account feature/fetch-build-in feature/genkit-v1.13.1 feature/github-mcp-in-sbx feature/go-1.27.3 feature/hooks feature/improve-init feature/interactive-prompt feature/last-question feature/llamacpp-fallback feature/llmman-provider feature/log-errors feature/lsp-builtin feature/mcp-gateway-session feature/mcp-support feature/memory-search feature/mini-me-as-a-sdk feature/mini-mes feature/mm-init feature/open-terminal feature/prompt-flag feature/response-request feature/resume-acp feature/sbx-template-kit feature/search-in-doc feature/selectors feature/skill-discovery feature/structured-output feature/system-one feature/talk-to-another-mm feature/tasks-builtin feature/try-again feature/unload-model feature/wasm-custom-tools feature/yolo feature/zed-skill-selector fix/generation-timeout fix/plan-mode fix/too-long-completion fix/tool-order fix/word-wrap improve-mem-usage main thinking
Tags
v0.0.0 v0.0.1 v0.0.2 v0.0.3 v0.0.4 v0.1.0 v0.1.1 v0.1.2 v0.1.3 v0.1.4 v0.1.5 v0.1.6 v0.1.7 v0.1.8 v0.1.9 v0.2.0 v0.2.1 v0.2.2 v0.2.3 v0.2.4 v0.2.5 v0.2.6 v0.3.0 v0.4.0 v0.5.0 v0.6.0 v0.7.0 v0.8.0 v0.9.0 v1.0.0 v1.0.1 v1.0.2 v1.0.3 v1.0.4 v1.0.5 v1.0.6 v1.0.7 v1.0.8 v1.0.9 v1.1.0 View all branches → View all tags →
Commits
⬇ Clone ▾
git clone https://git.rickub.com/bots-garden/mini-me.git
git clone ssh://git@rickub.com/bots-garden/mini-me.git

Host key fingerprint (ed25519): SHA256:iycHnxEyq0Q7uyVpB7JlznP0G7JrTPXLYRcAU5CSLhc — verify it before your first connect.

Download ZIP

🛟 Updated.

k33g committed 2026-11-14T19:49:40+01:00 Browse files
86f68f8 parent: 500dbbd
modified .tickets/issues/0211-check-if-llamacpp-1-0-0-is-working.yaml +2 -2
@@ -6,9 +6,9 @@ author:
66 name: k33g
77 email: ph.charriere@gmail.com
88 createdAt: 2026-11-14T05:56:02.153Z
9-updatedAt: 2026-11-14T06:35:10.000Z
9+updatedAt: 2026-11-14T18:45:26.000Z
1010 labels: []
11-milestone: v1.1.0
11+milestone: v2.0.0
1212 epic: Providers (LLM Engine)
1313 body: ""
1414 tasks: []
@@ -6,9 +6,9 @@ author:
6 name: k33g6 name: k33g
7 email: ph.charriere@gmail.com7 email: ph.charriere@gmail.com
8 createdAt: 2026-11-14T05:56:02.153Z8 createdAt: 2026-11-14T05:56:02.153Z
9-updatedAt: 2026-11-14T06:35:10.000Z9+updatedAt: 2026-11-14T18:45:26.000Z
10 labels: []10 labels: []
11-milestone: v1.1.011+milestone: v2.0.0
12 epic: Providers (LLM Engine)12 epic: Providers (LLM Engine)
13 body: ""13 body: ""
14 tasks: []14 tasks: []
modified .tickets/issues/0215-mm-acp-create-start-test-every-provider-address.yaml +34 -5
@@ -1,12 +1,12 @@
11 # Managed by IssueSpec. Hand edits are welcome; keep the schema valid.
22 id: 215
33 title: "mm acp-create (/start): test every provider - address used and model list"
4-state: open
4+state: closed
55 author:
66 name: k33g
77 email: ph.charriere@gmail.com
88 createdAt: 2026-11-14T07:41:16.000Z
9-updatedAt: 2026-11-14T08:07:53.000Z
9+updatedAt: 2026-11-14T18:45:13.000Z
1010 labels:
1111 - kind::chore
1212 - To be tested
@@ -79,12 +79,14 @@ tasks:
7979 closedAt: 2026-11-14T07:47:59.000Z
8080 - id: 3
8181 title: llamacpp - llama-server on 127.0.0.1:8080
82- state: open
82+ state: closed
8383 priority: medium
8484 author:
8585 name: k33g
8686 email: ph.charriere@gmail.com
8787 createdAt: 2026-11-14T07:41:16.000Z
88+ updatedAt: 2026-11-14T18:24:44.000Z
89+ closedAt: 2026-11-14T18:24:44.000Z
8890 - id: 4
8991 title: kronk
9092 state: closed
@@ -107,20 +109,24 @@ tasks:
107109 closedAt: 2026-11-14T08:07:53.000Z
108110 - id: 6
109111 title: Same checks with mm in a container or a Docker Sandbox (the fallback must be taken there)
110- state: open
112+ state: closed
111113 priority: medium
112114 author:
113115 name: k33g
114116 email: ph.charriere@gmail.com
115117 createdAt: 2026-11-14T07:41:16.000Z
118+ updatedAt: 2026-11-14T18:45:13.000Z
119+ closedAt: 2026-11-14T18:45:13.000Z
116120 - id: 7
117121 title: Write the results in a comment (version, client, address used, list ok or not); a ticket per bug found
118- state: open
122+ state: closed
119123 priority: low
120124 author:
121125 name: k33g
122126 email: ph.charriere@gmail.com
123127 createdAt: 2026-11-14T07:41:16.000Z
128+ updatedAt: 2026-11-14T18:45:13.000Z
129+ closedAt: 2026-11-14T18:45:13.000Z
124130 comments:
125131 - id: 1
126132 author:
@@ -148,3 +154,26 @@ comments:
148154 body: |
149155 User (2026-11-14): **openai works** too. Still to try: llamacpp, and
150156 inside a container / sandbox.
157+ - id: 4
158+ author:
159+ name: k33g
160+ email: ph.charriere@gmail.com
161+ createdAt: 2026-11-14T18:24:44.000Z
162+ body: |
163+ User (2026-11-14): **llamacpp works** too. Every provider now checked
164+ on the host: dmr, llmman, kronk, openai, llamacpp, and ollama after
165+ the fix of #214. Still to try: inside a container / Docker Sandbox
166+ (task 6).
167+ - id: 5
168+ author:
169+ name: k33g
170+ email: ph.charriere@gmail.com
171+ createdAt: 2026-11-14T18:45:13.000Z
172+ body: |
173+ Closed at the user's request. Summary: on the host, every provider
174+ lists its models in `/start` (dmr, llmman, kronk, openai, llamacpp;
175+ ollama after #214). In a container, the fallback works for ollama and
176+ llmman; llamacpp had none — added by #217 (on a branch, not in a
177+ release yet: a container built by `mm dockerinit` runs the latest
178+ release, v1.0.9, which does not have it). Docker Sandbox not tried.
179+
@@ -1,12 +1,12 @@
1 # Managed by IssueSpec. Hand edits are welcome; keep the schema valid.1 # Managed by IssueSpec. Hand edits are welcome; keep the schema valid.
2 id: 2152 id: 215
3 title: "mm acp-create (/start): test every provider - address used and model list"3 title: "mm acp-create (/start): test every provider - address used and model list"
4-state: open4+state: closed
5 author:5 author:
6 name: k33g6 name: k33g
7 email: ph.charriere@gmail.com7 email: ph.charriere@gmail.com
8 createdAt: 2026-11-14T07:41:16.000Z8 createdAt: 2026-11-14T07:41:16.000Z
9-updatedAt: 2026-11-14T08:07:53.000Z9+updatedAt: 2026-11-14T18:45:13.000Z
10 labels:10 labels:
11 - kind::chore11 - kind::chore
12 - To be tested12 - To be tested
@@ -79,12 +79,14 @@ tasks:
79 closedAt: 2026-11-14T07:47:59.000Z79 closedAt: 2026-11-14T07:47:59.000Z
80 - id: 380 - id: 3
81 title: llamacpp - llama-server on 127.0.0.1:808081 title: llamacpp - llama-server on 127.0.0.1:8080
82- state: open82+ state: closed
83 priority: medium83 priority: medium
84 author:84 author:
85 name: k33g85 name: k33g
86 email: ph.charriere@gmail.com86 email: ph.charriere@gmail.com
87 createdAt: 2026-11-14T07:41:16.000Z87 createdAt: 2026-11-14T07:41:16.000Z
88+ updatedAt: 2026-11-14T18:24:44.000Z
89+ closedAt: 2026-11-14T18:24:44.000Z
88 - id: 490 - id: 4
89 title: kronk91 title: kronk
90 state: closed92 state: closed
@@ -107,20 +109,24 @@ tasks:
107 closedAt: 2026-11-14T08:07:53.000Z109 closedAt: 2026-11-14T08:07:53.000Z
108 - id: 6110 - id: 6
109 title: Same checks with mm in a container or a Docker Sandbox (the fallback must be taken there)111 title: Same checks with mm in a container or a Docker Sandbox (the fallback must be taken there)
110- state: open112+ state: closed
111 priority: medium113 priority: medium
112 author:114 author:
113 name: k33g115 name: k33g
114 email: ph.charriere@gmail.com116 email: ph.charriere@gmail.com
115 createdAt: 2026-11-14T07:41:16.000Z117 createdAt: 2026-11-14T07:41:16.000Z
118+ updatedAt: 2026-11-14T18:45:13.000Z
119+ closedAt: 2026-11-14T18:45:13.000Z
116 - id: 7120 - id: 7
117 title: Write the results in a comment (version, client, address used, list ok or not); a ticket per bug found121 title: Write the results in a comment (version, client, address used, list ok or not); a ticket per bug found
118- state: open122+ state: closed
119 priority: low123 priority: low
120 author:124 author:
121 name: k33g125 name: k33g
122 email: ph.charriere@gmail.com126 email: ph.charriere@gmail.com
123 createdAt: 2026-11-14T07:41:16.000Z127 createdAt: 2026-11-14T07:41:16.000Z
128+ updatedAt: 2026-11-14T18:45:13.000Z
129+ closedAt: 2026-11-14T18:45:13.000Z
124 comments:130 comments:
125 - id: 1131 - id: 1
126 author:132 author:
@@ -148,3 +154,26 @@ comments:
148 body: |154 body: |
149 User (2026-11-14): **openai works** too. Still to try: llamacpp, and155 User (2026-11-14): **openai works** too. Still to try: llamacpp, and
150 inside a container / sandbox.156 inside a container / sandbox.
157+ - id: 4
158+ author:
159+ name: k33g
160+ email: ph.charriere@gmail.com
161+ createdAt: 2026-11-14T18:24:44.000Z
162+ body: |
163+ User (2026-11-14): **llamacpp works** too. Every provider now checked
164+ on the host: dmr, llmman, kronk, openai, llamacpp, and ollama after
165+ the fix of #214. Still to try: inside a container / Docker Sandbox
166+ (task 6).
167+ - id: 5
168+ author:
169+ name: k33g
170+ email: ph.charriere@gmail.com
171+ createdAt: 2026-11-14T18:45:13.000Z
172+ body: |
173+ Closed at the user's request. Summary: on the host, every provider
174+ lists its models in `/start` (dmr, llmman, kronk, openai, llamacpp;
175+ ollama after #214). In a container, the fallback works for ollama and
176+ llmman; llamacpp had none — added by #217 (on a branch, not in a
177+ release yet: a container built by `mm dockerinit` runs the latest
178+ release, v1.0.9, which does not have it). Docker Sandbox not tried.
179+
added .tickets/issues/0217-llamacpp-fall-back-on-host-docker-internal-8080.yaml +154 -0
new file mode 100644
@@ -0,0 +1,154 @@
1+# Managed by IssueSpec. Hand edits are welcome; keep the schema valid.
2+id: 217
3+title: "llamacpp: fall back on host.docker.internal:8080 like the other local providers"
4+state: closed
5+author:
6+ name: k33g
7+ email: ph.charriere@gmail.com
8+createdAt: 2026-11-14T18:36:24.000Z
9+updatedAt: 2026-11-14T18:45:13.000Z
10+labels:
11+ - enhancement
12+ - kind::feature
13+milestone: v1.1.0
14+epic: Providers (LLM Engine)
15+related:
16+ - 215
17+body: |
18+ ## Seen (user, 2026-11-14, while testing #215)
19+
20+ With mm in a container, the fallback address works for **ollama** and
21+ **llmman**, not for **llamacpp**.
22+
23+ ## Why (code read)
24+
25+ 1. **llamacpp declares no fallback** ([internal/engine/openai_compat.go](internal/engine/openai_compat.go)):
26+ dmr, ollama, llmman and kronk fall back on
27+ `http://host.docker.internal:<port>/...` when their default does not
28+ answer; llamacpp only has `http://127.0.0.1:8080/v1`. In a container,
29+ that is the container itself, and nothing else is tried. The container
30+ guide says so ("set `baseUrl: http://host.docker.internal:8080/v1`").
31+ The shipped `agent.llamacpp.yaml` even opts out on purpose
32+ (`fallback: ""`, "llama-server is wherever you started it").
33+ 2. **llama-server listens on `127.0.0.1` by default.** On Linux and in a
34+ Docker Sandbox, `host.docker.internal` is the host's gateway, not its
35+ loopback: start it with `--host 0.0.0.0` (as Kronk needs
36+ `--api-host 0.0.0.0:11435`). On Docker Desktop (Mac) the loopback is
37+ reached anyway (Ollama on 127.0.0.1 works there) — not checked for
38+ llama-server.
39+ 3. **Docker Sandbox: port 8080 is not in the mini-me kit's network
40+ policy** (12434, 11434, 17434 only): `sbx policy allow network
41+ host.docker.internal:8080`, or a change in the kit (other repository,
42+ mini-me-kit) — out of scope here.
43+
44+ ## Change (agreed with the user)
45+
46+ - llamacpp's fallback: `http://host.docker.internal:8080/v1`, the same
47+ rule as the others (taken when the default does not answer `/models`
48+ within 2 s; `fallback: ""` in a file still turns it off).
49+ - Its start hint mentions `--host 0.0.0.0` for mm in a container or a
50+ sandbox.
51+ - `agent.llamacpp.yaml` drops `fallback: ""`, so that a copy of it gets
52+ the fallback. Re-litigates that file's explicit choice — the user asked
53+ for llamacpp to behave like the others. `config.samples/` and
54+ `demos/struggle-signals/` (pinned to a fake server) are left alone.
55+ - `mm init` writes the fallback in the file's comment, as for the others
56+ (ticket 121); the wizard's address choice (#214) now also tries it.
57+
58+ ## Acceptance criteria
59+
60+ - mm in a container, no `baseUrl`, llama-server on the host started with
61+ `--host 0.0.0.0` → mm reaches it through `host.docker.internal:8080`.
62+ - On the host, nothing changes when llama-server answers on 127.0.0.1.
63+ - Tests and docs EN+FR.
64+tasks:
65+ - id: 1
66+ title: Engine - llamacpp falls back on http://host.docker.internal:8080/v1, like the other local providers
67+ state: closed
68+ priority: high
69+ author:
70+ name: k33g
71+ email: ph.charriere@gmail.com
72+ createdAt: 2026-11-14T18:36:24.000Z
73+ updatedAt: 2026-11-14T18:39:34.000Z
74+ closedAt: 2026-11-14T18:39:34.000Z
75+ - id: 2
76+ title: startHint - say --host 0.0.0.0 for mm in a container or a sandbox
77+ state: closed
78+ priority: medium
79+ author:
80+ name: k33g
81+ email: ph.charriere@gmail.com
82+ createdAt: 2026-11-14T18:36:24.000Z
83+ updatedAt: 2026-11-14T18:39:34.000Z
84+ closedAt: 2026-11-14T18:39:34.000Z
85+ - id: 3
86+ title: agent.llamacpp.yaml - drop fallback empty (it opted out on purpose)
87+ state: closed
88+ priority: medium
89+ author:
90+ name: k33g
91+ email: ph.charriere@gmail.com
92+ createdAt: 2026-11-14T18:36:24.000Z
93+ updatedAt: 2026-11-14T18:39:34.000Z
94+ closedAt: 2026-11-14T18:39:34.000Z
95+ - id: 4
96+ title: Tests - provider resolution, mm init's comment, the wizard's address choice
97+ state: closed
98+ priority: medium
99+ author:
100+ name: k33g
101+ email: ph.charriere@gmail.com
102+ createdAt: 2026-11-14T18:36:24.000Z
103+ updatedAt: 2026-11-14T18:39:34.000Z
104+ closedAt: 2026-11-14T18:39:34.000Z
105+ - id: 5
106+ title: Docs EN+FR - configuration (providers table), run-in-a-container, run-with-llama-cpp, sandbox (port 8080 policy)
107+ state: closed
108+ priority: medium
109+ author:
110+ name: k33g
111+ email: ph.charriere@gmail.com
112+ createdAt: 2026-11-14T18:36:24.000Z
113+ updatedAt: 2026-11-14T18:39:34.000Z
114+ closedAt: 2026-11-14T18:39:34.000Z
115+comments:
116+ - id: 1
117+ author:
118+ name: k33g
119+ email: ph.charriere@gmail.com
120+ createdAt: 2026-11-14T18:39:34.000Z
121+ body: |
122+ Implemented on `feature/llamacpp-fallback` (uncommitted).
123+
124+ - Engine: llamacpp's fallback `http://host.docker.internal:8080/v1`;
125+ start hint `… --port 8080 (add --host 0.0.0.0 when mm runs in a
126+ container or a sandbox)`.
127+ - `agent.llamacpp.yaml`: `fallback: ""` removed, the comment says what
128+ the fallback is and how to turn it off. `config.samples/` and
129+ `demos/struggle-signals/` keep their `fallback: ""`.
130+ - Tests updated where they asserted "no fallback"
131+ (`TestResolveURLPrecedence`: falls back when baseUrl is down, keeps
132+ it when up; `TestConfigCompat` pins `fallback: ""`;
133+ `TestDefaultURLs`; `TestRenderRecordsTheDefaultURLs`: mm init now
134+ writes the fallback comment for llamacpp). Whole suite green.
135+ - Docs EN+FR: configuration (providers table, agent.llamacpp.yaml),
136+ mcp-configuration (EN), cli (warning example), run-in-a-container,
137+ run-with-llama-cpp (container / sandbox variant), sandbox guide.
138+
139+ Side effect, same as dmr/ollama/llmman already: on the host with
140+ llama-server down, the start-up warning names the fallback address
141+ (`nothing answers at http://host.docker.internal:8080/v1`).
142+
143+ Not tried with a real llama-server in a container or a sandbox.
144+ - id: 2
145+ author:
146+ name: k33g
147+ email: ph.charriere@gmail.com
148+ createdAt: 2026-11-14T18:45:13.000Z
149+ body: |
150+ Closed at the user's request. mm init writes the fallback comment for
151+ llamacpp (checked by the user). In a container, checked only with
152+ v1.0.9 (the released binary the image downloads), which does not have
153+ the fallback yet: the fix reaches containers with the next release.
154+
new file mode 100644
@@ -0,0 +1,154 @@
1+# Managed by IssueSpec. Hand edits are welcome; keep the schema valid.
2+id: 217
3+title: "llamacpp: fall back on host.docker.internal:8080 like the other local providers"
4+state: closed
5+author:
6+ name: k33g
7+ email: ph.charriere@gmail.com
8+createdAt: 2026-11-14T18:36:24.000Z
9+updatedAt: 2026-11-14T18:45:13.000Z
10+labels:
11+ - enhancement
12+ - kind::feature
13+milestone: v1.1.0
14+epic: Providers (LLM Engine)
15+related:
16+ - 215
17+body: |
18+ ## Seen (user, 2026-11-14, while testing #215)
19+
20+ With mm in a container, the fallback address works for **ollama** and
21+ **llmman**, not for **llamacpp**.
22+
23+ ## Why (code read)
24+
25+ 1. **llamacpp declares no fallback** ([internal/engine/openai_compat.go](internal/engine/openai_compat.go)):
26+ dmr, ollama, llmman and kronk fall back on
27+ `http://host.docker.internal:<port>/...` when their default does not
28+ answer; llamacpp only has `http://127.0.0.1:8080/v1`. In a container,
29+ that is the container itself, and nothing else is tried. The container
30+ guide says so ("set `baseUrl: http://host.docker.internal:8080/v1`").
31+ The shipped `agent.llamacpp.yaml` even opts out on purpose
32+ (`fallback: ""`, "llama-server is wherever you started it").
33+ 2. **llama-server listens on `127.0.0.1` by default.** On Linux and in a
34+ Docker Sandbox, `host.docker.internal` is the host's gateway, not its
35+ loopback: start it with `--host 0.0.0.0` (as Kronk needs
36+ `--api-host 0.0.0.0:11435`). On Docker Desktop (Mac) the loopback is
37+ reached anyway (Ollama on 127.0.0.1 works there) — not checked for
38+ llama-server.
39+ 3. **Docker Sandbox: port 8080 is not in the mini-me kit's network
40+ policy** (12434, 11434, 17434 only): `sbx policy allow network
41+ host.docker.internal:8080`, or a change in the kit (other repository,
42+ mini-me-kit) — out of scope here.
43+
44+ ## Change (agreed with the user)
45+
46+ - llamacpp's fallback: `http://host.docker.internal:8080/v1`, the same
47+ rule as the others (taken when the default does not answer `/models`
48+ within 2 s; `fallback: ""` in a file still turns it off).
49+ - Its start hint mentions `--host 0.0.0.0` for mm in a container or a
50+ sandbox.
51+ - `agent.llamacpp.yaml` drops `fallback: ""`, so that a copy of it gets
52+ the fallback. Re-litigates that file's explicit choice — the user asked
53+ for llamacpp to behave like the others. `config.samples/` and
54+ `demos/struggle-signals/` (pinned to a fake server) are left alone.
55+ - `mm init` writes the fallback in the file's comment, as for the others
56+ (ticket 121); the wizard's address choice (#214) now also tries it.
57+
58+ ## Acceptance criteria
59+
60+ - mm in a container, no `baseUrl`, llama-server on the host started with
61+ `--host 0.0.0.0` → mm reaches it through `host.docker.internal:8080`.
62+ - On the host, nothing changes when llama-server answers on 127.0.0.1.
63+ - Tests and docs EN+FR.
64+tasks:
65+ - id: 1
66+ title: Engine - llamacpp falls back on http://host.docker.internal:8080/v1, like the other local providers
67+ state: closed
68+ priority: high
69+ author:
70+ name: k33g
71+ email: ph.charriere@gmail.com
72+ createdAt: 2026-11-14T18:36:24.000Z
73+ updatedAt: 2026-11-14T18:39:34.000Z
74+ closedAt: 2026-11-14T18:39:34.000Z
75+ - id: 2
76+ title: startHint - say --host 0.0.0.0 for mm in a container or a sandbox
77+ state: closed
78+ priority: medium
79+ author:
80+ name: k33g
81+ email: ph.charriere@gmail.com
82+ createdAt: 2026-11-14T18:36:24.000Z
83+ updatedAt: 2026-11-14T18:39:34.000Z
84+ closedAt: 2026-11-14T18:39:34.000Z
85+ - id: 3
86+ title: agent.llamacpp.yaml - drop fallback empty (it opted out on purpose)
87+ state: closed
88+ priority: medium
89+ author:
90+ name: k33g
91+ email: ph.charriere@gmail.com
92+ createdAt: 2026-11-14T18:36:24.000Z
93+ updatedAt: 2026-11-14T18:39:34.000Z
94+ closedAt: 2026-11-14T18:39:34.000Z
95+ - id: 4
96+ title: Tests - provider resolution, mm init's comment, the wizard's address choice
97+ state: closed
98+ priority: medium
99+ author:
100+ name: k33g
101+ email: ph.charriere@gmail.com
102+ createdAt: 2026-11-14T18:36:24.000Z
103+ updatedAt: 2026-11-14T18:39:34.000Z
104+ closedAt: 2026-11-14T18:39:34.000Z
105+ - id: 5
106+ title: Docs EN+FR - configuration (providers table), run-in-a-container, run-with-llama-cpp, sandbox (port 8080 policy)
107+ state: closed
108+ priority: medium
109+ author:
110+ name: k33g
111+ email: ph.charriere@gmail.com
112+ createdAt: 2026-11-14T18:36:24.000Z
113+ updatedAt: 2026-11-14T18:39:34.000Z
114+ closedAt: 2026-11-14T18:39:34.000Z
115+comments:
116+ - id: 1
117+ author:
118+ name: k33g
119+ email: ph.charriere@gmail.com
120+ createdAt: 2026-11-14T18:39:34.000Z
121+ body: |
122+ Implemented on `feature/llamacpp-fallback` (uncommitted).
123+
124+ - Engine: llamacpp's fallback `http://host.docker.internal:8080/v1`;
125+ start hint `… --port 8080 (add --host 0.0.0.0 when mm runs in a
126+ container or a sandbox)`.
127+ - `agent.llamacpp.yaml`: `fallback: ""` removed, the comment says what
128+ the fallback is and how to turn it off. `config.samples/` and
129+ `demos/struggle-signals/` keep their `fallback: ""`.
130+ - Tests updated where they asserted "no fallback"
131+ (`TestResolveURLPrecedence`: falls back when baseUrl is down, keeps
132+ it when up; `TestConfigCompat` pins `fallback: ""`;
133+ `TestDefaultURLs`; `TestRenderRecordsTheDefaultURLs`: mm init now
134+ writes the fallback comment for llamacpp). Whole suite green.
135+ - Docs EN+FR: configuration (providers table, agent.llamacpp.yaml),
136+ mcp-configuration (EN), cli (warning example), run-in-a-container,
137+ run-with-llama-cpp (container / sandbox variant), sandbox guide.
138+
139+ Side effect, same as dmr/ollama/llmman already: on the host with
140+ llama-server down, the start-up warning names the fallback address
141+ (`nothing answers at http://host.docker.internal:8080/v1`).
142+
143+ Not tried with a real llama-server in a container or a sandbox.
144+ - id: 2
145+ author:
146+ name: k33g
147+ email: ph.charriere@gmail.com
148+ createdAt: 2026-11-14T18:45:13.000Z
149+ body: |
150+ Closed at the user's request. mm init writes the fallback comment for
151+ llamacpp (checked by the user). In a container, checked only with
152+ v1.0.9 (the released binary the image downloads), which does not have
153+ the fallback yet: the fix reaches containers with the next release.
154+
added .tickets/issues/0218-llamacpp-llama-server-s-default-port-will-change.yaml +114 -0
new file mode 100644
@@ -0,0 +1,114 @@
1+# Managed by IssueSpec. Hand edits are welcome; keep the schema valid.
2+id: 218
3+title: "llamacpp: llama-server's default port will change from 8080 to 9931"
4+state: open
5+author:
6+ name: k33g
7+ email: ph.charriere@gmail.com
8+createdAt: 2026-11-14T18:37:23.000Z
9+updatedAt: 2026-11-14T18:37:23.000Z
10+labels:
11+ - kind::chore
12+milestone: v2.0.0
13+epic: Providers (LLM Engine)
14+related:
15+ - 217
16+body: |
17+ ## Notice (seen by the user, 2026-11-14)
18+
19+ llama-server prints at start-up:
20+
21+ ```text
22+ llama_server: NOTICE: server default port will be changed to :9931 in a future release
23+ ```
24+
25+ The llama.cpp version that printed it, and the release that will make
26+ the change, are not recorded yet.
27+
28+ ## What depends on 8080 in mini-me (read 2026-11-14)
29+
30+ - **The engine** ([internal/engine/openai_compat.go](internal/engine/openai_compat.go)):
31+ llamacpp's default address `http://127.0.0.1:8080/v1`, its fallback
32+ `http://host.docker.internal:8080/v1` (ticket #217) and its start hint
33+ (`llama-server -m <model.gguf> --jinja --port 8080`).
34+ - **Shipped files**: `agent.llamacpp.yaml` (`baseUrl:
35+ http://127.0.0.1:8080/v1`), `config.samples/agent.llamacpp.*.yaml`,
36+ and the demos run on llama-server (e.g. `demos/qwen-style/`,
37+ `demos/decision-skills/`) — to check one by one: most `8080` in
38+ `demos/` are web apps the agents build, unrelated.
39+ - **Docs EN+FR**: `how-to/run-with-llama-cpp.md`,
40+ `reference/configuration.md` (providers table, `agent.llamacpp.yaml`),
41+ `reference/cli.md` (the warning example), `how-to/run-in-a-container.md`,
42+ `how-to/run-in-a-docker-sandbox.md` (`sbx policy allow network
43+ host.docker.internal:8080`), `how-to/verify-tool-order.md`.
44+
45+ ## Who is affected
46+
47+ Only someone who starts llama-server **without** `--port`: the docs and
48+ the start hint always pass `--port 8080`, and a file with an explicit
49+ `baseUrl` keeps working. With the new default, an agent with no
50+ `baseUrl` would find nothing at 8080.
51+
52+ ## To decide
53+
54+ - Follow the new default (`9931`) once the release is out — and say so
55+ in the release notes, since files without `baseUrl` change server.
56+ - Keep `8080` and keep passing `--port 8080` everywhere.
57+ - Or, during a transition, try both on the host: the wizard already
58+ tries several local addresses in order (`localAddresses`, #214); the
59+ engine's own resolution tries one default and one fallback.
60+
61+ ## Done when
62+
63+ The release is identified, the choice is made and written here, and the
64+ engine, the shipped files and the docs agree on it.
65+tasks:
66+ - id: 1
67+ title: Find the llama.cpp release that changes the default port, and whether it is already out
68+ state: open
69+ priority: high
70+ author:
71+ name: k33g
72+ email: ph.charriere@gmail.com
73+ createdAt: 2026-11-14T18:37:23.000Z
74+ - id: 2
75+ title: Decide - follow the new default (9931), keep 8080, or try both during a transition
76+ state: open
77+ priority: high
78+ author:
79+ name: k33g
80+ email: ph.charriere@gmail.com
81+ createdAt: 2026-11-14T18:37:23.000Z
82+ - id: 3
83+ title: Engine - default address, fallback and start hint of llamacpp
84+ state: open
85+ priority: medium
86+ author:
87+ name: k33g
88+ email: ph.charriere@gmail.com
89+ createdAt: 2026-11-14T18:37:23.000Z
90+ - id: 4
91+ title: Shipped files - agent.llamacpp.yaml, config.samples/agent.llamacpp.*, demos using llamacpp
92+ state: open
93+ priority: medium
94+ author:
95+ name: k33g
96+ email: ph.charriere@gmail.com
97+ createdAt: 2026-11-14T18:37:23.000Z
98+ - id: 5
99+ title: Docs EN+FR - run-with-llama-cpp, configuration (providers table), cli (warning example), container and sandbox guides
100+ state: open
101+ priority: medium
102+ author:
103+ name: k33g
104+ email: ph.charriere@gmail.com
105+ createdAt: 2026-11-14T18:37:23.000Z
106+ - id: 6
107+ title: Docker Sandbox - the port in the mini-me kit's network policy, if 8080 is ever added there (mini-me-kit repository)
108+ state: open
109+ priority: low
110+ author:
111+ name: k33g
112+ email: ph.charriere@gmail.com
113+ createdAt: 2026-11-14T18:37:23.000Z
114+comments: []
new file mode 100644
@@ -0,0 +1,114 @@
1+# Managed by IssueSpec. Hand edits are welcome; keep the schema valid.
2+id: 218
3+title: "llamacpp: llama-server's default port will change from 8080 to 9931"
4+state: open
5+author:
6+ name: k33g
7+ email: ph.charriere@gmail.com
8+createdAt: 2026-11-14T18:37:23.000Z
9+updatedAt: 2026-11-14T18:37:23.000Z
10+labels:
11+ - kind::chore
12+milestone: v2.0.0
13+epic: Providers (LLM Engine)
14+related:
15+ - 217
16+body: |
17+ ## Notice (seen by the user, 2026-11-14)
18+
19+ llama-server prints at start-up:
20+
21+ ```text
22+ llama_server: NOTICE: server default port will be changed to :9931 in a future release
23+ ```
24+
25+ The llama.cpp version that printed it, and the release that will make
26+ the change, are not recorded yet.
27+
28+ ## What depends on 8080 in mini-me (read 2026-11-14)
29+
30+ - **The engine** ([internal/engine/openai_compat.go](internal/engine/openai_compat.go)):
31+ llamacpp's default address `http://127.0.0.1:8080/v1`, its fallback
32+ `http://host.docker.internal:8080/v1` (ticket #217) and its start hint
33+ (`llama-server -m <model.gguf> --jinja --port 8080`).
34+ - **Shipped files**: `agent.llamacpp.yaml` (`baseUrl:
35+ http://127.0.0.1:8080/v1`), `config.samples/agent.llamacpp.*.yaml`,
36+ and the demos run on llama-server (e.g. `demos/qwen-style/`,
37+ `demos/decision-skills/`) — to check one by one: most `8080` in
38+ `demos/` are web apps the agents build, unrelated.
39+ - **Docs EN+FR**: `how-to/run-with-llama-cpp.md`,
40+ `reference/configuration.md` (providers table, `agent.llamacpp.yaml`),
41+ `reference/cli.md` (the warning example), `how-to/run-in-a-container.md`,
42+ `how-to/run-in-a-docker-sandbox.md` (`sbx policy allow network
43+ host.docker.internal:8080`), `how-to/verify-tool-order.md`.
44+
45+ ## Who is affected
46+
47+ Only someone who starts llama-server **without** `--port`: the docs and
48+ the start hint always pass `--port 8080`, and a file with an explicit
49+ `baseUrl` keeps working. With the new default, an agent with no
50+ `baseUrl` would find nothing at 8080.
51+
52+ ## To decide
53+
54+ - Follow the new default (`9931`) once the release is out — and say so
55+ in the release notes, since files without `baseUrl` change server.
56+ - Keep `8080` and keep passing `--port 8080` everywhere.
57+ - Or, during a transition, try both on the host: the wizard already
58+ tries several local addresses in order (`localAddresses`, #214); the
59+ engine's own resolution tries one default and one fallback.
60+
61+ ## Done when
62+
63+ The release is identified, the choice is made and written here, and the
64+ engine, the shipped files and the docs agree on it.
65+tasks:
66+ - id: 1
67+ title: Find the llama.cpp release that changes the default port, and whether it is already out
68+ state: open
69+ priority: high
70+ author:
71+ name: k33g
72+ email: ph.charriere@gmail.com
73+ createdAt: 2026-11-14T18:37:23.000Z
74+ - id: 2
75+ title: Decide - follow the new default (9931), keep 8080, or try both during a transition
76+ state: open
77+ priority: high
78+ author:
79+ name: k33g
80+ email: ph.charriere@gmail.com
81+ createdAt: 2026-11-14T18:37:23.000Z
82+ - id: 3
83+ title: Engine - default address, fallback and start hint of llamacpp
84+ state: open
85+ priority: medium
86+ author:
87+ name: k33g
88+ email: ph.charriere@gmail.com
89+ createdAt: 2026-11-14T18:37:23.000Z
90+ - id: 4
91+ title: Shipped files - agent.llamacpp.yaml, config.samples/agent.llamacpp.*, demos using llamacpp
92+ state: open
93+ priority: medium
94+ author:
95+ name: k33g
96+ email: ph.charriere@gmail.com
97+ createdAt: 2026-11-14T18:37:23.000Z
98+ - id: 5
99+ title: Docs EN+FR - run-with-llama-cpp, configuration (providers table), cli (warning example), container and sandbox guides
100+ state: open
101+ priority: medium
102+ author:
103+ name: k33g
104+ email: ph.charriere@gmail.com
105+ createdAt: 2026-11-14T18:37:23.000Z
106+ - id: 6
107+ title: Docker Sandbox - the port in the mini-me kit's network policy, if 8080 is ever added there (mini-me-kit repository)
108+ state: open
109+ priority: low
110+ author:
111+ name: k33g
112+ email: ph.charriere@gmail.com
113+ createdAt: 2026-11-14T18:37:23.000Z
114+comments: []
modified agent.llamacpp.yaml +4 -2
@@ -9,8 +9,10 @@ model: jetbrains/mellum2-12b-a2.5b-instruct-gguf-q4_k_m:Q4_K_M
99 # llama-server's default is 127.0.0.1:8080; the OpenAI routes sit under /v1.
1010 baseUrl: http://127.0.0.1:8080/v1
1111
12-# No fallback: llama-server is wherever you started it.
13-fallback: ""
12+# No fallback line: when baseUrl does not answer, mm tries llama-server on the
13+# host, http://host.docker.internal:8080/v1 — what mm in a container or a
14+# sandbox needs (start llama-server with --host 0.0.0.0 there). fallback: ""
15+# turns that off.
1416
1517 # Only when llama-server was started with --api-key; unset otherwise.
1618 # apiKeyEnv: LLAMA_API_KEY
@@ -9,8 +9,10 @@ model: jetbrains/mellum2-12b-a2.5b-instruct-gguf-q4_k_m:Q4_K_M
9 # llama-server's default is 127.0.0.1:8080; the OpenAI routes sit under /v1.9 # llama-server's default is 127.0.0.1:8080; the OpenAI routes sit under /v1.
10 baseUrl: http://127.0.0.1:8080/v110 baseUrl: http://127.0.0.1:8080/v1
11 11
12-# No fallback: llama-server is wherever you started it.12+# No fallback line: when baseUrl does not answer, mm tries llama-server on the
13-fallback: ""13+# host, http://host.docker.internal:8080/v1 — what mm in a container or a
14+# sandbox needs (start llama-server with --host 0.0.0.0 there). fallback: ""
15+# turns that off.
14 16
15 # Only when llama-server was started with --api-key; unset otherwise.17 # Only when llama-server was started with --api-key; unset otherwise.
16 # apiKeyEnv: LLAMA_API_KEY18 # apiKeyEnv: LLAMA_API_KEY
modified docs/en/how-to/run-in-a-container.md +2 -2
@@ -68,9 +68,9 @@ The project is mounted at the same path as on your machine, so the paths the edi
6868
6969 ## Reach the model
7070
71-From the container, `localhost` is the container. `dmr`, `llmman` and `ollama` fall back on `host.docker.internal` by themselves when their default address does not answer, as long as the agent file sets no `baseUrl`. The compose file makes that name known on Linux too.
71+From the container, `localhost` is the container. `dmr`, `llmman`, `ollama`, `kronk` and `llamacpp` fall back on `host.docker.internal` by themselves when their default address does not answer, as long as the agent file sets no `baseUrl`. The compose file makes that name known on Linux too.
7272
73-- **`llamacpp`** has no such fallback: set `baseUrl: http://host.docker.internal:8080/v1`.
73+- **`llamacpp`**: on Linux, start llama-server with `--host 0.0.0.0`, as it listens on the loopback only by default (see the last point). Files written before v1.1.0 may say `fallback: ""`: remove that line.
7474 - **An agent file with a `baseUrl` on `localhost` or `127.0.0.1`:** replace the host with `host.docker.internal`, or add a `fallback`.
7575 - **A server on Linux that only listens on `127.0.0.1`:** make it listen on an address the container can reach. `host.docker.internal` is then the host's gateway, not its loopback.
7676
@@ -68,9 +68,9 @@ The project is mounted at the same path as on your machine, so the paths the edi
68 68
69 ## Reach the model69 ## Reach the model
70 70
71-From the container, `localhost` is the container. `dmr`, `llmman` and `ollama` fall back on `host.docker.internal` by themselves when their default address does not answer, as long as the agent file sets no `baseUrl`. The compose file makes that name known on Linux too.71+From the container, `localhost` is the container. `dmr`, `llmman`, `ollama`, `kronk` and `llamacpp` fall back on `host.docker.internal` by themselves when their default address does not answer, as long as the agent file sets no `baseUrl`. The compose file makes that name known on Linux too.
72 72
73-- **`llamacpp`** has no such fallback: set `baseUrl: http://host.docker.internal:8080/v1`.73+- **`llamacpp`**: on Linux, start llama-server with `--host 0.0.0.0`, as it listens on the loopback only by default (see the last point). Files written before v1.1.0 may say `fallback: ""`: remove that line.
74 - **An agent file with a `baseUrl` on `localhost` or `127.0.0.1`:** replace the host with `host.docker.internal`, or add a `fallback`.74 - **An agent file with a `baseUrl` on `localhost` or `127.0.0.1`:** replace the host with `host.docker.internal`, or add a `fallback`.
75 - **A server on Linux that only listens on `127.0.0.1`:** make it listen on an address the container can reach. `host.docker.internal` is then the host's gateway, not its loopback.75 - **A server on Linux that only listens on `127.0.0.1`:** make it listen on an address the container can reach. `host.docker.internal` is then the host's gateway, not its loopback.
76 76
modified docs/en/how-to/run-in-a-docker-sandbox.md +1 -1
@@ -203,7 +203,7 @@ Inside a sandbox `localhost` is the sandbox itself. Your model server keeps runn
203203
204204 - The `mini-me` kit's network policy allows **`host.docker.internal:12434`** (Docker Model Runner), **`:11434`** (Ollama) and **`:17434`** (llmman). `agent.dmr.yaml` already falls back to `http://host.docker.internal:12434/engines/v1`; the configurations shipped in `demos/use-with-sbx/` use llmman (`fallback: http://host.docker.internal:17434/v1`).
205205 - Kronk (`:11435`) is not in that list: allow `localhost:11435` in the policy, and start Kronk on the host with `--api-host 0.0.0.0:11435` — it listens on the loopback only otherwise. See [How to run mini-me on Kronk](run-with-kronk.md).
206-- For another port — llama-server's `8080`, for instance — add it to the kit's `network-policy` capability, or allow it for one sandbox:
206+- For another port — llama-server's `8080`, for instance, which mm tries on its own (`http://host.docker.internal:8080/v1`) when started on the host with `--host 0.0.0.0` — add it to the kit's `network-policy` capability, or allow it for one sandbox:
207207
208208 ```bash
209209 sbx policy allow network host.docker.internal:8080
@@ -203,7 +203,7 @@ Inside a sandbox `localhost` is the sandbox itself. Your model server keeps runn
203 203
204 - The `mini-me` kit's network policy allows **`host.docker.internal:12434`** (Docker Model Runner), **`:11434`** (Ollama) and **`:17434`** (llmman). `agent.dmr.yaml` already falls back to `http://host.docker.internal:12434/engines/v1`; the configurations shipped in `demos/use-with-sbx/` use llmman (`fallback: http://host.docker.internal:17434/v1`).204 - The `mini-me` kit's network policy allows **`host.docker.internal:12434`** (Docker Model Runner), **`:11434`** (Ollama) and **`:17434`** (llmman). `agent.dmr.yaml` already falls back to `http://host.docker.internal:12434/engines/v1`; the configurations shipped in `demos/use-with-sbx/` use llmman (`fallback: http://host.docker.internal:17434/v1`).
205 - Kronk (`:11435`) is not in that list: allow `localhost:11435` in the policy, and start Kronk on the host with `--api-host 0.0.0.0:11435` — it listens on the loopback only otherwise. See [How to run mini-me on Kronk](run-with-kronk.md).205 - Kronk (`:11435`) is not in that list: allow `localhost:11435` in the policy, and start Kronk on the host with `--api-host 0.0.0.0:11435` — it listens on the loopback only otherwise. See [How to run mini-me on Kronk](run-with-kronk.md).
206-- For another port — llama-server's `8080`, for instance — add it to the kit's `network-policy` capability, or allow it for one sandbox:206+- For another port — llama-server's `8080`, for instance, which mm tries on its own (`http://host.docker.internal:8080/v1`) when started on the host with `--host 0.0.0.0` — add it to the kit's `network-policy` capability, or allow it for one sandbox:
207 207
208 ```bash208 ```bash
209 sbx policy allow network host.docker.internal:8080209 sbx policy allow network host.docker.internal:8080
modified docs/en/how-to/run-with-llama-cpp.md +1 -0
@@ -23,6 +23,7 @@ This guide shows how to point `mm` at a `llama-server` you start yourself. It as
2323 ## Variants
2424
2525 - **Different port or host.** Set `baseUrl` in the YAML to `http://<host>:<port>/v1`, or override it for one run with `AGENT_BASE_URL=http://<host>:<port>/v1 ./mm agent.llamacpp.yaml`. An environment override is taken as-is, without the fallback probe.
26+- **From a container or a Docker Sandbox.** When `baseUrl` does not answer, mm tries `http://host.docker.internal:8080/v1`, llama-server on the host. Start it there with `--host 0.0.0.0`: by default it listens on the loopback only, which a container on Linux or a sandbox cannot reach. In a sandbox, also allow the port: `sbx policy allow network host.docker.internal:8080`. `fallback: ""` turns the fallback off. See [run mini-me in a container](run-in-a-container.md) and [run mini-me in a Docker Sandbox](run-in-a-docker-sandbox.md).
2627 - **`autoPull: true`.** llama-server cannot download a model over its API: it serves the one it was started with. `-hf <repo>:<quant>` (step 1) is what downloads it. With `autoPull` on and a `model:` the server does not list, mm says so at start-up and goes on. See [`autoPull`](../reference/configuration.md).
2728 - **Server started with `--api-key`.** Export the key in `LLAMA_API_KEY` (the provider's default variable) or name another variable under `apiKeyEnv`. The key itself never goes in the YAML.
2829 - **Keep `agent.yaml` and switch provider for one run.** `AGENT_PROVIDER=llamacpp ./mm` works as long as `agent.yaml` leaves `baseUrl` empty; an explicit DMR URL in the file would still be honoured and point at port 12434.
@@ -23,6 +23,7 @@ This guide shows how to point `mm` at a `llama-server` you start yourself. It as
23 ## Variants23 ## Variants
24 24
25 - **Different port or host.** Set `baseUrl` in the YAML to `http://<host>:<port>/v1`, or override it for one run with `AGENT_BASE_URL=http://<host>:<port>/v1 ./mm agent.llamacpp.yaml`. An environment override is taken as-is, without the fallback probe.25 - **Different port or host.** Set `baseUrl` in the YAML to `http://<host>:<port>/v1`, or override it for one run with `AGENT_BASE_URL=http://<host>:<port>/v1 ./mm agent.llamacpp.yaml`. An environment override is taken as-is, without the fallback probe.
26+- **From a container or a Docker Sandbox.** When `baseUrl` does not answer, mm tries `http://host.docker.internal:8080/v1`, llama-server on the host. Start it there with `--host 0.0.0.0`: by default it listens on the loopback only, which a container on Linux or a sandbox cannot reach. In a sandbox, also allow the port: `sbx policy allow network host.docker.internal:8080`. `fallback: ""` turns the fallback off. See [run mini-me in a container](run-in-a-container.md) and [run mini-me in a Docker Sandbox](run-in-a-docker-sandbox.md).
26 - **`autoPull: true`.** llama-server cannot download a model over its API: it serves the one it was started with. `-hf <repo>:<quant>` (step 1) is what downloads it. With `autoPull` on and a `model:` the server does not list, mm says so at start-up and goes on. See [`autoPull`](../reference/configuration.md).27 - **`autoPull: true`.** llama-server cannot download a model over its API: it serves the one it was started with. `-hf <repo>:<quant>` (step 1) is what downloads it. With `autoPull` on and a `model:` the server does not list, mm says so at start-up and goes on. See [`autoPull`](../reference/configuration.md).
27 - **Server started with `--api-key`.** Export the key in `LLAMA_API_KEY` (the provider's default variable) or name another variable under `apiKeyEnv`. The key itself never goes in the YAML.28 - **Server started with `--api-key`.** Export the key in `LLAMA_API_KEY` (the provider's default variable) or name another variable under `apiKeyEnv`. The key itself never goes in the YAML.
28 - **Keep `agent.yaml` and switch provider for one run.** `AGENT_PROVIDER=llamacpp ./mm` works as long as `agent.yaml` leaves `baseUrl` empty; an explicit DMR URL in the file would still be honoured and point at port 12434.29 - **Keep `agent.yaml` and switch provider for one run.** `AGENT_PROVIDER=llamacpp ./mm` works as long as `agent.yaml` leaves `baseUrl` empty; an explicit DMR URL in the file would still be honoured and point at port 12434.
modified docs/en/reference/cli.md +1 -1
@@ -104,7 +104,7 @@ Printed on stdout in terminal mode, on stderr in ACP mode. Dimmed when stdout is
104104 ### Example
105105
106106 ```
107-[warning: llamacpp: nothing answers at http://127.0.0.1:8080/v1 — start it: llama-server -m <model.gguf> --jinja --port 8080]
107+[warning: llamacpp: nothing answers at http://host.docker.internal:8080/v1 — start it: llama-server -m <model.gguf> --jinja --port 8080 (add --host 0.0.0.0 when mm runs in a container or a sandbox)]
108108 mm v0.1.7 | config: agent.llamacpp.yaml | provider: llamacpp | model: jetbrains/mellum2-12b-a2.5b-instruct-gguf-q4_k_m:Q4_K_M | ctx: unknown | tools: bash, read_file, write_file, edit_file | skills: 0 | yolo: on
109109 ```
110110
@@ -104,7 +104,7 @@ Printed on stdout in terminal mode, on stderr in ACP mode. Dimmed when stdout is
104 ### Example104 ### Example
105 105
106 ```106 ```
107-[warning: llamacpp: nothing answers at http://127.0.0.1:8080/v1 — start it: llama-server -m <model.gguf> --jinja --port 8080]107+[warning: llamacpp: nothing answers at http://host.docker.internal:8080/v1 — start it: llama-server -m <model.gguf> --jinja --port 8080 (add --host 0.0.0.0 when mm runs in a container or a sandbox)]
108 mm v0.1.7 | config: agent.llamacpp.yaml | provider: llamacpp | model: jetbrains/mellum2-12b-a2.5b-instruct-gguf-q4_k_m:Q4_K_M | ctx: unknown | tools: bash, read_file, write_file, edit_file | skills: 0 | yolo: on108 mm v0.1.7 | config: agent.llamacpp.yaml | provider: llamacpp | model: jetbrains/mellum2-12b-a2.5b-instruct-gguf-q4_k_m:Q4_K_M | ctx: unknown | tools: bash, read_file, write_file, edit_file | skills: 0 | yolo: on
109 ```109 ```
110 110
modified docs/en/reference/configuration.md +2 -2
@@ -197,7 +197,7 @@ tools:
197197 | Provider | `baseUrl` | `fallback` | `apiKeyEnv` | Key required |
198198 |----------|-----------|------------|-------------|--------------|
199199 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | none | no |
200-| `llamacpp` | `http://127.0.0.1:8080/v1` | none | `LLAMA_API_KEY` | no |
200+| `llamacpp` | `http://127.0.0.1:8080/v1` | `http://host.docker.internal:8080/v1` | `LLAMA_API_KEY` | no |
201201 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | no |
202202 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | no |
203203 | `kronk` | `http://127.0.0.1:11435/v1` | `http://host.docker.internal:11435/v1` | `KRONK_TOKEN` | no — only Kronk's `authenticated` and `full-protected` modes check it |
@@ -477,7 +477,7 @@ A start-up warning, not an error, is printed when `context.enabled` is `true`, t
477477 | File | Purpose |
478478 |------|---------|
479479 | `agent.yaml` | Docker Model Runner set-up: `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression off, a system prompt named "Bob" that describes the skills and file-editing rules. |
480-| `agent.llamacpp.yaml` | llama.cpp set-up: `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, `fallback: ""`, compression on. |
480+| `agent.llamacpp.yaml` | llama.cpp set-up: `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, compression on. |
481481
482482 ### Example
483483
@@ -197,7 +197,7 @@ tools:
197 | Provider | `baseUrl` | `fallback` | `apiKeyEnv` | Key required |197 | Provider | `baseUrl` | `fallback` | `apiKeyEnv` | Key required |
198 |----------|-----------|------------|-------------|--------------|198 |----------|-----------|------------|-------------|--------------|
199 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | none | no |199 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | none | no |
200-| `llamacpp` | `http://127.0.0.1:8080/v1` | none | `LLAMA_API_KEY` | no |200+| `llamacpp` | `http://127.0.0.1:8080/v1` | `http://host.docker.internal:8080/v1` | `LLAMA_API_KEY` | no |
201 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | no |201 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | no |
202 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | no |202 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | no |
203 | `kronk` | `http://127.0.0.1:11435/v1` | `http://host.docker.internal:11435/v1` | `KRONK_TOKEN` | no — only Kronk's `authenticated` and `full-protected` modes check it |203 | `kronk` | `http://127.0.0.1:11435/v1` | `http://host.docker.internal:11435/v1` | `KRONK_TOKEN` | no — only Kronk's `authenticated` and `full-protected` modes check it |
@@ -477,7 +477,7 @@ A start-up warning, not an error, is printed when `context.enabled` is `true`, t
477 | File | Purpose |477 | File | Purpose |
478 |------|---------|478 |------|---------|
479 | `agent.yaml` | Docker Model Runner set-up: `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression off, a system prompt named "Bob" that describes the skills and file-editing rules. |479 | `agent.yaml` | Docker Model Runner set-up: `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression off, a system prompt named "Bob" that describes the skills and file-editing rules. |
480-| `agent.llamacpp.yaml` | llama.cpp set-up: `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, `fallback: ""`, compression on. |480+| `agent.llamacpp.yaml` | llama.cpp set-up: `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, compression on. |
481 481
482 ### Example482 ### Example
483 483
modified docs/en/reference/mcp-configuration.md +2 -2
@@ -78,7 +78,7 @@ The shipped `agent.lsp.yaml` is `agent.dmr.yaml` with this block on (gopls activ
7878 | Provider | `baseUrl` | `fallback` | `apiKeyEnv` | Key required |
7979 |----------|-----------|------------|-------------|--------------|
8080 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | none | no |
81-| `llamacpp` | `http://127.0.0.1:8080/v1` | none | `LLAMA_API_KEY` | no |
81+| `llamacpp` | `http://127.0.0.1:8080/v1` | `http://host.docker.internal:8080/v1` | `LLAMA_API_KEY` | no |
8282 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | no |
8383 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | no |
8484
@@ -114,7 +114,7 @@ A start-up warning, not an error, is printed when `context.enabled` is `true`, t
114114 | File | Purpose |
115115 |------|---------|
116116 | `agent.yaml` | Docker Model Runner set-up: `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression off, a system prompt named "Bob" that describes the skills and file-editing rules. |
117-| `agent.llamacpp.yaml` | llama.cpp set-up: `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, `fallback: ""`, compression on. |
117+| `agent.llamacpp.yaml` | llama.cpp set-up: `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, compression on. |
118118
119119 ### Example
120120
@@ -78,7 +78,7 @@ The shipped `agent.lsp.yaml` is `agent.dmr.yaml` with this block on (gopls activ
78 | Provider | `baseUrl` | `fallback` | `apiKeyEnv` | Key required |78 | Provider | `baseUrl` | `fallback` | `apiKeyEnv` | Key required |
79 |----------|-----------|------------|-------------|--------------|79 |----------|-----------|------------|-------------|--------------|
80 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | none | no |80 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | none | no |
81-| `llamacpp` | `http://127.0.0.1:8080/v1` | none | `LLAMA_API_KEY` | no |81+| `llamacpp` | `http://127.0.0.1:8080/v1` | `http://host.docker.internal:8080/v1` | `LLAMA_API_KEY` | no |
82 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | no |82 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | no |
83 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | no |83 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | no |
84 84
@@ -114,7 +114,7 @@ A start-up warning, not an error, is printed when `context.enabled` is `true`, t
114 | File | Purpose |114 | File | Purpose |
115 |------|---------|115 |------|---------|
116 | `agent.yaml` | Docker Model Runner set-up: `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression off, a system prompt named "Bob" that describes the skills and file-editing rules. |116 | `agent.yaml` | Docker Model Runner set-up: `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression off, a system prompt named "Bob" that describes the skills and file-editing rules. |
117-| `agent.llamacpp.yaml` | llama.cpp set-up: `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, `fallback: ""`, compression on. |117+| `agent.llamacpp.yaml` | llama.cpp set-up: `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, compression on. |
118 118
119 ### Example119 ### Example
120 120
modified docs/fr/how-to/run-in-a-container.md +2 -2
@@ -68,9 +68,9 @@ Le projet est monté au même chemin que sur votre machine : les chemins envoyé
6868
6969 ## Joindre le modèle
7070
71-Depuis le conteneur, `localhost` désigne le conteneur. `dmr`, `llmman` et `ollama` se rabattent d'eux-mêmes sur `host.docker.internal` quand leur adresse par défaut ne répond pas, tant que le fichier d'agent ne définit pas de `baseUrl`. Le fichier compose fait aussi connaître ce nom sous Linux.
71+Depuis le conteneur, `localhost` désigne le conteneur. `dmr`, `llmman`, `ollama`, `kronk` et `llamacpp` se rabattent d'eux-mêmes sur `host.docker.internal` quand leur adresse par défaut ne répond pas, tant que le fichier d'agent ne définit pas de `baseUrl`. Le fichier compose fait aussi connaître ce nom sous Linux.
7272
73-- **`llamacpp`** n'a pas ce repli : définissez `baseUrl: http://host.docker.internal:8080/v1`.
73+- **`llamacpp`** : sous Linux, lancez llama-server avec `--host 0.0.0.0`, car par défaut il n'écoute que sur la boucle locale (voir le dernier point). Les fichiers écrits avant la v1.1.0 peuvent contenir `fallback: ""` : retirez cette ligne.
7474 - **Un fichier d'agent avec un `baseUrl` sur `localhost` ou `127.0.0.1` :** remplacez l'hôte par `host.docker.internal`, ou ajoutez un `fallback`.
7575 - **Un serveur sous Linux qui n'écoute que sur `127.0.0.1` :** faites-le écouter sur une adresse joignable par le conteneur. `host.docker.internal` y désigne la passerelle de l'hôte, pas sa boucle locale.
7676
@@ -68,9 +68,9 @@ Le projet est monté au même chemin que sur votre machine : les chemins envoyé
68 68
69 ## Joindre le modèle69 ## Joindre le modèle
70 70
71-Depuis le conteneur, `localhost` désigne le conteneur. `dmr`, `llmman` et `ollama` se rabattent d'eux-mêmes sur `host.docker.internal` quand leur adresse par défaut ne répond pas, tant que le fichier d'agent ne définit pas de `baseUrl`. Le fichier compose fait aussi connaître ce nom sous Linux.71+Depuis le conteneur, `localhost` désigne le conteneur. `dmr`, `llmman`, `ollama`, `kronk` et `llamacpp` se rabattent d'eux-mêmes sur `host.docker.internal` quand leur adresse par défaut ne répond pas, tant que le fichier d'agent ne définit pas de `baseUrl`. Le fichier compose fait aussi connaître ce nom sous Linux.
72 72
73-- **`llamacpp`** n'a pas ce repli : définissez `baseUrl: http://host.docker.internal:8080/v1`.73+- **`llamacpp`** : sous Linux, lancez llama-server avec `--host 0.0.0.0`, car par défaut il n'écoute que sur la boucle locale (voir le dernier point). Les fichiers écrits avant la v1.1.0 peuvent contenir `fallback: ""` : retirez cette ligne.
74 - **Un fichier d'agent avec un `baseUrl` sur `localhost` ou `127.0.0.1` :** remplacez l'hôte par `host.docker.internal`, ou ajoutez un `fallback`.74 - **Un fichier d'agent avec un `baseUrl` sur `localhost` ou `127.0.0.1` :** remplacez l'hôte par `host.docker.internal`, ou ajoutez un `fallback`.
75 - **Un serveur sous Linux qui n'écoute que sur `127.0.0.1` :** faites-le écouter sur une adresse joignable par le conteneur. `host.docker.internal` y désigne la passerelle de l'hôte, pas sa boucle locale.75 - **Un serveur sous Linux qui n'écoute que sur `127.0.0.1` :** faites-le écouter sur une adresse joignable par le conteneur. `host.docker.internal` y désigne la passerelle de l'hôte, pas sa boucle locale.
76 76
modified docs/fr/how-to/run-in-a-docker-sandbox.md +1 -1
@@ -203,7 +203,7 @@ Dans un sandbox, `localhost` est le sandbox lui-même. Votre serveur de modèles
203203
204204 - La politique réseau du kit `mini-me` autorise **`host.docker.internal:12434`** (Docker Model Runner), **`:11434`** (Ollama) et **`:17434`** (llmman). `agent.dmr.yaml` se replie déjà sur `http://host.docker.internal:12434/engines/v1` ; les configurations livrées dans `demos/use-with-sbx/` utilisent llmman (`fallback: http://host.docker.internal:17434/v1`).
205205 - Kronk (`:11435`) n'est pas dans cette liste : autorisez `localhost:11435` dans la politique, et lancez Kronk sur l'hôte avec `--api-host 0.0.0.0:11435` — sinon il n'écoute que sur la boucle locale. Voir [Faire tourner mini-me sur Kronk](run-with-kronk.md).
206-- Pour un autre port — le `8080` de llama-server, par exemple — ajoutez-le à la capacité `network-policy` du kit, ou autorisez-le pour un seul sandbox :
206+- Pour un autre port — le `8080` de llama-server, par exemple, que mm essaie de lui-même (`http://host.docker.internal:8080/v1`) quand il est lancé sur l'hôte avec `--host 0.0.0.0` — ajoutez-le à la capacité `network-policy` du kit, ou autorisez-le pour un seul sandbox :
207207
208208 ```bash
209209 sbx policy allow network host.docker.internal:8080
@@ -203,7 +203,7 @@ Dans un sandbox, `localhost` est le sandbox lui-même. Votre serveur de modèles
203 203
204 - La politique réseau du kit `mini-me` autorise **`host.docker.internal:12434`** (Docker Model Runner), **`:11434`** (Ollama) et **`:17434`** (llmman). `agent.dmr.yaml` se replie déjà sur `http://host.docker.internal:12434/engines/v1` ; les configurations livrées dans `demos/use-with-sbx/` utilisent llmman (`fallback: http://host.docker.internal:17434/v1`).204 - La politique réseau du kit `mini-me` autorise **`host.docker.internal:12434`** (Docker Model Runner), **`:11434`** (Ollama) et **`:17434`** (llmman). `agent.dmr.yaml` se replie déjà sur `http://host.docker.internal:12434/engines/v1` ; les configurations livrées dans `demos/use-with-sbx/` utilisent llmman (`fallback: http://host.docker.internal:17434/v1`).
205 - Kronk (`:11435`) n'est pas dans cette liste : autorisez `localhost:11435` dans la politique, et lancez Kronk sur l'hôte avec `--api-host 0.0.0.0:11435` — sinon il n'écoute que sur la boucle locale. Voir [Faire tourner mini-me sur Kronk](run-with-kronk.md).205 - Kronk (`:11435`) n'est pas dans cette liste : autorisez `localhost:11435` dans la politique, et lancez Kronk sur l'hôte avec `--api-host 0.0.0.0:11435` — sinon il n'écoute que sur la boucle locale. Voir [Faire tourner mini-me sur Kronk](run-with-kronk.md).
206-- Pour un autre port — le `8080` de llama-server, par exemple — ajoutez-le à la capacité `network-policy` du kit, ou autorisez-le pour un seul sandbox :206+- Pour un autre port — le `8080` de llama-server, par exemple, que mm essaie de lui-même (`http://host.docker.internal:8080/v1`) quand il est lancé sur l'hôte avec `--host 0.0.0.0` — ajoutez-le à la capacité `network-policy` du kit, ou autorisez-le pour un seul sandbox :
207 207
208 ```bash208 ```bash
209 sbx policy allow network host.docker.internal:8080209 sbx policy allow network host.docker.internal:8080
modified docs/fr/how-to/run-with-llama-cpp.md +1 -0
@@ -23,6 +23,7 @@ Ce guide montre comment pointer `mm` vers un `llama-server` que vous démarrez v
2323 ## Variantes
2424
2525 - **Autre port ou autre hôte.** Mettez `baseUrl` à `http://<hôte>:<port>/v1` dans le YAML, ou surchargez pour une seule exécution avec `AGENT_BASE_URL=http://<hôte>:<port>/v1 ./mm agent.llamacpp.yaml`. Une surcharge par l'environnement est prise telle quelle, sans sonde de repli.
26+- **Depuis un conteneur ou un Docker Sandbox.** Quand `baseUrl` ne répond pas, mm essaie `http://host.docker.internal:8080/v1`, llama-server sur l'hôte. Lancez-le avec `--host 0.0.0.0` : par défaut il n'écoute que sur la boucle locale, qu'un conteneur sous Linux ou un sandbox ne peuvent pas joindre. Dans un sandbox, autorisez aussi le port : `sbx policy allow network host.docker.internal:8080`. `fallback: ""` coupe le repli. Voir [faire tourner mini-me dans un conteneur](run-in-a-container.md) et [faire tourner mini-me dans un Docker Sandbox](run-in-a-docker-sandbox.md).
2627 - **`autoPull: true`.** llama-server ne sait pas télécharger un modèle par son API : il sert celui avec lequel il a été lancé. C'est `-hf <dépôt>:<quant>` (étape 1) qui le télécharge. Avec `autoPull` actif et un `model:` que le serveur ne liste pas, mm le dit au démarrage et continue. Voir [`autoPull`](../reference/configuration.md).
2728 - **Serveur démarré avec `--api-key`.** Exportez la clé dans `LLAMA_API_KEY` (la variable par défaut du fournisseur) ou nommez une autre variable dans `apiKeyEnv`. La clé elle-même ne va jamais dans le YAML.
2829 - **Garder `agent.yaml` et changer de fournisseur pour une exécution.** `AGENT_PROVIDER=llamacpp ./mm` fonctionne tant qu'`agent.yaml` laisse `baseUrl` vide ; une URL DMR explicite dans le fichier serait respectée et pointerait toujours sur le port 12434.
@@ -23,6 +23,7 @@ Ce guide montre comment pointer `mm` vers un `llama-server` que vous démarrez v
23 ## Variantes23 ## Variantes
24 24
25 - **Autre port ou autre hôte.** Mettez `baseUrl` à `http://<hôte>:<port>/v1` dans le YAML, ou surchargez pour une seule exécution avec `AGENT_BASE_URL=http://<hôte>:<port>/v1 ./mm agent.llamacpp.yaml`. Une surcharge par l'environnement est prise telle quelle, sans sonde de repli.25 - **Autre port ou autre hôte.** Mettez `baseUrl` à `http://<hôte>:<port>/v1` dans le YAML, ou surchargez pour une seule exécution avec `AGENT_BASE_URL=http://<hôte>:<port>/v1 ./mm agent.llamacpp.yaml`. Une surcharge par l'environnement est prise telle quelle, sans sonde de repli.
26+- **Depuis un conteneur ou un Docker Sandbox.** Quand `baseUrl` ne répond pas, mm essaie `http://host.docker.internal:8080/v1`, llama-server sur l'hôte. Lancez-le avec `--host 0.0.0.0` : par défaut il n'écoute que sur la boucle locale, qu'un conteneur sous Linux ou un sandbox ne peuvent pas joindre. Dans un sandbox, autorisez aussi le port : `sbx policy allow network host.docker.internal:8080`. `fallback: ""` coupe le repli. Voir [faire tourner mini-me dans un conteneur](run-in-a-container.md) et [faire tourner mini-me dans un Docker Sandbox](run-in-a-docker-sandbox.md).
26 - **`autoPull: true`.** llama-server ne sait pas télécharger un modèle par son API : il sert celui avec lequel il a été lancé. C'est `-hf <dépôt>:<quant>` (étape 1) qui le télécharge. Avec `autoPull` actif et un `model:` que le serveur ne liste pas, mm le dit au démarrage et continue. Voir [`autoPull`](../reference/configuration.md).27 - **`autoPull: true`.** llama-server ne sait pas télécharger un modèle par son API : il sert celui avec lequel il a été lancé. C'est `-hf <dépôt>:<quant>` (étape 1) qui le télécharge. Avec `autoPull` actif et un `model:` que le serveur ne liste pas, mm le dit au démarrage et continue. Voir [`autoPull`](../reference/configuration.md).
27 - **Serveur démarré avec `--api-key`.** Exportez la clé dans `LLAMA_API_KEY` (la variable par défaut du fournisseur) ou nommez une autre variable dans `apiKeyEnv`. La clé elle-même ne va jamais dans le YAML.28 - **Serveur démarré avec `--api-key`.** Exportez la clé dans `LLAMA_API_KEY` (la variable par défaut du fournisseur) ou nommez une autre variable dans `apiKeyEnv`. La clé elle-même ne va jamais dans le YAML.
28 - **Garder `agent.yaml` et changer de fournisseur pour une exécution.** `AGENT_PROVIDER=llamacpp ./mm` fonctionne tant qu'`agent.yaml` laisse `baseUrl` vide ; une URL DMR explicite dans le fichier serait respectée et pointerait toujours sur le port 12434.29 - **Garder `agent.yaml` et changer de fournisseur pour une exécution.** `AGENT_PROVIDER=llamacpp ./mm` fonctionne tant qu'`agent.yaml` laisse `baseUrl` vide ; une URL DMR explicite dans le fichier serait respectée et pointerait toujours sur le port 12434.
modified docs/fr/reference/cli.md +1 -1
@@ -104,7 +104,7 @@ Affichée sur stdout en mode terminal, sur stderr en mode ACP. Grisée quand std
104104 ### Exemple
105105
106106 ```
107-[warning: llamacpp: nothing answers at http://127.0.0.1:8080/v1 — start it: llama-server -m <model.gguf> --jinja --port 8080]
107+[warning: llamacpp: nothing answers at http://host.docker.internal:8080/v1 — start it: llama-server -m <model.gguf> --jinja --port 8080 (add --host 0.0.0.0 when mm runs in a container or a sandbox)]
108108 mm v0.1.7 | config: agent.llamacpp.yaml | provider: llamacpp | model: jetbrains/mellum2-12b-a2.5b-instruct-gguf-q4_k_m:Q4_K_M | ctx: unknown | tools: bash, read_file, write_file, edit_file | skills: 0 | yolo: on
109109 ```
110110
@@ -104,7 +104,7 @@ Affichée sur stdout en mode terminal, sur stderr en mode ACP. Grisée quand std
104 ### Exemple104 ### Exemple
105 105
106 ```106 ```
107-[warning: llamacpp: nothing answers at http://127.0.0.1:8080/v1 — start it: llama-server -m <model.gguf> --jinja --port 8080]107+[warning: llamacpp: nothing answers at http://host.docker.internal:8080/v1 — start it: llama-server -m <model.gguf> --jinja --port 8080 (add --host 0.0.0.0 when mm runs in a container or a sandbox)]
108 mm v0.1.7 | config: agent.llamacpp.yaml | provider: llamacpp | model: jetbrains/mellum2-12b-a2.5b-instruct-gguf-q4_k_m:Q4_K_M | ctx: unknown | tools: bash, read_file, write_file, edit_file | skills: 0 | yolo: on108 mm v0.1.7 | config: agent.llamacpp.yaml | provider: llamacpp | model: jetbrains/mellum2-12b-a2.5b-instruct-gguf-q4_k_m:Q4_K_M | ctx: unknown | tools: bash, read_file, write_file, edit_file | skills: 0 | yolo: on
109 ```109 ```
110 110
modified docs/fr/reference/configuration.md +2 -2
@@ -197,7 +197,7 @@ tools:
197197 | Fournisseur | `baseUrl` | `fallback` | `apiKeyEnv` | Clé requise |
198198 |-------------|-----------|------------|-------------|-------------|
199199 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | aucune | non |
200-| `llamacpp` | `http://127.0.0.1:8080/v1` | aucun | `LLAMA_API_KEY` | non |
200+| `llamacpp` | `http://127.0.0.1:8080/v1` | `http://host.docker.internal:8080/v1` | `LLAMA_API_KEY` | non |
201201 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | non |
202202 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | non |
203203 | `kronk` | `http://127.0.0.1:11435/v1` | `http://host.docker.internal:11435/v1` | `KRONK_TOKEN` | non — seuls les modes `authenticated` et `full-protected` de Kronk la vérifient |
@@ -477,7 +477,7 @@ Un avertissement au démarrage, et non une erreur, est affiché quand `context.e
477477 | Fichier | Rôle |
478478 |---------|------|
479479 | `agent.yaml` | Configuration Docker Model Runner : `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression désactivée, un prompt système nommé « Bob » qui décrit les skills et les règles d'édition de fichiers. |
480-| `agent.llamacpp.yaml` | Configuration llama.cpp : `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, `fallback: ""`, compression activée. |
480+| `agent.llamacpp.yaml` | Configuration llama.cpp : `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, compression activée. |
481481
482482 ### Exemple
483483
@@ -197,7 +197,7 @@ tools:
197 | Fournisseur | `baseUrl` | `fallback` | `apiKeyEnv` | Clé requise |197 | Fournisseur | `baseUrl` | `fallback` | `apiKeyEnv` | Clé requise |
198 |-------------|-----------|------------|-------------|-------------|198 |-------------|-----------|------------|-------------|-------------|
199 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | aucune | non |199 | `dmr` | `http://localhost:12434/engines/v1` | `http://host.docker.internal:12434/engines/v1` | aucune | non |
200-| `llamacpp` | `http://127.0.0.1:8080/v1` | aucun | `LLAMA_API_KEY` | non |200+| `llamacpp` | `http://127.0.0.1:8080/v1` | `http://host.docker.internal:8080/v1` | `LLAMA_API_KEY` | non |
201 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | non |201 | `llmman` | `http://127.0.0.1:17434/v1` | `http://host.docker.internal:17434/v1` | `LLMMAN_API_KEY` | non |
202 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | non |202 | `ollama` | `http://localhost:11434/v1` | `http://host.docker.internal:11434/v1` | `OLLAMA_API_KEY` | non |
203 | `kronk` | `http://127.0.0.1:11435/v1` | `http://host.docker.internal:11435/v1` | `KRONK_TOKEN` | non — seuls les modes `authenticated` et `full-protected` de Kronk la vérifient |203 | `kronk` | `http://127.0.0.1:11435/v1` | `http://host.docker.internal:11435/v1` | `KRONK_TOKEN` | non — seuls les modes `authenticated` et `full-protected` de Kronk la vérifient |
@@ -477,7 +477,7 @@ Un avertissement au démarrage, et non une erreur, est affiché quand `context.e
477 | Fichier | Rôle |477 | Fichier | Rôle |
478 |---------|------|478 |---------|------|
479 | `agent.yaml` | Configuration Docker Model Runner : `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression désactivée, un prompt système nommé « Bob » qui décrit les skills et les règles d'édition de fichiers. |479 | `agent.yaml` | Configuration Docker Model Runner : `provider: dmr`, `maxTurns: 40`, `displayCommands: true`, compression désactivée, un prompt système nommé « Bob » qui décrit les skills et les règles d'édition de fichiers. |
480-| `agent.llamacpp.yaml` | Configuration llama.cpp : `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, `fallback: ""`, compression activée. |480+| `agent.llamacpp.yaml` | Configuration llama.cpp : `provider: llamacpp`, `baseUrl: http://127.0.0.1:8080/v1`, compression activée. |
481 481
482 ### Exemple482 ### Exemple
483 483
modified internal/agentinit/decide_test.go +1 -1
@@ -170,7 +170,7 @@ func TestRenderRecordsTheDefaultURLs(t *testing.T) {
170170 t.Errorf("llmman's default URLs missing:\n%s", llmman)
171171 }
172172 llamacpp := Render(Decide(p, Choices{Provider: "llamacpp", Model: "qwen3", Usage: "assistant", Machine: "small"}, Facts{}))
173- if !strings.Contains(llamacpp, "# baseUrl: \"http://127.0.0.1:8080/v1\"\n# (llamacpp has no fallback address)\n") {
173+ if !strings.Contains(llamacpp, "# baseUrl: \"http://127.0.0.1:8080/v1\"\n# fallback: \"http://host.docker.internal:8080/v1\" # tried when baseUrl does not answer\n") {
174174 t.Errorf("llamacpp's default URL missing:\n%s", llamacpp)
175175 }
176176 own := Render(Decide(p, Choices{Provider: "llmman", BaseURL: "http://box:17434/v1", Model: "qwen3", Usage: "assistant", Machine: "small"}, Facts{}))
@@ -170,7 +170,7 @@ func TestRenderRecordsTheDefaultURLs(t *testing.T) {
170 t.Errorf("llmman's default URLs missing:\n%s", llmman)170 t.Errorf("llmman's default URLs missing:\n%s", llmman)
171 }171 }
172 llamacpp := Render(Decide(p, Choices{Provider: "llamacpp", Model: "qwen3", Usage: "assistant", Machine: "small"}, Facts{}))172 llamacpp := Render(Decide(p, Choices{Provider: "llamacpp", Model: "qwen3", Usage: "assistant", Machine: "small"}, Facts{}))
173- if !strings.Contains(llamacpp, "# baseUrl: \"http://127.0.0.1:8080/v1\"\n# (llamacpp has no fallback address)\n") {173+ if !strings.Contains(llamacpp, "# baseUrl: \"http://127.0.0.1:8080/v1\"\n# fallback: \"http://host.docker.internal:8080/v1\" # tried when baseUrl does not answer\n") {
174 t.Errorf("llamacpp's default URL missing:\n%s", llamacpp)174 t.Errorf("llamacpp's default URL missing:\n%s", llamacpp)
175 }175 }
176 own := Render(Decide(p, Choices{Provider: "llmman", BaseURL: "http://box:17434/v1", Model: "qwen3", Usage: "assistant", Machine: "small"}, Facts{}))176 own := Render(Decide(p, Choices{Provider: "llmman", BaseURL: "http://box:17434/v1", Model: "qwen3", Usage: "assistant", Machine: "small"}, Facts{}))
modified internal/engine/openai_compat.go +6 -1
@@ -118,11 +118,16 @@ func init() {
118118 register(&openaiCompat{
119119 name: "llamacpp",
120120 baseURL: "http://127.0.0.1:8080/v1", // llama-server's default host:port, plus the OpenAI prefix
121+ // From a container or a sandbox, like the other local servers
122+ // (ticket 217). llama-server listens on the loopback only unless
123+ // started with --host 0.0.0.0, which Linux and a Docker Sandbox need
124+ // for this name to reach it (Docker Desktop reaches the loopback).
125+ fallback: "http://host.docker.internal:8080/v1",
121126 // llama-server only checks a key when started with --api-key; most
122127 // local set-ups are not, so an unset LLAMA_API_KEY is normal.
123128 apiKeyEnv: "LLAMA_API_KEY",
124129 dummyKey: "not-needed",
125- startHint: "start it: llama-server -m <model.gguf> --jinja --port 8080",
130+ startHint: "start it: llama-server -m <model.gguf> --jinja --port 8080 (add --host 0.0.0.0 when mm runs in a container or a sandbox)",
126131 notFoundHint: "llama-server serves one model; check baseUrl ends with /v1 (model: %s)",
127132 probe: probeLlamaCpp,
128133 progress: true,
@@ -118,11 +118,16 @@ func init() {
118 register(&openaiCompat{118 register(&openaiCompat{
119 name: "llamacpp",119 name: "llamacpp",
120 baseURL: "http://127.0.0.1:8080/v1", // llama-server's default host:port, plus the OpenAI prefix120 baseURL: "http://127.0.0.1:8080/v1", // llama-server's default host:port, plus the OpenAI prefix
121+ // From a container or a sandbox, like the other local servers
122+ // (ticket 217). llama-server listens on the loopback only unless
123+ // started with --host 0.0.0.0, which Linux and a Docker Sandbox need
124+ // for this name to reach it (Docker Desktop reaches the loopback).
125+ fallback: "http://host.docker.internal:8080/v1",
121 // llama-server only checks a key when started with --api-key; most126 // llama-server only checks a key when started with --api-key; most
122 // local set-ups are not, so an unset LLAMA_API_KEY is normal.127 // local set-ups are not, so an unset LLAMA_API_KEY is normal.
123 apiKeyEnv: "LLAMA_API_KEY",128 apiKeyEnv: "LLAMA_API_KEY",
124 dummyKey: "not-needed",129 dummyKey: "not-needed",
125- startHint: "start it: llama-server -m <model.gguf> --jinja --port 8080",130+ startHint: "start it: llama-server -m <model.gguf> --jinja --port 8080 (add --host 0.0.0.0 when mm runs in a container or a sandbox)",
126 notFoundHint: "llama-server serves one model; check baseUrl ends with /v1 (model: %s)",131 notFoundHint: "llama-server serves one model; check baseUrl ends with /v1 (model: %s)",
127 probe: probeLlamaCpp,132 probe: probeLlamaCpp,
128 progress: true,133 progress: true,
modified internal/engine/provider_test.go +11 -6
@@ -108,10 +108,13 @@ func TestResolveURLPrecedence(t *testing.T) {
108108 {"an explicit empty fallback disables it", "dmr", config.Config{BaseURL: down, Fallback: ptr("")}, nil, down, "not-needed"},
109109 {"AGENT_BASE_URL wins without probing", "dmr", config.Config{BaseURL: up.URL, Fallback: ptr("")}, map[string]string{"AGENT_BASE_URL": down}, down, "not-needed"},
110110 {"DMR_BASE_URL still works for dmr", "dmr", config.Config{BaseURL: up.URL, Fallback: ptr("")}, map[string]string{"DMR_BASE_URL": down}, down, "not-needed"},
111- {"llamacpp default is llama-server's port", "llamacpp", config.Config{}, nil, "http://127.0.0.1:8080/v1", "not-needed"},
112- {"llamacpp ignores DMR_BASE_URL", "llamacpp", config.Config{}, map[string]string{"DMR_BASE_URL": down}, "http://127.0.0.1:8080/v1", "not-needed"},
113- {"llamacpp reads LLAMA_API_KEY when set", "llamacpp", config.Config{}, map[string]string{"LLAMA_API_KEY": "secret"}, "http://127.0.0.1:8080/v1", "secret"},
114- {"apiKeyEnv names another variable", "llamacpp", config.Config{APIKeyEnv: "MY_KEY"}, map[string]string{"MY_KEY": "k2"}, "http://127.0.0.1:8080/v1", "k2"},
111+ // llamacpp falls back on the host like the other local servers
112+ // since ticket 217 (its default address: TestDefaultURLs).
113+ {"llamacpp falls back to the host's llama-server", "llamacpp", config.Config{BaseURL: down}, nil, "http://host.docker.internal:8080/v1", "not-needed"},
114+ {"llamacpp keeps a reachable baseUrl", "llamacpp", config.Config{BaseURL: up.URL}, nil, up.URL, "not-needed"},
115+ {"llamacpp ignores DMR_BASE_URL", "llamacpp", config.Config{BaseURL: up.URL}, map[string]string{"DMR_BASE_URL": down}, up.URL, "not-needed"},
116+ {"llamacpp reads LLAMA_API_KEY when set", "llamacpp", config.Config{BaseURL: up.URL}, map[string]string{"LLAMA_API_KEY": "secret"}, up.URL, "secret"},
117+ {"apiKeyEnv names another variable", "llamacpp", config.Config{BaseURL: up.URL, APIKeyEnv: "MY_KEY"}, map[string]string{"MY_KEY": "k2"}, up.URL, "k2"},
115118 {"llmman falls back to the host's daemon", "llmman", config.Config{BaseURL: down}, nil, "http://host.docker.internal:17434/v1", "not-needed"},
116119 {"llmman keeps a reachable baseUrl", "llmman", config.Config{BaseURL: up.URL}, nil, up.URL, "not-needed"},
117120 {"llmman reads LLMMAN_API_KEY when set", "llmman", config.Config{BaseURL: up.URL}, map[string]string{"LLMMAN_API_KEY": "k3", "LLAMA_API_KEY": "secret"}, up.URL, "k3"},
@@ -445,7 +448,9 @@ func TestConfigCompat(t *testing.T) {
445448 }
446449
447450 config.Cfg = saved
448- llama := write("llama.yaml", "provider: llamacpp\nmodel: qwen2.5-coder\ncontextWindow: 16384\n")
451+ // fallback: "" keeps the default address whether llama-server runs or
452+ // not (it has a fallback since ticket 217).
453+ llama := write("llama.yaml", "provider: llamacpp\nmodel: qwen2.5-coder\ncontextWindow: 16384\nfallback: \"\"\n")
449454 if _, err := config.Load(llama); err != nil {
450455 t.Fatalf("Load(llamacpp config): %v", err)
451456 }
@@ -559,7 +564,7 @@ func TestDefaultURLs(t *testing.T) {
559564 {"ollama", "http://localhost:11434/v1", "http://host.docker.internal:11434/v1"},
560565 {"kronk", "http://127.0.0.1:11435/v1", "http://host.docker.internal:11435/v1"},
561566 {"dmr", "http://localhost:12434/engines/v1", "http://host.docker.internal:12434/engines/v1"},
562- {"llamacpp", "http://127.0.0.1:8080/v1", ""},
567+ {"llamacpp", "http://127.0.0.1:8080/v1", "http://host.docker.internal:8080/v1"},
563568 {"openai", "https://api.openai.com/v1", ""},
564569 }
565570 for _, c := range cases {
@@ -108,10 +108,13 @@ func TestResolveURLPrecedence(t *testing.T) {
108 {"an explicit empty fallback disables it", "dmr", config.Config{BaseURL: down, Fallback: ptr("")}, nil, down, "not-needed"},108 {"an explicit empty fallback disables it", "dmr", config.Config{BaseURL: down, Fallback: ptr("")}, nil, down, "not-needed"},
109 {"AGENT_BASE_URL wins without probing", "dmr", config.Config{BaseURL: up.URL, Fallback: ptr("")}, map[string]string{"AGENT_BASE_URL": down}, down, "not-needed"},109 {"AGENT_BASE_URL wins without probing", "dmr", config.Config{BaseURL: up.URL, Fallback: ptr("")}, map[string]string{"AGENT_BASE_URL": down}, down, "not-needed"},
110 {"DMR_BASE_URL still works for dmr", "dmr", config.Config{BaseURL: up.URL, Fallback: ptr("")}, map[string]string{"DMR_BASE_URL": down}, down, "not-needed"},110 {"DMR_BASE_URL still works for dmr", "dmr", config.Config{BaseURL: up.URL, Fallback: ptr("")}, map[string]string{"DMR_BASE_URL": down}, down, "not-needed"},
111- {"llamacpp default is llama-server's port", "llamacpp", config.Config{}, nil, "http://127.0.0.1:8080/v1", "not-needed"},111+ // llamacpp falls back on the host like the other local servers
112- {"llamacpp ignores DMR_BASE_URL", "llamacpp", config.Config{}, map[string]string{"DMR_BASE_URL": down}, "http://127.0.0.1:8080/v1", "not-needed"},112+ // since ticket 217 (its default address: TestDefaultURLs).
113- {"llamacpp reads LLAMA_API_KEY when set", "llamacpp", config.Config{}, map[string]string{"LLAMA_API_KEY": "secret"}, "http://127.0.0.1:8080/v1", "secret"},113+ {"llamacpp falls back to the host's llama-server", "llamacpp", config.Config{BaseURL: down}, nil, "http://host.docker.internal:8080/v1", "not-needed"},
114- {"apiKeyEnv names another variable", "llamacpp", config.Config{APIKeyEnv: "MY_KEY"}, map[string]string{"MY_KEY": "k2"}, "http://127.0.0.1:8080/v1", "k2"},114+ {"llamacpp keeps a reachable baseUrl", "llamacpp", config.Config{BaseURL: up.URL}, nil, up.URL, "not-needed"},
115+ {"llamacpp ignores DMR_BASE_URL", "llamacpp", config.Config{BaseURL: up.URL}, map[string]string{"DMR_BASE_URL": down}, up.URL, "not-needed"},
116+ {"llamacpp reads LLAMA_API_KEY when set", "llamacpp", config.Config{BaseURL: up.URL}, map[string]string{"LLAMA_API_KEY": "secret"}, up.URL, "secret"},
117+ {"apiKeyEnv names another variable", "llamacpp", config.Config{BaseURL: up.URL, APIKeyEnv: "MY_KEY"}, map[string]string{"MY_KEY": "k2"}, up.URL, "k2"},
115 {"llmman falls back to the host's daemon", "llmman", config.Config{BaseURL: down}, nil, "http://host.docker.internal:17434/v1", "not-needed"},118 {"llmman falls back to the host's daemon", "llmman", config.Config{BaseURL: down}, nil, "http://host.docker.internal:17434/v1", "not-needed"},
116 {"llmman keeps a reachable baseUrl", "llmman", config.Config{BaseURL: up.URL}, nil, up.URL, "not-needed"},119 {"llmman keeps a reachable baseUrl", "llmman", config.Config{BaseURL: up.URL}, nil, up.URL, "not-needed"},
117 {"llmman reads LLMMAN_API_KEY when set", "llmman", config.Config{BaseURL: up.URL}, map[string]string{"LLMMAN_API_KEY": "k3", "LLAMA_API_KEY": "secret"}, up.URL, "k3"},120 {"llmman reads LLMMAN_API_KEY when set", "llmman", config.Config{BaseURL: up.URL}, map[string]string{"LLMMAN_API_KEY": "k3", "LLAMA_API_KEY": "secret"}, up.URL, "k3"},
@@ -445,7 +448,9 @@ func TestConfigCompat(t *testing.T) {
445 }448 }
446 449
447 config.Cfg = saved450 config.Cfg = saved
448- llama := write("llama.yaml", "provider: llamacpp\nmodel: qwen2.5-coder\ncontextWindow: 16384\n")451+ // fallback: "" keeps the default address whether llama-server runs or
452+ // not (it has a fallback since ticket 217).
453+ llama := write("llama.yaml", "provider: llamacpp\nmodel: qwen2.5-coder\ncontextWindow: 16384\nfallback: \"\"\n")
449 if _, err := config.Load(llama); err != nil {454 if _, err := config.Load(llama); err != nil {
450 t.Fatalf("Load(llamacpp config): %v", err)455 t.Fatalf("Load(llamacpp config): %v", err)
451 }456 }
@@ -559,7 +564,7 @@ func TestDefaultURLs(t *testing.T) {
559 {"ollama", "http://localhost:11434/v1", "http://host.docker.internal:11434/v1"},564 {"ollama", "http://localhost:11434/v1", "http://host.docker.internal:11434/v1"},
560 {"kronk", "http://127.0.0.1:11435/v1", "http://host.docker.internal:11435/v1"},565 {"kronk", "http://127.0.0.1:11435/v1", "http://host.docker.internal:11435/v1"},
561 {"dmr", "http://localhost:12434/engines/v1", "http://host.docker.internal:12434/engines/v1"},566 {"dmr", "http://localhost:12434/engines/v1", "http://host.docker.internal:12434/engines/v1"},
562- {"llamacpp", "http://127.0.0.1:8080/v1", ""},567+ {"llamacpp", "http://127.0.0.1:8080/v1", "http://host.docker.internal:8080/v1"},
563 {"openai", "https://api.openai.com/v1", ""},568 {"openai", "https://api.openai.com/v1", ""},
564 }569 }
565 for _, c := range cases {570 for _, c := range cases {