0latency CLI

Verbatim CLI capture wrapper for Claude Code, Codex, Gemini CLI, and Aider. Captures all user inputs and agent outputs as role-tagged atoms, stored locally or synced to the 0Latency cloud memory platform.

Installation

pip install -e .

Usage

# Wrap Claude Code session (writes to cloud if authenticated)
0latency claude

# Wrap Claude Code with local-only storage
0latency --local claude

# Dry-run mode (show what would be captured)
0latency --explain claude

# Authenticate with 0Latency cloud
0latency login

# Check status
0latency status

# Pass arguments to Claude Code
0latency claude --print "what is 2+2"
0latency claude /path/to/project

Commands

• 0latency claude [args...] - Wrap Claude Code session with verbatim capture
• 0latency login - Authenticate with OAuth device-code flow
• 0latency status - Show auth state, storage info, and sync status

Flags

• --local - Force local-only storage (override cloud writes)
• --explain - Dry-run mode showing what would be captured
• --version - Show version information
• --help - Show help message

License

MIT

Reliability Features (v0.3.0)

Production-hardened for long-running sessions and adverse conditions:

• Crash Recovery: Zero atom loss via rolling buffer + auto-import
• Backpressure Handling: Local queue (10K cap) with exponential backoff retry
• Large Paste Support: UTF-8-safe chunking up to 1M+ characters
• Long-Session Stability: Ring buffers keep RSS < 500MB in 4+ hour sessions
• Interactive Prompts: Y/N and password prompts pass through (not captured)
• Tool-Call Tracking: Multi-tool chains atomized with sequence metadata
• Async Background: Non-blocking capture for long-running bash commands
• Atom Batching: API efficiency via 10-atom batches or 2s flush

See docs/reliability.md for details and verification.