Zero
OfficialServer Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| KG_DB_NS | No | SurrealDB namespace | kg |
| KG_DB_URL | Yes | SurrealDB WebSocket URL, e.g. ws://127.0.0.1:8000 | |
| KG_DB_NAME | No | SurrealDB database name | main |
| KG_DB_PASS | Yes | SurrealDB password | |
| KG_DB_USER | Yes | SurrealDB username | |
| KG_AUTH_MODE | No | Use local for single-user stdio: one fixed local user, no login. | local |
| KG_JWT_SECRET | Yes | At least 32 characters. Secret for the kg_server SurrealDB sign-in; its sha256 must be stored in the database (docs/signin-secret.md). | |
| EMBEDDING_API_KEY | No | API key when the provider needs one (cohere) | |
| EMBEDDING_BASE_URL | No | Ollama base URL, e.g. http://127.0.0.1:11434 | |
| EMBEDDING_PROVIDER | No | cohere, ollama or fixture | ollama |
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
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| kg_node_createA | Create a knowledge node (observation, decision, problem, approach, ...). Requires an active session. New nodes start fully active. Generates an embedding for semantic search and can link to existing nodes via relates_to. |
| kg_node_updateA | Update the status, content or tags of an existing knowledge node and reactivate it. Tags replace the whole tag list. Changing content overwrites the previous content and regenerates the embedding; if that fails the update is rejected and the node is unchanged. Requires an active session. |
| kg_queryA | Search knowledge by semantic similarity, type, status, or recency. Returns nodes ranked by relevance. Use spread=true to find related nodes via spreading activation. With an active session, returned nodes are reactivated (their usage-based decay score is boosted; node content is not changed); pass read_only=true to search without changing anything. |
| kg_relateA | Create a relationship between two knowledge nodes, or reinforce it if it already exists (the edge weight increases). Both nodes are reactivated (their decay scores are boosted). Requires an active session; limited to 100 edges per session. |
| kg_session_startA | Start a new working-memory session. Required before creating nodes or edges. Applies decay to the usage-based scores of existing nodes (node content is not changed), and can optionally create the named project (plus the default workspace records on first use) if it does not exist. |
| kg_session_endA | End the current session. Nodes and edges are kept; a new session must be started before further writes. |
| kg_session_infoA | Get details of the current session (number, start time, project, notes) and node counts by type. Read-only. |
| kg_statusA | Get an overview of current working memory: open todos and problems, recent decisions and approaches, node count, and active session info. Read-only. Optionally filter by project. |
| kg_meta_modelA | Read the stored meta-model of a container, project or person: methodology patterns, resiliency metrics, or process flows depending on lens. Models are computed ahead of time (nightly on the hosted service), not on request; a scope with no stored model yet returns status "not_computed" and "computed_at" says how fresh a model is. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 9 tools
Most tools target clearly distinct operations: node create/update/query, edge relate, and session lifecycle. The main overlap is between kg_status and kg_session_info, both of which report session information, and kg_query's optional reactivation side effect could be mistaken for a pure read. Descriptions do enough to separate them, but the boundaries are not perfectly crisp.
All tools share a consistent kg_ prefix, and action tools follow a predictable verb_noun pattern (kg_node_create, kg_node_update, kg_session_start, kg_relate). A few tools are noun-oriented labels (kg_status, kg_session_info, kg_meta_model), which is a minor deviation but still readable and coherent.
Nine tools is well-scoped for a knowledge-graph memory server, covering nodes, relationships, sessions, status, and meta-model access without obvious padding. Each tool appears to earn its place.
The set covers creation, update, search, and relationship creation for knowledge nodes, plus session lifecycle. However, there is no explicit deletion or forgetting operation for nodes or edges, and no direct node retrieval beyond query. These are notable gaps for a knowledge-graph domain where cleanup and precise retrieval matter.