engram
by devinmlowe
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| OLLAMA_HOST | No | Ollama endpoint; used first if reachable | http://localhost:11434 |
| ENGRAM_DB_PATH | No | SQLite database location | |
| ENGRAM_DATA_DIR | No | Root for database, archive, and logs | |
| ENGRAM_LOGS_DIR | No | Log directory | |
| ENGRAM_MCP_HOST | No | Bind address. Anything but loopback is refused unless ENGRAM_MCP_TOKEN is set | 127.0.0.1 |
| ENGRAM_MCP_TOKEN | No | When set, /mcp requires Authorization: Bearer <token> (401 otherwise); /health stays open for supervisors. Required before the daemon will bind anything but loopback | |
| ANTHROPIC_API_KEY | No | Enables the Anthropic provider (final tier of the LLM cascade) | |
| ENGRAM_ARCHIVE_DIR | No | Conversation archive directory | |
| ENGRAM_LOCAL_MODEL | No | Ollama model name (must be pulled on the Ollama host) | qwen2.5:7b |
| OPENROUTER_API_KEY | No | Enables the OpenRouter provider | |
| ENGRAM_HTTP_WORKERS | No | Worker count in HTTP mode (0 = run tool calls inline on the main thread) | 2 |
| ENGRAM_OPENAI_MODEL | No | Model served by that route; setting it activates the `openai` tier | |
| ENGRAM_LLM_PROVIDERS | No | Tier order, comma-separated. A tier left out is never tried; unknown names are ignored with a warning | ollama,openai,openrouter,anthropic |
| ENGRAM_EMBEDDING_DIMS | No | Matryoshka embedding dimensions (must match the existing DB) | 256 |
| ENGRAM_LLM_TIMEOUT_MS | No | Per-request timeout for every tier. Raise it for a slow local model | 120000 |
| ENGRAM_MCP_STANDALONE | No | 1 makes a stdio start run inline even when the daemon is healthy | |
| ENGRAM_RERANK_ENABLED | No | Set to false or 0 to disable the cross-encoder reranker | true |
| ENGRAM_SKIP_PREFLIGHT | No | Set to 1 to silence the npm install platform preflight | |
| ENGRAM_MODEL_CACHE_DIR | No | Where model weights are downloaded/cached; never inside node_modules | |
| ENGRAM_NO_UPDATE_CHECK | No | Set to 1 to disable the automatic daily version lookup and notice | |
| ENGRAM_OPENAI_BASE_URL | No | Generic OpenAI-compatible route: OpenAI, a LiteLLM gateway, a self-hosted server. A bare origin gets `/v1` appended | https://api.openai.com/v1 |
| ENGRAM_OPENROUTER_MODEL | No | OpenRouter model name | google/gemini-2.5-flash-lite |
| ENGRAM_CHUNKING_STRATEGY | No | fixed or adaptive (content-aware boundaries) | fixed |
| ENGRAM_WORKER_TIMEOUT_MS | No | Per-call timeout; remember, remember_batch, index_file_structure and reflect --refresh use higher floors. | 8000 |
| ENGRAM_OPENAI_API_KEY_ENV | No | Name of the environment variable holding the route's credential. Engram reads it at call time and never stores, prints or commits the value | OPENAI_API_KEY |
| ENGRAM_OPENAI_TEMPERATURE | No | Sent only when set; the generic route omits `temperature` by default because many gateways and reasoning models reject it | |
| ENGRAM_CLAUDE_PROJECTS_DIR | No | Where engram sync looks for Claude Code conversations | |
| ENGRAM_FORGET_RETENTION_DAYS | No | Days a forgotten memory is kept before hard-delete | 30 |
| ENGRAM_LOCAL_MODEL_FALLBACKS | No | Comma-separated Ollama models tried in order when ENGRAM_LOCAL_MODEL is not pulled |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Server capabilities have not been inspected yet.
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
No tools | |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
This server cannot be deployed
Maintenance
ActivityMaintained
ResponsivenessResponsive