Give your agents a map, not a context dump.

Agentic coding made context a bottleneck. Damper's context graph models your project as typed nodes and relationships — with authority and confidence on every edge — so an agent resolving a task knows which rules are hard, which docs are hints, and where to stop reading.

Typed structure

Everything that shapes a change becomes a node.

Not files and folders — meaning. Routes, path patterns, modules, surfaces, decisions, critical rules and features each exist as a typed node, connected by relationships an agent can reason about instead of guessing from keyword search.

  • Typed relations — governs, depends_on, implements, documented_by, supersedes and more.
  • Express what routes can't — “this decision governs every checkout change” is one edge, not a README paragraph.
  • Additive by design — the graph runs alongside your existing context docs. Nothing to migrate.
Honest retrieval

Authority and confidence on every edge.

A context dump flattens everything into equally-plausible text. The graph keeps the differences that matter: an edge is mandatory, advisory or generated, and carries a confidence score from 0 to 1. Agents can tell a hard rule from a hunch — and so can you, when you audit what they read.

  • Explainable — every edge has a source. No black-box retrieval deciding what your agent believes.
  • Trust, calibrated — ship-generated hints don't masquerade as human mandates.
Bounded traversal

One call from “what changed” to “what governs it”.

An agent working a task calls resolve_context_graph from behind the change. The walk runs to a capped depth and returns ranked candidates — never an unbounded dump, never a token explosion.

  1. Add a relationship

    Model the constraint your route patterns can't express.

  2. Ask for context behind a change

    Resolve from a file, route or module node.

  3. Compare before you trust

    Run it in shadow mode against your existing context routing first.

See the MCP context tools

Your agents are already reading. Give them something true.

Model the graph once, and every agent — in every session, on every machine — starts with the map.