Mission
Mission: Agent thread storage
Section titled “Mission: Agent thread storage”Build a clear storage architecture for Gio’s own Vercel AI SDK / Eve-style agent app by studying how real agent systems persist threads, turns, branches, summaries, and delegated agents. The goal is not just to understand the internals of Codex, OpenCode, and Pi, but to extract a design that is practical to implement in Gio’s app.
Success looks like
Section titled “Success looks like”- Explain the storage model used by Codex, OpenCode, and Pi in concrete terms.
- Identify the key entities, relationships, and SQL tables involved in durable agent threads.
- Understand how branching, compaction, and delegated/sub-agent activity should be represented.
- Finish with a recommended schema and architecture Gio can implement for a Vercel AI SDK / Eve agent.
Constraints
Section titled “Constraints”- Use the local teach-workspace structure for this topic.
- Ground claims in source code and repo docs, not vague summaries.
- Focus on thread storage, entities, schema, lineage, compaction, and sub-agent behavior.
- Prefer patterns that fit a modern app stack with SQL storage and streamed agent turns.
Out of scope
Section titled “Out of scope”- Training or fine-tuning models.
- Frontend chat UI details beyond what the storage model requires.
- Deep provider-specific transport details unless they affect storage design.