Internals

How Oda works inside: the agent runtime, tools and their catalog, session events and chat log format, storage and builds.

• The Oda agent runtime

  How the Oda agent runs: creating, resuming and disposing agents, their statuses and phases, and the methods that drive a conversation.

• Tools and how they are offered

  How tools are offered to the model and how calls execute: modes, restrictions, guards and the execution pipeline.

• Tool catalog and session events

  The tools DataMind Oda can use, how its work is recorded, and the codes and token counts those records carry.

• Tool availability

  Why some Oda tools appear only in some deployments, and what happens when one is missing.

• Session events and projections

  The durable events behind plan mode, presets, the sandbox and scheduled runs, and the projections read from them.

• Chat log format

  What a chat log contains, how it is read, and how its format is versioned so older logs stay readable, for people who inspect, support or process chat logs.

• How Oda stores workspaces

  Where Oda keeps its workspace list, how chats join a workspace, and what removal, hiding and archiving never touch.

• The suggest-task tool and service

  For Oda developers: the suggestion tool's fields and refusals, the service behind the cards, and the authority rules for started chats.

• The Cordis API catalog

  How the API catalog and its documentation regions are generated, gated and served.

• Building and shipping Oda

  Run, build and package Oda from source, and how builds are identified and published.