Feature request: Persistent memory for coding agent sessions
Nobody has claimed this yet.
- Dominant language
- Go
- Stars
- 5.1k
- Forks
- 475
- Avg merge
- 1d 11h
- Merged PRs (30d)
- 178
Description
Problem or use case
Summary
Add support for persistent, self-managed memory across coding agent sessions — similar to what OneContext provides — so that agents can record their trajectory, share context with teammates, and resume work from exactly where they left off.
Add a memory layer to coding agent sessions that saves important context to a memory.md file so that critical information is never lost between sessions.
Problem
When working on long or complex tasks, important decisions, reasoning, and progress get lost when a session ends or the context window fills up. Starting fresh every time means the agent has no recollection of what was already done, decided, or explored.
Proposed Solution
Introduce a simple persistent memory file (memory.md) that the agent reads at session start and writes to during the session. When the context window approaches its limit (e.g. 128K tokens), the agent automatically summarizes the current session, saves the key information to memory.md, clears the context, and starts fresh — loading the memory back in so nothing important is lost.
Memory Files
∙ main.md — shared memory (global, always loaded)
∙ branches/{branch-name}/commit.md — milestone summaries per branch
∙ branches/{branch-name}/log.md — session event log
∙ branches/{branch-name}/metadata.yaml — structured project info
Desired behavior
Proposed solution
Implementation Approach:
Since Entire already uses Claude Code hooks to capture agent sessions on every git push, it could extend this existing hook infrastructure — specifically the SessionStart and Stop hooks — to read and write a structured memory layout that hydrates the agent at session start and flushes updates at session end, making persistent cross-session and cross-teammate memory a natural extension of the checkpoint system already in place. This memory would be stored in the following files:
∙ main.md — shared memory (global, always loaded)
∙ branches/{branch-name}/commit.md — milestone summaries per branch
∙ branches/{branch-name}/log.md — session event log
∙ branches/{branch-name}/metadata.yaml — structured project info
CopyCat 🐱: https://github.com/TheAgentContextLab/OneContext
Alternatives or workarounds
No response
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reading the existing SessionStart and Stop hooks and the checkpoint flow described in the issue. Trace how session data is captured, then define how main.md and the branches/{branch-name} memory files would be loaded and written; done means context survives session boundaries and can be resumed across branches and teammates.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- git, go, markdown, yaml
- Domain
- cli, developer-experience, tooling
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100