Configuration
Navi Agent reads YAML configuration from ~/.navi-agent/config.yaml by
default. Environment variables override values from the file.
Configuration file
model:
name: gpt-4o-mini
api_key: replace-with-your-api-key
base_url: https://api.openai.com/v1
context_limit_tokens: 128000
runtime:
max_iterations: 30
web:
search_api_key: replace-with-your-brave-search-api-key
gateway:
weixin:
token: replace-with-your-weixin-token
account_id: replace-with-your-weixin-account-id
base_url: https://ilinkai.weixin.qq.com
poll_interval_seconds: 1.0
dm_policy: pairing
allowed_users: []
telemetry:
langfuse:
enabled: false
public_key: replace-with-your-langfuse-public-key
secret_key: replace-with-your-langfuse-secret-key
host: https://cloud.langfuse.com
Environment variables
| Variable | Purpose |
|---|---|
NAVI_HOME |
Replace the default ~/.navi-agent data directory |
NAVI_PROFILE |
Use ~/.navi-agent/profiles/<name> as an isolated profile |
NAVI_MODEL |
Model name |
NAVI_API_KEY |
Model API key |
NAVI_BASE_URL |
OpenAI-compatible API base URL |
NAVI_CONTEXT_LIMIT_TOKENS |
Context window limit used by compaction |
NAVI_MAX_ITERATIONS |
Maximum runtime iterations per request |
NAVI_WEB_SEARCH_API_KEY |
Brave Search API key |
NAVI_WEIXIN_TOKEN |
Weixin gateway token |
NAVI_WEIXIN_ACCOUNT_ID |
Weixin gateway account ID |
NAVI_WEIXIN_BASE_URL |
Weixin gateway API endpoint |
NAVI_WEIXIN_POLL_INTERVAL_SECONDS |
Gateway polling interval |
NAVI_WEIXIN_DM_POLICY |
Direct-message access policy |
NAVI_WEIXIN_ALLOWED_USERS |
Comma-separated allowed user IDs |
NAVI_LANGFUSE_ENABLED |
Enable Langfuse export |
LANGFUSE_PUBLIC_KEY |
Langfuse public key |
LANGFUSE_SECRET_KEY |
Langfuse secret key |
LANGFUSE_HOST |
Langfuse endpoint |
OPENAI_API_KEY, OPENAI_BASE_URL, BRAVE_SEARCH_API_KEY, and the unprefixed
Weixin variables are also accepted as fallbacks.
Runtime data
The Navi home directory contains configuration and persistent state: