Skip to Content (Press Enter)
Community › Forge › Assets list › OKF Agent
Login to follow
OKF Agent

OKF Agent (ODC)

Stable version 0.1.0 (Compatible with ODC)
Uploaded on 3 Aug (2 weeks ago) by Michael Guzman
OKF Agent

OKF Agent (ODC)

Details
Detailed Description

OKF Agent is an ODC AI Agent app that answers questions over an OKF v0.2 bundle ingested by OKF Store, using an OpenAI model connection and a tool-calling interface.

The agent is wired to six tools, each a thin wrapper over an OKF Store service action: browsing bundles and their concept index, retrieving a concept by path, listing concepts filtered by type, status, staleness, and trust tier, and traversing links in both directions. The trust tier filter is exposed to the model as plain text and translated to OKF Store's underlying identifier before the call is made.

The agent's system prompt is not stored locally. It is fetched from a separate app, PromptAssembler, on every message. The prompt itself, and a set of test questions for exercising the agent, ship with this component as reference files.

A caller sends a message and a session identifier to CallOKFAgent and receives the agent's response. The most recent call's tool-call trace can be retrieved separately for debugging.

Limitations
  • OKF Agent has no screens or UI of its own. It is consumed entirely through its service actions; a consuming app or flow must be built separately to call it.
  • Trace data reflects only the most recent call. The Trace entity is cleared at the start of every invocation, so DownloadLatestTrace cannot retrieve history from an earlier call once a new one has run.
  • The system prompt is fetched from PromptAssembler on every message, with no caching. This adds a call on each turn and means the agent's behavior depends on PromptAssembler being installed and reachable.
  • The agent has no built-in way to search across bundles. Every tool call operates within one BundleId; a caller must already know which bundle to query, typically via ListBundles first.
  • Response output is free text. No structured output format is configured, so responses are not validated against a schema.