Knowledge Graphs

Decisions, code, and docs in one queryable graph.

Connect decisions, code, and documentation in a queryable graph. Understand how your project's knowledge is interconnected.

Graph tiers by plan

Pick the depth you need.

Pro · Graph-Lite

Module-level import graph

Supports dependencies and 1-hop impact for files/modules. No call/dataflow/type layers.

Elite / Team · Full Graph

Full multi-layer graph

Module + call + dataflow + type with deeper impact analysis and graph ingestion via graph(action="ingest").

Search + graph workflow

Find a target with search, then traverse.

Use search to identify the best target, then graph queries to answer dependency and impact questions. Keeps prompts small and avoids unnecessary local scans.

terminal · search → graph

search(mode="hybrid", query="token refresh flow", limit=3)
graph(action="impact", target={ "type": "module", "path": "src/auth.rs" }, max_depth=2)

What are knowledge graphs?

Nodes + edges, queryable.

Knowledge graphs represent information as nodes (entities) connected by edges (relationships). In ContextStream, knowledge graphs help you:

Track Dependencies

See how decisions affect other parts of your system.

Connect Context

Link related memories, code, and documentation.

Trace Impact

Understand what changes when you modify something.

Discover Insights

Find non-obvious connections in your knowledge base.

Creating knowledge nodes

Capture durable context first.

Knowledge graph nodes come from the same durable context you already store in ContextStream: decisions, lessons, docs, tasks, todos, and facts. Use the MCP tools to create those records, then query the graph relationships around them.

terminal · memory · create_node

memory(action="create_node",
  node_type="decision",
  title="Use PostgreSQL for primary database",
  content="Chose PostgreSQL for JSON support and operational maturity.",
  project_id="<project-id>")

Connecting nodes

Edges express relationships.

ContextStream derives useful relationships from stored context and indexed code. Query those links directly instead of manually assembling raw request payloads.

terminal · graph · related

graph(action="related",
  query="PostgreSQL database decision and schema design",
  max_depth=2)

Common relation types:

influencesdepends_onsupersedesrelated_toimplements

Querying the graph

Traverse, filter, and walk paths.

Find related knowledge nodes:

terminal · graph · related

graph(action="related",
  query="PostgreSQL decision",
  max_depth=2,
  relation_types=["influences", "depends_on"])

Get all decisions in a workspace:

terminal · memory · decisions

memory(action="decisions", query="database architecture")

Code graph analysis

Files, functions, types, variables.

Code graph endpoints analyze indexed code entities. They are separate from knowledge graph endpoints; memory node types like decision or fact are not valid here. Use /api/v1/graph/knowledge/path with UUID node IDs (no type) for knowledge graph queries.

Pro Graph-Lite supports module-level targets only (files/paths) and 1-hop impact. Elite/Team unlocks function/type/variable layers and deeper impact analysis.

Accepted code entity types (v0.4.x consolidated graph tool):

graph(action="dependencies") target.type: module (file/path), function (method), type (struct/enum/trait/class), variable (data/const/constant)

graph(action="call_path") source.type/target.type: function (method) only (type optional; ids are function IDs)

graph(action="impact") target.type: module (file/path), function (method), type (struct/enum/trait/class), variable (data/const/constant)

Type payload examples (code graph only):

module/file/path: {"target":{"type":"module","path":"src/auth.rs"}}function/method: {"target":{"type":"function","id":"fn_login"}}type/struct/enum/trait/class: {"target":{"type":"type","id":"User"}}variable/data/const/constant: {"target":{"type":"variable","id":"API_KEY"}}

Dependencies example:

terminal · graph · dependencies

graph(action="dependencies",
  target={ "type": "module", "path": "src/auth.rs" },
  max_depth=3)

Call path example:

terminal · graph · call_path

graph(action="call_path",
  source={ "id": "fn_login", "type": "function" },
  target={ "id": "fn_verify_token", "type": "function" },
  max_depth=5)

Impact analysis example:

terminal · graph · impact

graph(action="impact",
  change_type="modify_signature",
  target={ "id": "User", "type": "type" })

Graph ingestion (Elite/Team full graph):

terminal · graph · ingest

graph(action="ingest", project_id="<project-id>", wait=true)

Knowledge graph path example (memory node UUIDs):

terminal · graph · path

graph(action="path",
  source={ "id": "7e3b0b5e-2c76-4f49-8e5a-2f2bfe7eb7f3" },
  target={ "id": "5c3a6b1a-0f89-4ddc-bb29-0a2a9a3c2c4e" },
  max_depth=5)

Next steps