Skip to main content
nanograph

Context Graph Example

The canonical RevOps / context-graph example lives in examples/revops/. It models signals, decisions, actions, opportunities, and projects in one graph, and it now showcases:

  • schema @description(...) and @instruction(...)
  • query @description(...) and @instruction(...)
  • checked-in nanograph.toml defaults and aliases
  • lexical and semantic signal search scoped by graph traversal

Setup

cd examples/revops

nanograph init
nanograph load --data revops.jsonl --mode overwrite
nanograph check --query revops.gq

The example config gives you:

  • db.default_path = "omni.nano"
  • query.roots = ["."]
  • aliases like why, trace, value, pipeline, and signals
  • deterministic mock embeddings for Signal.summaryEmbedding

See Project Config for the full config and env model.

Describe the schema

The schema carries agent-oriented metadata on types like Decision and Signal.

nanograph describe --type Signal --format json

That JSON includes:

  • description
  • instruction
  • key_property
  • property descriptions
  • edge summaries and endpoint-key metadata

Useful aliases

Why did an opportunity happen?

nanograph run why opp-stripe-migration

Trace a signal through decision and action:

nanograph run trace sig-hates-vendor

Show value created by a signal:

nanograph run value sig-hates-vendor

Pipeline summary:

nanograph run pipeline

Semantic signal search scoped to one client:

nanograph run signals cli-priya-shah "procurement approval timing"

Query metadata in action

Queries like decision_trace and semantic_signals_for_client include @description(...) and @instruction(...), so table output gives agents extra context before the rows:

Query: decision_trace
Description: Trace an opportunity back to the decision, actor, and signal that created motion.
Instruction: Use as the default 'why did this happen?' query for an opportunity slug.

Why this example matters

This example shows how NanoGraph can represent operational reasoning directly:

  • Signal captures observed evidence
  • Decision records why work should happen
  • Action proves what actually happened
  • graph traversal lets you move across those layers without flattening everything into one table

With the new search fields, you can also ask retrieval-style questions inside that reasoning graph instead of keeping search and graph logic separate.

See also

Star Wars