You return a week later
The next session spends time reconstructing what you were doing instead of continuing it.
Repository memory for coding agents
Your coding agent should not start from scratch every time.
LoopCode keeps the goals, decisions, constraints, current work, and proof of what happened inside your repository—so Codex or Claude Code can pick up the real work with the right context.
- Why important decisions were made
- What is active, blocked, and next
- Which checks prove the work is ready
01 — The problem
Your repository stores the code. It rarely stores the whole story.
A coding agent can read thousands of files and still miss the reason behind a decision, the work another session already tried, the business goal a change protects, or the check that must pass before release.
Another agent takes over
It sees the current code, but not every decision, rejected option, risk, or unfinished thread behind it.
The repository grows
Product, release, reliability, and company work drift into separate lists with no dependable shared picture.
LoopCode gives the repository a durable working memory—not just another chat history.
02 — What changes
The next session starts with what the last one learned.
LoopCode adds a small, structured collection of Markdown files and receipts to the repository. They remain readable by people, versioned with the work, and available to supported coding-agent hosts.
“What was I working on?”
Context lives mainly in the previous conversation and has to be rediscovered.
“Here is the current work and why it matters.”
The repository records active work, governing goals, decisions, evidence, and the next verified action.
“The change looks done.”
A plausible answer can be mistaken for a verified result.
“These checks passed; this limitation remains.”
Material work closes with evidence, an honest residual, and a receipt that can be inspected later.
03 — What it keeps together
The parts of software work that are easiest to lose between conversations.
A
Direction
What the product and repository are trying to achieve—and the constraints that cannot be traded away.
B
Continuity
What is happening now, what was already tried, what is blocked, and what should happen next.
C
Decisions
The options considered, why one path was chosen, and when that choice should be reviewed again.
D
Proof
The tests, checks, observations, limitations, and recovery information behind a completion claim.
E
Coordination
Who or what is working where, so parallel agents do not quietly overwrite or duplicate each other.
F
Learning
What this work taught the repository and where that lesson may—or may not—apply next.
04 — Designed to stay understandable
You can inspect it, change it, pause it, or remove its prompt behavior.
LoopCode is not a new model and it does not replace Codex or Claude Code. Its current core uses local Python and Markdown. The host provides intelligence and tools; the repository keeps its own operating memory.
- Local and readable
- Canonical memory is ordinary Markdown in the repository, not a hidden remote database.
- Bounded authority
- Deployment, spending, credentials, destructive actions, and production changes still require their own authority.
- Evidence before “done”
- Required checks and recorded observations matter more than an agent sounding confident.
- Host-neutral
- The same repository memory can support Codex, Claude Code, Cursor, and future compatible hosts without making one conversation the source of truth.
05 — Availability
LoopCode is in private beta.
The plugin is working across Codex and Claude Code, but its source marketplace is currently private. Public installation will open only when the distribution repository, license, release channel, and onboarding path are ready for people outside the development team.
Current status
Verified locally. Public distribution is being prepared.
This page will publish the real installation command when it works for everyone—not before.
Development repository Access currently requires permission to the private repository.Questions
The short version.
Is LoopCode another coding agent?
No. Codex, Claude Code, or another compatible host remains the coding agent. LoopCode gives that host durable repository context, operating rules, verification gates, and receipts.
Does it upload my repository?
The current core has no network dependency and canonical memory stays in local Markdown. Any future remote analytics must remain separately opt-in and cannot contain prompts, code, paths, diffs, secrets, or repository identifiers.
Does it make every task complicated?
No. Small reversible work stays lightweight. Deeper review is reserved for consequential, cross-system, uncertain, repeated-failure, or hard-to-reverse work.
Can I stop using it?
Yes. LoopCode can be paused or deactivated while preserving the repository's Markdown memory and receipts.