VerifiedLocalInspectable

A repository-aware workflow for changes you can review.

Repository-Aware Coding Workflow

Describe a task.
Inspect the change.

minitok researches your repository, plans and applies a change, runs your configured checks, and records the result. Failed checks can guide bounded repair; if limits are reached, it stops for review.

Choose a plan
npm install -g @flotic/minitok@1.4.6
Read the docs
Local-firstYour repository stays yours.
Bring your own LLMYour keys, your model.
Provider-agnosticFirst-class vendors plus custom OpenAI-compatible endpoints.

Why minitok?

AI coding agents generate code. minitok adds a repository-aware workflow around them: plan, implement, verify, review, repair, and record the result.

Without minitok

  • Prompt every change manually
  • Review and correct output repeatedly
  • No automated verification

With minitok

  • Describe the task and inspect the plan
  • Run checks and bounded repair cycles
  • Keep local evidence of the outcome

minitok is not another AI model. It is a workflow runtime that structures the task, controls execution, verifies results, and handles retry and recovery within configured limits — around whatever model you choose.

How It Works

A closed loop for every repository change, with explicit verification and safe-stop boundaries.

1

Describe

Tell minitok what you want changed in plain language.

Goal
2

Execute

minitok directs your coding agent.

Agent
3

Verify

Your checks validate the result.

Checks
4

Adapt

Failures guide targeted retries.

Retry
5

Deliver

An approved, verified result is applied to your working directory.

Result
✓ Verified result

Failsafe: if checks fail or configured limits are reached, minitok stops without merging the change. The normal approval applies one complete change set; file-by-file cherry-picking is not currently supported. If the final patch cannot be applied cleanly, it is preserved at .minitok/last-run.patch for inspection.

Inspect the result

See the workflow in action

A representative run, local measurements, and the commands that connect a goal to a verified change.

Representative run

From goal to verified change

Repository context, bounded execution, checks, and a result you can inspect.

Illustrative fixture, not live production evidence. Verified means the configured checks passed; it does not guarantee correctness or production readiness.
Measured product outcomes

Clearer workflows, measured locally

Context compaction and bounded local run history, shown as implementation fixtures rather than performance promises.

Measured character compaction and bounded local outcome persistence.
Context control

Keep what matters

Compacts over-budget context while reporting the before-and-after size.

100,000 → 4,000 characters Fixture result, not a provider-token benchmark.
Inspectable outcomes

Carry verified results forward

Stores bounded local outcome history for targeted retries.

Bounded outcomes · evidence recorded locally Fixture persistence shown; model learning and performance are not claimed.

From Install to Verified Result

Install, initialize the repository, configure a provider and verification gate, then inspect the recorded result.

bash
# Check environment and configure your LLM provider
$ minitok doctor
✓ Node.js >=22.19.0
✓ LLM provider configured
✓ Repository detected
 
# Initialize repository config and verification gate
$ minitok migrate
# Configure minitok.yml and keep VERIFY_CMD.mjs in the repo
 
# Run the repository-aware workflow
$ minitok run "Add input validation to the signup form"

Requires Node.js ≥ 22.19.0 and git. Configure at least one provider in minitok.yml; non-dry-run tasks also require a passing VERIFY_CMD.mjs gate and active entitlement.

Built for Developers Who Ship

Core tools for real developer workflows.

Provider Agnostic

Use Anthropic, OpenAI, Google, or a custom OpenAI-compatible endpoint without changing the workflow.

Repository Intelligence

Research and planning use context from your actual codebase before implementation.

Local Workflow Runtime

Run the workflow in your repository with explicit limits.

Budget Control

Set token limits and control usage.

Review & Repair

Verification and review failures guide bounded retries and targeted repair.

VS Code & MCP

Use the VS Code sidebar or connect the local MCP stdio runtime to an external agent.

What Would You Build?

Describe a task in plain language; minitok runs the workflow.

Fix a bug

"Prevent duplicate signup accounts and add a clear error."

Ship a feature

"Add signup validation and tests; verify before completion."

Maintain a dependency

"Upgrade auth and fix the tests that break."

Security by Design

Protect credentials and control where code runs.

Signed entitlements

Only authorized installations can access protected minitok capabilities.

Encrypted credentials

Activation keys are encrypted at rest so sensitive credentials stay protected.

Protected sessions

Session boundaries help keep authenticated workflow access isolated and controlled.

Abuse-resistant service

Rate limits help protect public endpoints and keep access reliable.

Simple Pricing

Choose the telemetry boundary for your workflow.

Open
$3.99/ month

Local-first workflow runtime

  • Consent-required per-run telemetry
  • 30-day retention
  • Repository-aware planning
  • Verify, review, and repair loop
  • Bring your own LLM provider
Choose Open
RecommendedSelect
$4.99/ month

Local-first workflow runtime

  • Consent-required aggregate-only telemetry
  • 14-day retention
  • Repository-aware planning
  • Verify, review, and repair loop
  • Bring your own LLM provider
Choose Select
Private
$6.99/ month

Local-first workflow runtime

  • No telemetry upload or storage
  • Telemetry stays local
  • Repository-aware planning
  • Verify, review, and repair loop
  • Bring your own LLM provider
Choose Private

All prices are in USD. LLM provider usage is billed separately. Taxes, if applicable, are shown before payment. Monthly subscriptions renew automatically until cancelled. Cancel anytime through the Dodo customer portal; access continues through the current paid billing period, and cancellation stops the next renewal. The current paid billing period is not automatically refunded. No free trial. See the Cancellation Policy.

Frequently Asked Questions

What is minitok?

minitok is a local-first workflow runtime around the coding model you choose. It separates repository research, planning, implementation, deterministic verification, review, repair, and evidence recording. It does not replace the model or guarantee a correct production change.

Do I have to pay for an LLM separately?

Yes. minitok is the workflow runtime, not an LLM provider. Bring credentials for Anthropic, OpenAI, Google, or a custom OpenAI-compatible endpoint and pay that provider directly for model usage.

Which providers are supported?

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

Where does minitok run?

The CLI runs the workflow locally and edits your working directory. Model requests go to the provider configured in minitok.yml. The VS Code Extension uses the local CLI, and MCP access uses explicit scopes and approvals.

What do the plans include?

Open includes consented per-run telemetry for 30 days; Select, consented aggregate-only telemetry for 14 days; Private stores no telemetry. Each includes one installation and no free trial. Cancel anytime through the Dodo customer portal; access continues through the current paid billing period and the next renewal is canceled. The current paid billing period is not automatically refunded. See the Cancellation Policy.

Ready to Automate Your Workflow?

Run a repository task with checks, limits, and a result you can inspect.

Choose a plan