About

Why Argon exists

Argon began in 2025 with a problem in the MongoDB workflow for AI agents. An agent needed to branch its data, test a change, return to an earlier state, and try another approach. Putting those steps together took too much work.

Backups and rollback helped with recovery. Agents also needed a way to inspect what they had changed and bring useful results back after review. Argon brings branching, time travel, and reviewed merges to MongoDB, so agents can experiment in their own sandboxes and teams can decide what to keep.

Read more about the branch-per-agent workflow and time travel versus recovery.

Current status

Working today

  • Deterministic WAL engine; public reproducible benchmarks
  • Branches check out as real MongoDB databases — real-driver validated in CI
  • argon diff / merge — reviewable data PRs; per-actor undo
  • TTL sandboxes, dataset pins, MCP server, argon-agents

What's next

  • GCS chunk-store backend
  • Read-your-writes barrier for the wire proxy

Team

Jake Wang

Jake Wang

Founder

Software engineer at Pinterest. Previously at MongoDB, LinkedIn, and Bloomberg.

Noot Noot

Noot Noot

Chief Emotional Support Officer

Official team cat and professional morale booster. Keeps the workplace pawsitive. His value: beyond measure (and very fluffy).

Argon is licensed under the MIT License. Ready to take it for a spin or help shape its future?