How far can
one person go with AI?
Amoeba Labs builds tools for people who ship alone. One person should be able to do what used to take a team.
Projects
Real software, built solo. Rust backends, real databases, paying users.
CMK
Claude Memory Kit. Persistent memory layer that intercepts conversations via MCP, classifies memories through five semantic gates, and enables recall across sessions. Local-first with optional cloud sync to Qdrant.
Onboard
Personalized codebase onboarding powered by Opus. Ingests full repository context and generates role-specific documentation. An SRE sees production risks. A new grad sees learning paths. Adapts in real-time.
Airlock
Preflight security analysis for pull requests. A GitHub App that scores risk, enforces policies, and blocks dangerous changes before they land. Fail-closed architecture with emergency bypass and full audit trail.
The best software will be built by tiny teams
AI changes what one person can build. The bottleneck used to be headcount. Now it is taste, speed, and willingness to ship.
Every project here was built by one person. Not demos, not prototypes. These are real products with real users. Amoeba Labs exists to see how far that can go.
Approach
Build real things
No toy demos. Every project has auth, billing, and deployment. Rust backends, real databases, paying users.
Stay small
More people means more coordination, not more output. Working solo forces you to pick better tools and simpler architecture. Zero meetings, zero overhead.
Ship constantly
When you own the whole stack, the feedback loop shrinks to minutes. No PRs to review, no deploys to schedule. Just push and see what happens.