Local-first AI coding agent

The open coding agent
for your terminal.

Plan, Build, Verify, in your repo. Every step recorded, nothing merged without your approval.

npm install -g @flotic/minitok@1.5.2
  • Runs locally, inside your repository
  • Your model, Anthropic, OpenAI, or Google
  • Every run leaves evidence in .minitok/
Capabilities

Everything an agent should do.

Open source. Plan-and-Act. Every model. No vendor lock-in.

Edits across your project

Coordinated multi-file changes with linter-aware fixes, diffs, checkpoints, and one-click undo.

Runs bash commands

Executes in your terminal and reacts to output as it appears. Tests, deploys, all handled live.

Plan, then Act

Toggle Plan mode to align on a strategy, then Act to execute. Approve every step or flip auto-approve.

Your checks gate everything

Tests, lints, build, and verification gates you configure decide the outcome, not the model confidence.

Workflow

See minitok at work.

One agent across your IDE and terminal. Pick a task and watch it run.

1

Understand

Research the codebase structure, dependencies, and behavior before planning.

2

Plan

Generate a structured task plan with safety boundaries and scope limits.

3

Build

Implement changes with your configured LLM provider, file by file.

4

Verify

Run your checks, record evidence in .minitok/, and repair if needed.

Evidence

Don't take our word for it. Read the evidence.

minitok — autonomous run
minitok Verified result

Task

Add signup validation and regression coverage

  • Repository context loaded
  • Plan created from project structure
  • Implementation completed
  • Targeted retry after check feedback

Run summary

Verified

Changed
2 files
Checks
lint · tests · build
Cycles
2 autonomous cycles
Evidence
checkable result recorded
  1. Goal
  2. Inspect
  3. Plan
  4. Execute
  5. Verify
  6. Adapt
  7. Done

Illustrative sequence of one run. Every step maps to a locally recorded result.

$ minitok doctor
✓ Node.js ≥ 22.19.0
✓ Provider configured
✓ Checks validated
Ready to run.
Pricing

Simple, transparent pricing.

No subscriptions for the agent. Pay only for what you use.

LLM provider usage billed separately. Free 14-day trial available. See Cancellation Policy.

FAQ

Frequently asked questions

Is telemetry enabled by default?

No. Telemetry uploads are off by default and require your explicit consent.

Which models can I use?

Anthropic, OpenAI, and Google are first-class providers. OpenRouter, Azure OpenAI, Ollama, vLLM, and similar gateways can be configured.

Can anything change without my approval?

No. minitok plans, builds, and verifies, but nothing is merged without your explicit approval.

What exactly is verified?

Your own checks gate every result: tests, lints, build, and verification gates you configure decide the outcome.

Can I audit a run afterward?

Yes. Every run writes plans, diffs, check results, retries, and cost to .minitok/ on your machine.

Bring a task. Leave with proof.

Your repo, your model, your checks, and a record of everything in between.