turbo-editors/configs · agent.mm-dmr-gemma.yaml

⑂ main

agent.mm-dmr-gemma.yaml · 110 lines · 3.8 KBYAML Blame History Raw

🛟 Updated. a3c787d k33g 8h ago	1	# Agent settings.
	2	#
	3	provider: dmr
	4
	5	model: huggingface.co/unsloth/gemma-4-e2b-it-gguf:Q4_K_M
	6	baseUrl: http://localhost:12434/engines/v1
	7	fallback: http://host.docker.internal:12434/engines/v1
	8
	9	bashTool: true
	10	# The built-in file tools: read_file, write_file, edit_file.
	11	editTools: true
	12
	13	lsp:
	14	servers:
	15	- name: gopls
	16	command: gopls
	17	extensions: [".go"]
	18	- name: rust # rustup component add rust-analyzer
	19	command: rust-analyzer
	20	extensions: [".rs"]
	21	# How many tokens the server actually serves.
	22	# 0 = ask the server (llama-server tells it on /props)
	23	contextWindow: 0
	24
	25	# Max number of characters a tool returns to the model.
	26	# Context safeguard: beyond that, the output is truncated (beginning + end kept).
	27	maxOutput: 16000
	28
	29	# Max number of model <-> tools round trips for a single question.
	30	# Raise it for tasks that need long chains of commands.
	31	maxTurns: 40
	32
	33	# Persistent sessions, per project.
	34	# When enabled, every conversation is saved as one JSON file under `dir`.
	35	# `/sessions` lists them, `/resume` continues one; in an editor, "Reload Agent" replays the thread.
	36	sessions:
	37	enabled: true
	38	dir: .mm/sessions
	39
	40	# How many lines of a command's output are echoed to the terminal.
	41	# 0 disables it.
	42	previewLines: 20
	43
	44	displayCommands: true
	45
	46	# The system prompt: what the agent is, and what it is allowed to do.
	47	# This is the knob to play with — it is the shortest path to changing behaviour.
	48	system: \|
	49	Your name is Riker.
	50	You are a coding agent working in a terminal.
	51	You have a "bash" tool to run shell commands.
	52	Use it to explore files, run tests, inspect the repository, etc.
	53	Chain several commands if needed, then answer clearly in English.
	54
	55	FILE EDITING
	56	You have three tools for files: `read_file`, `edit_file` and `write_file`.
	57	They are how a file gets read and changed here: each change is exact,
	58	checked before it is written, and comes back as a diff with line numbers.
	59	bash is for running things — building, testing, listing, searching.
	60
	61	RULES
	62	- Keep everything the file already does, unless the user asked to remove it.
	63	- Touch only the files the request is about. Do not add tests, files or
	64	features that were not asked for.
65	- Never run a git command unless the user says git, commit or push.
66	- Never move, rename or delete a file unless the user asked for it.
67	- Then answer in English, in a few lines.
68	- If you don't know how to use a <cli>, run `<cli> --help` (or `<cli> help`)
69	to understand the options, then run the command.
70
71	BACKGROUND JOBS
72	Never let a command block the answer. Anything that serves, watches or runs
73	long goes to the background, with BOTH streams redirected and its pid kept:
74
75	nohup <command> > /tmp/<job>.log 2>&1 & echo $! > /tmp/<job>.pid
76
77
78	# Generation settings (OpenAI API keys). Low temperature for a coding agent:
79	# we want precise and reproducible answers, not creativity.
80	sampling:
81	temperature: 0.0
82	parallel_tool_calls: false
83	top_p: 0.9
84	max_tokens: 4096
85
86	# llama-server processes the whole prompt before the first token;
87	# on a laptop a 32k context can take a while. Raise it if it fires.
88	watchdogTimeout: 30s
89
90	# Context compression.
91	# ON here: the agent knows the window (read from /props), so `threshold` has a number to apply to.
92	context:
93	enabled: true
94
95	# Compress when the history reaches this share of the window.
96	threshold: 75
97
98	# Fallback on the message count, for when the window is unknown.
99	# One command costs 2 messages (call + response), so 80 is roughly 30 commands of history.
100	maxMessages: 80
101
102	# How many recent questions (with their commands and answers) are kept as-is instead of being summarised.
103	# The model works on the last few turns, so summarising them would make it redo what it just did.
104	keepLastTurns: 3
105
106	# max_tokens of the summary request.
107	summaryMaxTokens: 1200
108
109	showStats: true
110