Skip to main content

Maintain your Momentum as you build.

The Momentum workspace in light theme: a City Builder session where Claude Code resumes the goal from Momentum Memory, the commit graph docked beside the transcript

Start to done, across pauses and runtimes.1

Bind the goal once.

“Build a city builder in three.js” becomes a durable goal: title, criteria, constraints, bound to the session, not the runtime. Codex runs it at GPT-6.1 Sol, Ultra effort, Approve for me.

  1. Bind the goal once.

    “Build a city builder in three.js” becomes a durable goal: title, criteria, constraints, bound to the session, not the runtime. Codex runs it at GPT-6.1 Sol, Ultra effort, Approve for me.

  2. Work lands as files.

    Sol scaffolds the street graph and zoned blocks. Every file it touches shows up in Review as it lands: staged, edited, untracked.

  3. The graph keeps the receipts.

    Real commits on real branches: the work is a tree you can inspect, not a transcript you have to trust.

  4. Switch the harness. Say continue.

    New runtime: Claude Code, Opus 5.5, Auto mode, Ultracode effort. Type “continue” and nothing else; the goal comes back from Momentum Memory.

  5. Every turn lands on the Ledger.

    Receipts record which harness ran, how the turn ended, and what it changed, across the handoff too.

  6. Goal met. The artifact takes over.

    All four criteria pass, the goal flips to met, and the preview fills the whole frame.

No vendor lock-in.

The goal binds to the session itself: swap the harness or the model mid-flight.

Connect it with your agents via MCP&ACP.2

Your work stays on your machine.

Goals, memory, and checkpoints live in a local store you control. We don't process or retain any of it on our end.

Harnesses like Claude Code or Codex talk to their own model providers. That traffic is between you and the provider under their terms.

Use Cases

Multi-day builds

  • Goals persist across restarts
  • Criteria define what done means
  • Pick up where the agent left off
See it in the story

Harness switching

  • Change runtime or model mid-session
  • Memory follows the session
  • No re-briefing the new agent
See it in the story

Work you can audit

  • Every step lands on the record
  • Conflicts stay open until resolved
  • Handoffs carry their evidence
See it in the story

Build with Momentum

Talk to us

Questions

What is Momentum?

Local archive-and-resume infrastructure for agents: versioned goals, typed memory, and atomic SQLite checkpoints, so work continues after disconnects, restarts, or a full context window.

Which agent harnesses does it attach to?

Momentum scans the machine for installed harnesses (Codex, Claude Code, Grok Build, and other ACP-compatible runtimes) and connects locally, binding continuity to the session itself.

Where is my data stored?

Goals, memory, and checkpoints live in a local store you control. We do not process or retain any of it on our end.

Is there a hosted version?

Momentum Cloud (a self-hosted enterprise console) and Momentum Enterprise (priority support and fleet licensing) are coming soon and not generally available.

Join the Momentum waitlist.

What would you like?

Loading security verification…

Notes

  1. Emulated Momentum session shown for illustration: it depicts Codex on GPT-6.1 Sol (Ultra effort, Approve for me) handing off to Claude Code on Opus 5.5 (Ultracode, Auto). Model and mode labels are illustrative; neither depicted runtime produced the artifact render. The interface may change between versions.↩
  2. Claude and Claude Code are trademarks of Anthropic, PBC. Codex and GPT-6 Astra are trademarks of OpenAI Group PBC. Grok, Grok CLI, and Grok Build are trademarks of X.AI Corp. Gemini and Antigravity are trademarks of Google LLC. Cursor is a trademark of Anysphere, Inc. OpenCode, Pi, Cline, Windsurf, Trae, and other names and marks belong to their respective owners. TACTICO is independent and is not affiliated with, sponsored by, supported by, or endorsed by any of these companies.↩