Graphify: Query the Team Knowledge Base from Any Repo

graphify turns this knowledge-base repo into a navigable knowledge graph. Instead of grepping or reading raw Markdown, your AI agent asks a question and graphify returns a small, relevant sub-graph of doc nodes and the edges between them — roughly 70x fewer tokens per answer, and it surfaces cross-document connections you would not have thought to search for.

The graph is committed in this repo at graphify-out/graph.json, so everyone who clones knowledge-base gets the same map with no rebuild and no API cost.

Setup

Run the team opencode bootstrap once (opencode-ai-setup) and answer Y to the graphify step. It:

installs graphify and wires it into opencode,
registers this repo’s graph into a global graph so it is reachable from any repo,
installs the graphify-kb-link skill (lets the agent link the KB graph into whatever project you are in),
installs a git hook that auto-rebuilds the graph after commits (free, no LLM).

Using it in opencode

In opencode, run the /graphify command with your question — or just ask the agent a docs question in plain language and it does this for you. No paths, no flags:

/graphify query "how does the relay bot store tokens?"

The graphify skill checks for a graph in the current project. If you are not inside the knowledge-base repo, it links the shared KB graph first (via the graphify-kb-link skill), then runs the query and answers from the result.

Other commands work the same way:

/graphify explain "Bedrock Knowledge Base"
/graphify path "Relay Bot Architecture and Operations" "Bedrock Knowledge Base"

Worked example: from the `hh-server` repo

This is a real opencode session (model: MiniMax M3) run from the hh-server repo — a Rails app that has nothing to do with the docs. The user typed one command:

/graphify query "how does the relay bot store tokens?"

opencode linked the knowledge-base graph and ran the query, returning a 13-node sub-graph of the relevant docs:

Traversal: BFS depth=2 | Start: ['Relay Bot Architecture and Operations',
  'Relay CI Stack Pattern', 'relay-eval-ci IAM Role'] | 13 nodes found

NODE Relay Bot Architecture and Operations  [src=src/DevOps/Relay_Bot_Architecture_and_Operations.md community=119]
NODE Slack Bolt App                         [src=src/DevOps/Relay_Bot_Architecture_and_Operations.md community=119]
NODE Amazon Bedrock                         [src=src/DevOps/Relay_Bot_Architecture_and_Operations.md community=119]
NODE Bedrock Knowledge Base                 [src=src/DevOps/Knowledge_Base_DOCX_Drive_Retirement.md community=119]
NODE HungryHub IAM Architecture and Operations [src=src/DevOps/HungryHub_IAM_Architecture_and_Operations.md community=9]
NODE Relay CI Stack Pattern                 [src=src/DevOps/Relay_CI_Stack_Pattern.md community=9]
...
EDGE Relay Bot Architecture and Operations --references--> Amazon Bedrock
EDGE Relay Bot Architecture and Operations --references--> Slack Bolt App
EDGE Knowledge Base DOCX Drive Retirement  --references--> Bedrock Knowledge Base

Then the agent answered, grounded in that sub-graph:

Relay bot lives at src/DevOps/Relay_Bot_Architecture_and_Operations.md (community 119). Token-storage adjacents in the graph: Slack Bolt App (bot token / signing secret source), Amazon Bedrock (retrieval path), Knowledge Base DOCX Drive Retirement (KB token/role handoff). Token handling is detailed inside the relay bot doc (SLACK_BOT_TOKEN / SLACK_SIGNING_SECRET, SSM / Secrets Manager).

One command, from an unrelated repo: the agent found the right doc, the related services, and the likely secret-handling mechanism — without reading any file directly.

opencode commands

Command	Use
`/graphify query "<q>"`	Ask a natural-language question about the docs
`/graphify explain "<node>"`	Plain-language explanation of one node + its neighbors
`/graphify path "<A>" "<B>"`	Show how two concepts connect

Notes

You never pass a path or --graph flag in opencode — the skill resolves the knowledge-base graph for you, linking it into the project first if needed.
The linked graph reflects the knowledge-base checkout on disk; git pull there to pick up teammates’ doc updates.
Most of the time you do not even type /graphify — just ask the agent a docs question in plain language and it runs the query for you.

Keyboard shortcuts