Close the loop

Every agent today writes code and hopes. Mosaic gives agents what they've never had — the ability to verify their own work inside a running application.

Runs inside your app

Tests execute in your application's runtime — same process, same state, same stores. Not from outside. Not against mocks. Inside.

Direct store access

Zustand, Redux, MobX, custom state. Mosaic subscribes to your stores in real-time, traces every mutation, and asserts against actual state.

Zero mocks

Real backend. Real data. Real race conditions. If the test passes in Mosaic, it works in your app. No fake anything.

One command

mosaic test <name> — the agent runs it, results come back, the loop closes. Standalone binary, no runtime dependency.

Agent-first

Structured output, machine-readable logs, no interactive prompts. Every design decision optimizes for the agent, not the human.

Any agent

Claude Code, Codex, Gemini CLI, Orbit — if it can run a CLI command and read the output, it can use Mosaic.

The agent builds. Mosaic verifies.

The workflow is simple. User gives a prompt. Agent writes the code. Agent writes a test. Mosaic runs it inside the live app. Results come back — structured, machine-readable. The agent reads them and decides: pass, or try again.

That's the entire thesis. Every agent today operates in an open loop. Mosaic closes it.

1User prompt

2Agent builds

3mosaic test

4Pass?

YesDone

NoIterate → step 2

Test from inside, not outside

Playwright clicks a button and checks if the DOM changed. Mosaic calls the function behind the button, verifies the store updated, confirms the file system reflects the change, and checks the DOM — all from inside the application.

A Mosaic test script has direct access to application stores, internal functions, the file system, and the DOM. Same runtime. Same process. Same state.

Application storesZustand, Redux, MobX

Internal functionshandleSend(), handleRewind()

File systemread, write, verify

DOMwhen needed, not limited to it

Desktop today. Web and backend next.

Desktop application testing is proven internally — script injection into the application's runtime, direct store imports, file system verification. Battle-tested on Orbit with 18-step stress tests across 3 phases.

Web and browser application testing is planned with a Rust-built automation layer. Backend testing is confirmed scope. The injection strategy adapts to the platform.

ProvenDesktopScript injection into app runtime

PlannedWebRust-built browser automation

PlannedBackendApproach TBD

Mosaic is in development

Request early access