← all repositories
aoci-spec/aoci-code

Git-versioned context that survives agent reboots

AOCI-CODE builds a persistent, Git-versioned plain-text index of your codebase and database schema so coding agents read system context once instead of grepping the repo on every task.

aoci-code
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does

This is a local-first Go tool that sits between your codebase and AI coding agents via the Model Context Protocol. It governs the creation of a dense, plain-text index—one entry per file—that records responsibilities, dependencies, public contracts, and hidden constraints agents must not get wrong. The index lives in the repository, is versioned by Git, and is read by the agent before it acts, eliminating the need to re-scan hundreds of thousands of lines on every new task or conversation.

The interesting bit

The index is not a static summary but a living artifact maintained by the agent itself: after each task, the MCP server detects what changed and prompts the agent to update the affected entries. This creates a feedback loop where the agent writes its own long-term memory in a structured format—complete with database schema entries—so any later agent or session can pick up exactly where the last one left off.

Key highlights

  • One-line entries per file use a dense FRAS notation (Function, Related, API, Safety) plus layer tags, allowing an agent to ingest a 700,000-line commercial system in roughly 300K tokens.
  • Supports code and database cognition together: PostgreSQL, MySQL, and constrained openGauss 6.0.5 schemas can be indexed alongside the codebase.
  • Runs entirely offline with no stored credentials; database connections are made via named environment variables for catalog metadata only, and business data is never touched.
  • The index is plain text and portable across agents—Claude Code, Codex, Cursor, OpenCode, or any host that speaks stdio MCP.

Caveats

  • Building the first index is deliberately slow—about an hour per 200,000 lines—and the project is still at v0.1.0-rc14.
  • Host compatibility depends on MCP stdio support and adherence to the tool contract; a model name alone does not guarantee it will work.

Verdict

Teams running large or long-lived codebases with AI agents should look here; if you want a drop-in autocomplete or one-shot generator, this is the wrong tool.

Frequently asked

What is aoci-spec/aoci-code?
AOCI-CODE builds a persistent, Git-versioned plain-text index of your codebase and database schema so coding agents read system context once instead of grepping the repo on every task.
Is aoci-code open source?
Yes — aoci-spec/aoci-code is an open-source project tracked on heatdrop.
What language is aoci-code written in?
aoci-spec/aoci-code is primarily written in Go.
How popular is aoci-code?
aoci-spec/aoci-code has 513 stars on GitHub.
Where can I find aoci-code?
aoci-spec/aoci-code is on GitHub at https://github.com/aoci-spec/aoci-code.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.