It remembers every error, warns before repeats, and blocks coding without a spec.
Claude Code is fast. But it forgets what it broke yesterday and breaks it again today.
Clouvel remembers.
Free and open source. 20 tools included.
pip install clouvel · works everywhere MCP is supported
Claude Code is fast. But without memory, speed creates chaos.
No error memory between sessions. Same mistake, different day.
No second pair of eyes. Blind spots go straight to production.
Decisions lost when context resets. Waste hours rediscovering context.
AI starts from zero every time. No learning, no improvement.
Three things. All automatic.
AI that learns from mistakes.
Warning: This error happened before.
Root cause: Missing null check on DB query
Prevention: Always validate query results
(Memory #7 - prevented 3 times)
Think before AI codes.
You: "Build login"
BLOCKED - No PRD found.
You: *writes PRD*
PASS - Ready to code.
Blind spots in 10 seconds.
PM: "What happens when login fails 5 times?"
CTO: "Rate limiting needed for brute force."
Real output from Clouvel
claude> Build a login page with OAuth support
🚫 BLOCK | Missing: PRD, Acceptance
Fix: start(path=".")
Found: None | Missing (required): PRD.md, Acceptance Criteria
Next: write PRD with Acceptance Criteria section first
Why spec first? 10 min spec = 2 hours saved (no rework)
claude> ▊
claude> Build a login page with OAuth support
✅ PASS | Required: PRD ✓ Acceptance ✓ | 12 tests
Ready to code
✓ docs/PRD.md (OAuth, password reset, rate limiting defined)
✓ Acceptance Criteria: 8 items
✓ Tests: 12 test files
Claude
I'll implement the login page based on your PRD specifications:
Creating src/auth/login.ts...
claude> ▊
# Error Learning Analysis
Pattern Analysis:
import_error 3x Learning needed
null_check 2x Observing
CLAUDE.md Updated:
+ NEVER: import from deleted module
+ ALWAYS: check query result before use
Learned: 2 rules · Analyzed: 5 errors
# C-Level Meeting: Auth System
PM: "What happens when the token expires mid-session?"
CTO: "Are we hashing passwords with bcrypt or argon2?"
CSO: "Rate limiting on login? Brute force is day-one risk."
QA: "Edge case: user logs in on two devices simultaneously?"
4 managers · 8 blind spots found
SHIP READY
[████████████░] 3/4 PASS
PASS LINT 2.5s
PASS TYPECHECK 1.8s
PASS TEST 4.2s
SKIP BUILD
Evidence: .claude/evidence/auth_0221.md
Everything included. No tiers, no limits.
| Feature | Included |
|---|---|
| Error history | Full history + patterns |
| Context slots | 50 + timeline |
| Manager feedback | 8 managers, 2+ questions |
| Error learning | Auto-generates rules |
| Cross-project memory | Share lessons everywhere |
| Drift detection | Catches scope creep |
| Ship pipeline | Full verify + evidence |
20 tools. All included.
can_codeBlocks coding without a specstartSet up project with PRD templatessave_prdSave PRD from conversationerror_checkWarns before repeating mistakeserror_recordRecords errors with root causecontext_saveSaves state before context runs outcontext_loadRestores state in new sessionquick_perspectivesQuick blind-spot checkgateRun lint, test, build in sequencelicense_statusCheck installation statuserror_learnAuto-generates NEVER/ALWAYS rulesmemory_statusError memory dashboardmemory_searchSearch past errors by keywordmemory_global_searchCross-project error patternsdrift_checkDetects scope drift from goalsplanDetailed execution plansmeetingFull 8-manager C-Level reviewshipOne-click lint+test+build+evidencerecord_decisionPersistent knowledge basesearch_knowledgeSearch past decisions3 commands. Under 60 seconds.
# 1. Install
pip install clouvel
# 2. Auto-configure for your client
clouvel install
# If 'clouvel' not recognized:
# py -m clouvel install (Windows)
# python3 -m clouvel install (Mac/Linux)
# 3. Start coding
claude
# Clouvel runs automatically
That's it. Clouvel runs automatically when you use Claude Code.
Day 1: Install → start → write PRD → can_code PASS → code
Day 3: Error happens → error_record saves it
Day 5: Same file → error_check warns "this broke before"
Day 7: 5+ errors → error_learn auto-generates rules
Day 10: Context runs out → context_save/load preserves everything
Day 14: Cross-project memory shares lessons across repos
Join early adopters using Clouvel in production
Using Clouvel in your project? We'd love to hear about it.
Share your experienceClouvel is an MCP server that runs alongside Claude Code. It records every error with root cause analysis, warns you before repeating past mistakes, and blocks coding until you have a spec document. Think of it as persistent memory for your AI coding assistant.
CLAUDE.md is static rules you write by hand. Clouvel learns automatically from your actual errors and decisions. It detects patterns across sessions and warns you before you repeat a mistake. You can't write a CLAUDE.md rule for a bug you haven't encountered yet.
Yes. Everything works fully offline. All data stays local on your machine.
No. Clouvel is an MCP server - it provides tools to Claude Code but never modifies your source code. It stores its data in a .claude/ folder. Your project structure stays untouched.
All 20 tools are included: spec gating, full error history with pattern learning, context save/load, cross-project memory, 8-manager C-Level review, one-click ship with evidence, and more.
Nothing. Clouvel runs entirely offline. Your code, documents, and error history never leave your machine.
Just install Clouvel on your new machine with pip install clouvel. Your project-level data lives in .claude/ inside each repo, so it travels with your code.
Machine reset, license issues - we'll help you quickly
Inquiry received! We'll get back to you soon.
Failed to send. Try again later or use GitHub Issues.
Or reach us at GitHub Issues
I built Clouvel because I was tired of Claude Code breaking the same thing twice. Every error I fixed yesterday came back tomorrow. So I built a memory layer that actually learns.
Clouvel is actively maintained and updated weekly. Questions, bugs, or ideas — I read every GitHub Issue.
If Clouvel saves you time, help others find it
A GitHub star helps independent dev tools get discovered
Star on GitHub