You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
docs: add per-purpose inference env vars to deploy config
Exposes MEM0_LLM_BASE_URL, MEM0_LLM_API_KEY, MEM0_EMBED_BASE_URL,
MEM0_EMBED_API_KEY, MEM0_GRAPH_LLM_BASE_URL, MEM0_GRAPH_LLM_API_KEY,
and MEM0_GRAPH_EXTRACTION_STRATEGY across all 4 compose files,
.env.example, and README.
All vars are optional and fall back to OPENAI_BASE_URL / OPENAI_API_KEY.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|`QDRANT_COLLECTION`| Collection name for memory vectors |`store` service |`foxmemory`|
99
99
100
+
### Per-purpose inference (optional)
101
+
102
+
Each inference purpose can target a different endpoint. All fall back to `OPENAI_BASE_URL` / `OPENAI_API_KEY` when not set — existing deployments are unaffected.
103
+
104
+
| Variable | Purpose | Fallback | Example |
105
+
|---|---|---|---|
106
+
|`MEM0_LLM_BASE_URL`| Main LLM endpoint |`OPENAI_BASE_URL`|`http://exo:8000/v1`|
107
+
|`MEM0_LLM_API_KEY`| Main LLM API key |`OPENAI_API_KEY`|`not-needed`|
`MEM0_GRAPH_EXTRACTION_STRATEGY` controls how entities and relationships are extracted from text. Use `tool_calling` (default) for models that support OpenAI-style function calling. Use `json_prompting` for models that don't (e.g. small local models served via EXO).
115
+
100
116
Backward-compatible variables still present in some scripts/builds:
0 commit comments