GitMir guide
GitMir has read OpenAI Codex: what it does, what depends on what, and what a change could affect. Here is one question it answers — ask your own below.
See the product behind the code
GitMir turns OpenAI Codex into a living model of how the product works — what is connected, what depends on what, and what changes affect.
603,461 lines → 9.5 MB of Product Intelligence
←
OpenAI Codex
OPENAI CODEX · 603,461 LINES READ
You
Where is it riskiest to make a change?
GitMir guide
The riskiest place to change is “Terminal UI”: more of the product depends on App than on anything else.
242 inbound dependencies · computed from the reach graph, not estimated
The drawing
Terminal UI
Your existing stack + GitMir
Your tools already plan, find, build and review
GitMir gives them the same product understanding
Claude Code, Cursor, Sourcegraph, Augment and the rest of your engineering stack already help people and AI reach the right implementation, dependencies and history. GitMir adds the understanding of how the product works that would otherwise have to be rebuilt for every new task.
Your stack today
- Task
- Find the relevant implementation, dependencies and history
- Understand how this part of the product works
- Reason about the change
- Implement
For each new task, the relevant product understanding still has to be reconstructed from implementation context.
Add GitMir
- Task
- Start with existing Product Intelligence
- Reason about what exists and what needs to change
- Open the exact implementation needed
- Implement with the tools you already use
The product understanding is already there — for this task, the next task and the next person or agent.
Keep your tools. Give them the same understanding of the product.
GitMir turns that shared understanding into reusable Product Intelligence.
Claude Code · Cursor · Codex · Linear · Jira · Sourcegraph · Augment · CodeRabbit
See how GitMir works with your stackProduct history
The product has a history too
Git shows how the code changed. GitMir keeps those changes connected to the product. Across analyzed commits, you can see which product areas, behaviours, screens and customer paths each change reached. Implementation-only commits remain visible without being presented as product changes.
15 MOST RECENT COMMITS · 14 TOUCHED MODELLED BEHAVIOUR · 1 MOVED CODE ONLY · MODEL BUILT 2026-08-30
- Organize bundled Rust resources under asset directories (#41477)
Multi-agent · Agent core · MCP · Command execution
Screens · 3Slash command dispatch, Inline visualization viewer, StartupTooltips
Server · 7Agent spawn and resume, Agent role overrides, Model context fragments, World state sections, Guardian prompt builder, MCP tool approval message templates +1 more
Functions · 40agent_nickname_candidates, default_agent_nickname_list, restore_v2_agent_metadata, spawn_agent_with_metadata, spawn_agent_with_communication, spawn_agent_internal +34 more
Events · 2guardian.node_repl_policy, persistent_mode.instructions
1 file outside the current model — picked up on the next update.
- Use rules_rs platforms for release binaries (#41476)
Code moved; nothing the model describes changed.
- Refresh the TUI model picker from the app server (#41467)
Terminal UI
Screens · 11App event dispatch, AppServerSession, Model catalog fetch, Selection list popup, Bottom pane, Delayed approval prompts +5 more
Events · 16AppEvent::SelectAgentThread, AppEvent::RequestOlderScrollbackHistory, AppEvent::OlderThreadHistoryLoaded, AppEvent::DynamicToolCallCompleted, AppEvent::RunningTaskExit, AppEvent::PetSelectionLoaded +10 more
Each commit of OpenAI Codex is mapped onto the model through the files the model was read from. The cabinet goes further on your own repository: every push updates the model, and the history compares two builds to say what was added, changed, renamed or removed.
Measured on supabase
Compact enough to reuse
603,461lines of source covered
9.5 MBof Product Intelligence · 1.43 MB gzipped
19.4×smaller than the source it came from
up to 2,247×less text served to answer a measured question
MEASURED ON THE PUBLISHED SUPABASE SCOPE — APPS/STUDIO + PACKAGES/PG-META — AT THIS VERSION AND THIS QUESTION WORKLOAD.
The founder-attested record exists for the Supabase run; for this one the figures come from the run ledger and the model index — verified by tooling, not attested.
Why this repository
An agent, explained to agents
The most asked-about codebase of the year
Everyone extends, wraps or competes with coding agents. The docs and the source are both public; the model adds the answers that sit between them — which state machines govern a session, what the approval policy really permits, and which journeys a tool call walks.
A stress test for the method
Dense Rust internals with a different shape from Supabase Studio's product surface. The same method covered it completely, with zero integrity failures.
Record status
The measurements above come from the run ledger and the model index. The founder-attested measurement record for this run is in preparation — until it is signed, treat the figures as verified-by-tooling, not attested.
Your turn
See your own product this way
Connect one repository. GitMir builds a reusable understanding of how your product works, so your people, AI agents and existing tools can work from the same product picture.
Current product logicwhat exists now
Requirementswhat you want to change
Development taskswhat needs to be done
Product historywhat actually changed over time
Shared product understandingthe same product picture for your people and AI
One understanding of your product. Reused by every person, agent and task that needs it.
10,000 STARTING INTELLIGENCE CREDITS INCLUDED
What would make this more useful for you?
Tell us what you would want to ask, analyze, compare or understand next. We use this feedback to decide what capabilities to build next.
— received— reviewed— actioned
Something wrong in the model of GitMir Model — OpenAI Codex? ·