Install CVC, meet Sofia (your AI agent), and give your AI a time machine. Two commands. That's it.
From zero to time machine
curl -fsSL https://jaimeena.com/cvc/install.sh | bashInstalls Python 3.11+ via pyenv if needed, then installs CVC. Works on macOS 12+ and Ubuntu 20.04+.
cvcThe setup wizard guides you through provider selection and API key configuration. Takes about 30 seconds.
User: Help me build an Express API with auth
Sofia: [designs and builds the API]
/commit "API with auth complete" ← Save the AI's brain
/branch "risky-refactor" ← Try something new
/restore abc1234 ← Go back if it fails
/log ← See full historyYour AI agent with a time machine
Creates step-by-step plans for complex tasks before writing code
Records corrections in lessons.md — never makes the same mistake twice
Runs tests and checks before declaring a task complete
7 sub-agents for Security, UI, Data, AI, and more
Configure your LLM provider
export ANTHROPIC_API_KEY=your-key-hereThree interfaces, one brain, one database
cvcFull AI coding assistant in your terminal. Sofia + 17 tools + 31 slash commands.
cvc launch claudeTransparent API proxy wrapping your existing tools. Claude Code, Aider, Codex.
cvc mcpModel Context Protocol server for VS Code + Copilot, Cursor, Windsurf.
Your quick reference for CVC
/commitSave a cognitive checkpoint/branchCreate an exploration branch/restoreRewind to any previous state/mergeCombine branch insights into main/logView full commit history/diffCompare two cognitive states/searchSemantic search across all sessions/recallFind what the AI decided before/compactIntelligently compress context/costShow token usage and $ spent/exportExport session as markdown/toolsList available agent tools/agentsList specialized sub-agents/providerSwitch LLM provider/modelSwitch model/helpQuick reference for all commands/doctorDiagnose system health/auditView compliance audit trail/statsSession and workspace statsCommon patterns for CVC power users
Never lose a stable state. Every risky idea gets its own isolated branch.
/commit "stable state"/branch risky-experiment...try risky approach.../restore <stable-commit> (if failed)/merge risky-experiment (if succeeded)Build a searchable knowledge base across all your AI sessions.
Work on feature → AI learns context/commit "feature complete"Next day: cvc → auto-restores session/search "auth pattern" → finds past solutionCut API costs by 90% with prompt caching and intelligent model switching.
/cost → check spending/compact → reduce context, save tokens/provider ollama → switch to free local model/model haiku → use cheaper model for simple tasksSame AI brain across CLI, VS Code, Cursor, Aider — zero context loss.
9 AM: cvc (CLI agent) → design architecture10 AM: cvc launch cursor → implement in IDE2 PM: cvc launch aider → refactor codeOne continuous conversation across all tools