VerifyThink

AI assistants for your business

Ready-made assistants for medicine, law, marketing, and more. Try free in the browser — subscribe when you need full access. No coding required.

Open marketplace VTChat Build your assistant (step by step)

Browse the marketplace · quick chat · step-by-step builder

Reason before you code — in Cursor

Structured plans, verification against the diff, and a learning loop over MCP or HTTP. One ReasonService for every project in your editor.

VT plans and verifies — you decide and write the code.

Connect hosted MCP Add to Cursor (local) Full MCP guide

reason_plan → code → reason_verify → feedback — same taskId per batch

What works today

Honest split: local developer setup vs public production. No promised launch date.

Local: works after start-local.ps1 Public production: on roadmap
What works today (guide) OCD demo: Reviews →

Two hosts — one login

Browser chat and marketplace live on the product host. Billing, API keys, and this landing page live on the account host. Cursor MCP is not the site chat.

Host Open for
Product · ReasonService VTChat, marketplace, assistants, PromptHub, help
Account · AccountServer Sign-in, pricing, cabinet, MCP onboarding

Fast chat in the browser ≠ reason_plan in Cursor. Developers: use MCP tools with the same taskId per batch.

One platform — four clear areas

Each product has its own page and settings inside it. Sign in for API key, billing, and subscriptions — not required to try VTChat or browse prompts.

VTChat

Browser chat with Plain and VT on one page.

Without account: No account: try chat (Free tier — queue and ads).

After sign-in: With account: your tier limits; Pro/Max without ads.

Settings: mode and model on the chat page.

Open

Assistants & marketplace

Domain playbooks — medicine, law, marketing.

Without account: No account: browse marketplace and try demos.

After sign-in: With account: subscriptions, saved builder, your assistants.

Settings: inside each assistant; subscriptions in cabinet.

Open marketplace Build your assistant (step by step)

PromptHub

Describe a task — get a ready prompt. Public catalog.

Without account: No account: generator and catalog.

After sign-in: With account: save drafts and publish (when enabled).

Settings: language and filters on PromptHub pages.

Open the generator Prompt catalog

MCP for developers MCP

reason_plan → code → reason_verify in Cursor.

Without account: No account: docs and one-click local install.

After sign-in: With account: API key for hosted MCP and HTTP /v1/reason.

Settings: API key, BYOK, model — in Account cabinet.

Setup Developers

Guest vs signed-in account

Explore most products without registering. An account ties billing, API access, and assistant subscriptions.

Area Without account After sign-in
VTChat Try in browser Tier limits; higher tiers — no ads or queue
Marketplace & assistants Browse catalog and demos Subscribe, manage assistants, builder progress
PromptHub Generate and copy prompts Save drafts; publish when enabled
MCP & HTTP API Local install + documentation API key, hosted MCP, usage quota
Account cabinet — Billing, tier, keys, assistant subscriptions

Simple pricing

The higher the tier, the higher the limits and the deeper the reasoning.

Free

0 ₽

With a wait in the queue and built-in ads · basic reasoning depth

Pro

≈ $20 / 1500 ₽

≈ $20 / 1500 ₽ per month · higher limits, deeper reasoning, no ads or waiting

Max

≈ $200 / 15000 ₽

Maximum limits and the deepest verification with a judge ensemble

More than a coding tool

The same reasoning core powers domain assistants, a marketplace, and privacy-first data handling.

Assistants for your task

Build an assistant with its own scenario set: it runs analysis by your rules and keeps records.

Marketplace

Publish and connect ready-made assistants — by subscription or one-off.

Privacy

We store the minimum: email, password hash, and request metadata. Export and deletion on request.

PromptHub: prompt generator and catalog

Describe the task in plain words — get a well-built prompt. Ready-made prompts are searchable by category and copyable.

Open the generator Prompt catalog

How it thinks before it acts

The product does not jump to an answer. It follows a short loop you can read.

First it turns your request into a plan: what done looks like, and what could go wrong.

Only then is the work written — to that plan, not to a guess.

Then the result is checked against the plan. Gaps go back to fix.

If something is unclear, it asks you. It does not invent your choice.

Limits are written in plain language: what works today and what is still in progress.

The same loop repeats: plan → work → check, so the next pass is not a lottery.

An assistant without structure vs. with our reasoning

Same request, different result. A comparison with no reference to any specific product.

What happens Assistant without structured reasoning With our structured reasoning
Understanding the task Takes the request literally and codes right away Clarifies the goal and expands it into a plan with a DoD
Edge cases Often forgotten — surface in production Listed up front: errors, empty data, limits
Verification None — “looks right” Diff is checked against the plan; gaps go back to fix
Tests Optional, if it remembers A test matrix is proposed with the plan
Reproducibility Different every time One loop: plan → code → verify
Improving over time Doesn’t learn from your tasks Builds experience and strengthens weak categories from feedback
Honest limits Hidden caps and surprise gaps Published LIMITATIONS — what works today and what does not

Read honest product limits (LIMITATIONS) → · See static before/after examples →

See the difference (static examples)

Pre-recorded pairs: flat model answer vs the same request with plan→verify. No live generation on this page.

All comparisons

Product quality check (hidden tasks)

We compare product versions on a hidden task set. Numbers appear here only after an honest run — not marketing guesses.

Figures will appear after we finish an honest measurement on a release tag.

Hidden product quality check — not GSM8K or a public model leaderboard.

For developers — MCP & API

Verifiable reasoning for AI agents in your editor: structured plans, verification, and learning loop over MCP or HTTP.

Connect hosted MCP Add to Cursor (local) Start free (API key) Full MCP guide

Do not use mcp-sampling in Cursor — use hosted MCP or local with your OpenAI API key in Cursor.

Connect MCP in 3 steps

  1. Register → copy API key on onboarding (for hosted MCP).
  2. Click “Add to editor” or run: npx -y verithink-mcp
  3. Open a project in Cursor — the agent calls reason_plan and reason_verify.
npx -y verithink-mcp

ReasonService on 127.0.0.1:8787 — one instance for all Cursor projects.

What you get

Structured reasoning

Multi-step plans with coverage checks, risks, and edge cases — before the first line of code.

MCP integration

Connect once — the agent calls planning and verification right in its coding loop.

Flexible model choice

Service mode with pay-as-you-go, your own key, or a local model — switch it in the cabinet.

How it works: plan → code → verify

One worked example of how a vague request becomes a verified result.

Request

Vague task

“Add file upload to the profile.”

reason_plan

Structured plan

The system expands the task into steps with a DoD: type and size limits, content checks, link privacy, edge cases, and tests.

Code

Implementation

The agent writes code to the plan — covering every point, not just the happy path.

reason_verify

Verification

The finished diff is checked against the plan. A missing size limit? Verification catches it and sends it back to fix.

Request: "add file upload"

Plan (DoD):
  1. Validation: allowed types, 10 MB limit
  2. Security: content check, private links
  3. Edge cases: empty file, duplicate name, dropped connection
  4. Tests: success, over-limit, invalid type

Verify: step 2 "size limit" not found in code -> fix
Result: plan closed 4/4, verification passed

Example output: a vague request turned into a checked plan.

The MCP reasoning cycle

Four steps of one tool, closed into a loop. Everything stays inside our system.

reason_plan

A plan with coverage, risks, and a DoD.

code

The agent implements the plan’s steps.

reason_verify

The result is checked against the plan.

Then feedback teaches the system, and the loop repeats until the task is 100% done.

Honest benchmarks

Metrics below are from our internal verification suite. Each row shows the latest verified or in-progress result.

Metric Result
Planning quality Plan→verify→feedback cycle operational
Knowledge Q&A accuracy Internal GPQA diagnostic: SYM-G1 FAIL (ops only)
Creative tasks quality VerifyThink +15 pp vs direct (73% vs 58% WI, n=3)

Methodology available on request. Small sample runs are labeled; large hidden checks are for internal diagnostics only (SYM-G1).

Reasoning tools in your editor

All of these are capabilities of one system of ours, available to the agent over MCP.

Planning

A step-by-step plan with a DoD before any code

reason_plan

Verification

The finished code is checked against the plan and goal

reason_verify

Test matrix

Edge cases and tests for the feature

reason_testplan

Learning

Each task’s outcome strengthens the system for next time

reason_feedback

Frequently asked

How is this different from a regular assistant?

We don’t code right away. First we build a verifiable plan with edge cases, then check the finished result against that plan.

Do I need a card to start?

No. The free tier is available right away — with a wait in the queue and built-in ads.

Where does the system run?

In your editor via MCP and over the HTTP API. The model can be service-provided, your own key, or local.

What do I pay for on paid tiers?

Higher limits and deeper reasoning with stricter verification. No ads and no waiting.

What about my data?

We keep the minimum needed to run the service. Prompts are processed by the model you choose. Export and deletion on request.

Do I need an account to use VTChat or prompts?

No — VTChat, PromptHub, and marketplace demos work without signing in. Account unlocks tier limits, API key for MCP, and assistant subscriptions.

Where do I change settings?

Inside each product: chat page, assistant UI, PromptHub. Billing, API keys, tier, and subscriptions — in Account cabinet (index.html).

Create your account

Email and password — we issue an API key on the next step.