RecallLoom Project Memory
Keep project context, progress, and next steps in files so AI sessions and tools can pick up where work left off.
- Source repo
- Frappucc1no/recall-loom
- Stars
- ★ 154
- Last updated
- 2mo ago
- License
- Apache-2.0
- Primary language
- Python
- FA score
- Insufficient evidence
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A software developer returning to a long-running project after changing sessions or models needs the current implementation state, decisions, and next steps.
- Not a fit if
- Teams that need task orchestration or multi-agent execution
- Users who only want chat transcript summaries
- Source review
- Insufficient evidence 8 safety controls not found
What does this agent do, and when should you use it?
RecallLoom is a file-based project memory skill for AI coding tools, storing project state in Markdown and JSON files beside or inside a project. Install it with npx skills add, then initialize a project from a tool that can install or read skills; use natural language or rl-* triggers to resume, record progress, and validate files. It requires Python 3.10 or later and uses files for the project brief, current summary, daily logs, configuration, and state. On resumption, it starts with current facts and next steps, consulting older material as needed; writes and repairs follow validation and confirmation paths. It needs no database, RAG layer, or background service, and keeps project facts in local workspace files.
After installing the recallloom skill, initialize memory for the current project in an AI coding tool; projects with a valid .recallloom/ or recallloom/ directory can resume directly. A resume reads context_brief.md, rolling_summary.md, daily logs, and relevant configuration and state so work can start from current facts and next steps. Users can ask it to record progress or decisions; controlled scripts write durable facts to project memory and check revisions, freshness, and structure around writes. Use rl-status to inspect memory status and rl-validate to check continuity files. An abnormal daily-log cursor can be previewed before repair is confirmed.
- A software developer returning to a long-running project after changing sessions or models needs the current implementation state, decisions, and next steps.
- A product team maintaining a PRD or RFC over time needs to preserve scope changes, their rationale, and open questions.
- A researcher continuing work across days needs to retain claims, evidence boundaries, and writing progress.
- A user switching among tools such as Codex, Claude Code, and Gemini needs one project memory to hand off between them.
- An individual or team wants project memory in local workspace files that can be reviewed and moved with the project.
How do you install or deploy this agent?
Prerequisites: Node/npm/npx, Python 3.10+, and an AI coding tool that can install or read skills. Install the skill:
npx skills add https://github.com/Frappucc1no/recall-loom --skill recallloomUpdate installed skills with:
npx skills updateHow do you use this agent?
For first use in a project, select the recallloom skill and initialize it:
@recallloom Initialize this projectIf the tool does not support @ skill mentions, choose recallloom from its skill selector or say, “Use RecallLoom to initialize this project.” To continue, say “Continue this project” or “Resume from where we left off.” The short triggers include:
rl-resume
rl-status
rl-validateTo record progress, say, “Record today’s key progress.”What are this agent's strengths and limitations?
- Project state lives in readable Markdown and JSON files that can be reviewed, diffed, reverted, and moved with the project.
- Resumption starts with the current summary, recent progress, and next steps instead of requiring a full repository or history read.
- Codex, Claude Code, Gemini, and other tools can hand off the same project memory when their relevant skill or entry support is available.
- Writes include revision, freshness, and structure checks; repair paths can preview changes before confirmation.
- It requires Node/npm/npx, Python 3.10+, and an AI coding tool that supports skills.
- Users need to record and maintain important project facts; RecallLoom does not automatically organize the entire repository.
- It does not provide task orchestration, issue tracking, RAG, a knowledge base, or a background execution service.
- Handoffs across AI tools depend on each tool supporting skills or a corresponding entry file.
How does this agent compare with similar options?
RecallLoom complements rule files such as AGENTS.md and CLAUDE.md: those describe how a tool should work, while RecallLoom records project progress and what comes next. Unlike built-in memory in tools such as Claude Code, its project facts live in files that can move with the project. It does not replace task graphs such as Beads, workflow execution frameworks such as LangGraph, or RAG and vector databases such as Chroma and Qdrant. For a one-time handoff, a manually maintained HANDOFF.md may be enough.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| RecallLoom Project Memory This agent | Insufficient evidence | Agent plugin / skillFree | ★ 154 | 2mo ago | Python | Codex · Claude Code |
| Agent Deck | 76 · Good | CLIFree + model costs | ★ 1.1k | today | Go | Codex · Claude Code |
| Coucou Agent Companion | 73 · Some gaps | Desktop appFree + model costs | ★ 5.5k | today | Swift | Codex · Claude Code |
| mex Project Memory | 71 · Some gaps | CLIFree | ★ 1.8k | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
- Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
- Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
- The supplied material consists mainly of a README excerpt, license, security policy, and documentation-sync workflow; the referenced skill, scripts, protocol, and product implementation are missing, preventing a product-level static assessment.
- Local storage, confirmation, write guards, and compatibility described in the README are documentation claims that cannot be verified against implementation materials.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.5.0 RecallLoom v0.5.0
Release notes excerpt
RecallLoom v0.5.0 RecallLoom v0.5.0 is a stable write-path architecture release. It consolidates the previously distributed write flow into explicit final owners while preserving the protocol 1.0 sidecar contract and existing workspace compatibility. Changed - Managed write, daily-log append, and post-append summary sync now use one transaction orchestration boundary with explicit preview/apply, cut-point, post-write verification, provenance, and final-evidence handling. - Input acquisition is consolidated behind a host-neutral transport contract for …View on GitHub ↗ -
v0.4.8.3 RecallLoom v0.4.8.3
Release notes excerpt
RecallLoom v0.4.8.3 RecallLoom v0.4.8.3 is a Windows write-recovery hotfix for the workspace write lock and for the failure guidance shown when a workspace becomes write-locked. Fixed - On Windows, a leftover .recallloom.write.lock file could make every later write fail. The process-liveness probe used to reclaim a stale lock relied on os.kill(pid, 0), which on Windows can terminate an unrelated live process or fail with an uncaught error that was reported as a damaged sidecar. Liveness is now probed through the Windows process API with a fail-safe …View on GitHub ↗ -
v0.4.8.2 RecallLoom v0.4.8.2
Release notes excerpt
RecallLoom v0.4.8.2 RecallLoom v0.4.8.2 is a narrow compatibility hotfix for ordinary pre-D5 reviewed baselines created by the supported recovery-review path. Fixed - Full provenance validation could incorrectly treat an ordinary review_imported_baseline as current receipt-backed evidence and report a receipt/state mismatch. - Validation now verifies the exact stored proposal and review material against the metadata produced by the supported promotion path, including its legacy terminal-newline normalization, before recognizing that baseline as reviewed …View on GitHub ↗
FAQ
Does it require paid models or services?
Will it automatically change application code or keep recording in the background?
Where is project memory stored?
.recallloom/ or recallloom/ directory beside or inside the project, using Markdown and JSON files. Privacy depends on how you sync, commit, host, or share those files.Can it manage tasks or multi-agent execution?
What if both supported memory directories already exist?
.recallloom/ and recallloom/ both exist, stop initialization and inspect the project state first.