Your AI coding agents. One continuous workflow.
Move between coding agents without losing your project's context, decisions, or progress. Baton captures active git diffs and task state so the next agent resumes immediately without a cold repository re-read.
Every new agent starts with missing context.
When switching between coding tools, the new agent knows nothing about your previous session, active task branch, or why specific trade-offs were made.
- ✕Prompt repetition: You manually re-type 3 paragraphs of project context.
- ✕Massive token burn: Agent wastes 40,000–80,000 tokens scanning unneeded files.
- ✕Silent regressions: The new agent rewrites working code because decisions were lost.
- ✓Zero prompt typing: Baton passes the exact task slice, diffs, and tests automatically.
- ✓99.3% token reduction: 396 tokens instead of a 60,000-token repository dump.
- ✓Preserved decisions: Frozen files and settled architecture remain intact.
Three simple steps.
Designed to fit into your existing development workflow. No servers, no signups, and no cloud dependencies.
Capture your work
Baton automatically tracks the current task, modified files, decisions made, and test status from your local repository.
Choose your next agent
Select which coding agent receives the task next. Baton formats the prompt packet specifically for that tool.
Continue where you left off
The receiving agent gets concise instructions, inspects modified files, and continues the task without re-reading.
See Baton in action.
A real workflow passing an active authentication migration from Claude Code directly into OpenAI Codex CLI.
# Baton Handoff Packet: auth-session-migration source: claude-code | target: codex | status: ready ## Active Task - Migrate session storage from localStorage to secure HTTP-only cookies. ## Files Touched - lib/auth/session.ts (added cookie serialization) - app/api/auth/login/route.ts (switched to Set-Cookie header) ## Decisions & Invariants - Do NOT rewrite crypto algorithms in lib/auth/crypto.ts. - Next step: Update client fetch interceptor in lib/api-client.ts to drop manual Authorization header.
Everything you need to keep work moving.
Engineered for developers who use multiple coding agents and refuse to waste tokens re-explaining their repository.
Workflow orchestration
Coordinate complex engineering tasks across different specialized agents without dropping tasks, tests, or context.
Context continuity
Lock architectural invariants, modified files, and passing test results so the next agent builds forward instead of backwards.
Local-first architecture
Every handoff packet is written directly to your project in .baton/. No third-party servers, no account required, no token telemetry.
Agent flexibility
Use Claude Code for architectural design, Codex for rapid edits, and Cursor for in-editor debugging—seamlessly in one flow.
Explore the system behind Baton.
Detailed visual instruments showing how Baton works internally: context capture, token compression, multi-agent orchestration, and task lifecycle.
Context Capture
How Baton inspects the active session, modified git hunks, and decisions from Claude Code and bridges them smoothly to Cursor.
Handoff Packet
How Baton distills a full 60,000-token repository scan down to an exact 396-token markdown packet with 98.6% compression.
Agent Orchestration
How Baton routes work across different coding agents: Claude Code, OpenAI Codex, Cursor IDE, Gemini CLI, and Antigravity.
Workflow Lifecycle
How task jobs, touched files, architectural invariants, and next steps stay organized and physically inspectable across runs.
Works with your coding agents.
Baton supports all leading CLI and IDE agents. It generates standard markdown and JSON packets that any LLM or coding harness understands.
Keep your AI coding work continuous.
One continuous workflow across multiple coding agents. Zero token waste, zero cloud telemetry, and 100% local on your machine.