👍 👎 Local-first · open-source · enterprise governance when agents touch real systems

Catch AI agents before they spend money, leak secrets, or break production.

Claude Code, Cursor, Codex, Gemini CLI, Amp, Cline, OpenCode, and MCP tools can touch terminals, repos, databases, SaaS APIs, and cloud systems. After a configured agent proposes a tool call, ThumbGate evaluates it before execution and records the decision. The default policy denies detected secret exfiltration plus commands that kill the ThumbGate gate process or enable its bypass environment variable; other matched high-risk classes such as destructive deletes, force-pushes, and fetch-and-run commands warn by default. Strict mode turns matching warning-mode checks into denies. The beachhead buyer is an enterprise engineering, security, or platform team that needs governed AI-agent execution before agents touch regulated or high-blast-radius workflows.

MCP tool risk Managed-agent rollout Token spend pressure Regulated agent governance

ThumbGate flagging an AI agent's rm -rf, git push --force, and chmod 777 in real time — flagging them by default and hard-blocking under strict mode — while letting safe commands through

Pay $__SPRINT_DIAGNOSTIC_PRICE_DOLLARS__ diagnostic Pay $__WORKFLOW_SPRINT_PRICE_DOLLARS__ sprint Start Pro — $19/mo Install Free CLI

Money path today: Workflow Hardening Diagnostic ($__SPRINT_DIAGNOSTIC_PRICE_DOLLARS__) or Sprint ($__WORKFLOW_SPRINT_PRICE_DOLLARS__) for one repeated agent failure. Pro stays the solo self-serve lane. Prefer scoping first? Send workflow first.

Team / enterprise: buy proof this week

One workflow. One repeated failure. Paid diagnostic or sprint with Stripe — credit applies when you continue to implementation.

Pay $__SPRINT_DIAGNOSTIC_PRICE_DOLLARS__ diagnostic → Pay $__WORKFLOW_SPRINT_PRICE_DOLLARS__ sprint →

Solo operator: Start Pro

Unlimited rules, recall, proof, exports after one real caught repeat.

Pay $19/mo with Stripe →

Still evaluating: Free CLI or intake

2 captures/day, 3 active rules. Or send the workflow if you need fit review before paying.

Send workflow first →

Ideal customer Enterprise engineering, security, and platform teams already using AI coding agents in repos, CI/CD, SaaS APIs, cloud operations, or audited workflows.

Pain Security is the pain: agents can repeat dangerous actions. Enterprise governance is the budget: policy, auditability, approvals, and shared prevention before execution.

Partnership fit Detection and coordination layers can identify drift. ThumbGate turns approved findings into enforceable PreToolUse rules with decision evidence.

MIT open source Local-first Works with MCP-compatible agents Verification evidence in GitHub

“A better dashboard doesn’t make the agents more reliable. The hard part isn’t visibility. It’s trust.”

— Rob May, CEO & co-founder, Neurometric AI, in The New Stack on Anthropic’s Claude Code Agent View (May 2026). This is Rob May describing the industry, not an endorsement of ThumbGate. ThumbGate addresses that trust problem with configured PreToolUse gates, accepted feedback, explicit prevention rules, and local decision records.

Local-first is the new default

Perplexity demoed a Personal Computer agent at Computex 2026 that asks the user before sending sensitive content to the cloud. Automation Anywhere shipped EnterpriseClaw with Cisco, NVIDIA, Okta, and OpenAI — central policy, locally enforced, audit reported back. The architectural debate is settled. ThumbGate runs the same pattern for the coding agent already on your laptop.

Sources: Perplexity at Intel Computex keynote (2026-06-02), Automation Anywhere EnterpriseClaw launch (Imagine 2026), The New Stack coverage.

First-Dollar Activation Path

Run your first pre-action check locally.

Install ThumbGate, capture concrete feedback, and let the configured PreToolUse hook flag a matching action before execution. Upgrade only when you need higher limits, recall, a personal dashboard, or exports.

1. Install ThumbGate

Run npx thumbgate init in your repo. Claude, Cursor, Codex, Gemini, Amp, Cline, and OpenCode are supported through local hook or MCP-compatible setup paths.

2. Give feedback

Use thumbs up for a good pattern and thumbs down with concrete context for a failure. Accepted feedback becomes a local lesson; recurring failures can be promoted into prevention rules.

3. The repeat gets flagged

A matching tool call is evaluated and recorded before execution, with a reason the agent can use to choose a safer plan. Detected secret exfiltration and commands that kill the gate process or enable its bypass variable are denied by default; strict mode also denies matching warning-mode checks.

👍👎

Accepted feedback becomes local lessons Evaluate proposed tool calls before execution

For consultancies, platform teams, and AI product teams that want workflow governance, CLI-first rollout, and a reliable operator.

Have one AI-agent failure that keeps repeating? Start with one real workflow, one repeated failure pattern, enforceable pre-action gates, and a short audit trail your team can keep.

Prove one evaluated repeat before asking anyone to buy. Give thumbs up when the agent follows your standards and a concrete thumbs down when it misses. Upgrade when the individual operator needs higher limits, recall, a personal dashboard, or exports.

Workflow governance for isolated execution: ThumbGate pairs policy checks with Docker Sandboxes, signed hosted sandbox dispatch, Changeset evidence, and exact main-branch merge commit verification before release claims ship.

June 2026 Market Signal

The market moved from “agents can code” to “who is allowed to let agents act?”

The high-ROI buyer story is not another chatbot memory claim. June 2026 coverage is pushing the same pattern from multiple directions: tool lockdown, MCP security, managed agents, context budgets, and agent-authored production code all need pre-action governance at the execution boundary.

MCP and tool lockdown

As MCP servers and hosted tools spread, the failure mode is no longer just a bad answer. It is an agent calling the wrong tool with real developer permissions. ThumbGate provides a local allow, warn, or deny decision before a configured tool call executes.

Managed agents need receipts

Managed background agents can multiply output, but every extra autonomous run needs evidence, rollback boundaries, and approval rules. ThumbGate records why a configured pre-action check allowed, warned on, or denied a proposed action.

Tokenmaxxing backlash

Long-context agents are expensive when they repeatedly reload irrelevant history. ThumbGate keeps paid value above the free wedge with recall, search, context gating, and proof exports instead of dumping everything into every prompt.

Production code by AI

When teams claim most new production code is AI-authored, the real enterprise question becomes auditability. ThumbGate can promote accepted recurring corrections into prevention rules and retain the local gate evidence available to reviewers.

Data-science posture: treat Pro checkout starts and workflow-intake submissions as intent, not revenue. Make conversion or traction claims only from authenticated billing and analytics telemetry; this page does not publish a current customer or revenue count.

Deterministic Prevention

Native thumbs are a black box. ThumbGate is the inspectable control layer.

Codex, Claude Code, ChatGPT, and other agent surfaces can collect preference signals, but you usually cannot see exactly what changed, which rule will fire, or why a future tool call is allowed. ThumbGate keeps the prevention layer outside the model: accepted typed feedback becomes a local lesson, recurring failures can become explicit rules, and gate results expose the matched check and decision reason available at evaluation time.

Black-box memory

Native thumbs and vendor memories may improve future behavior, but they do not provide ThumbGate's explicit local allow, warn, or deny contract at the moment a configured agent touches files, terminals, APIs, or CI.

Inspectable ThumbGate memory

Lessons live in your ThumbGate store, can be searched, exported as JSONL or DPO pairs, and traced back to the exact correction that created the rule.

Rules before execution

The final decision is not another model opinion. ThumbGate checks tool name, arguments, working directory, command shape, confidence, and required evidence before the action runs.

Why this matters now

Agent security is now mainstream risk. Coding agents run shell commands, write files, query databases, and chain actions with developer permissions, so unattended autonomy needs a local policy boundary.
MCP adoption is accelerating. More tools are becoming agent-callable through shared protocols, which means one cross-agent governance layer beats one-off prompt rules per app.
Managed-agent output needs audit evidence. Background agents, subagents, and workflow runners can ship more work, but buyers need receipts for what was blocked, approved, and verified.
Repeated failures waste cash and trust. Repeats can burn tokens, review time, and release confidence. ThumbGate can promote accepted recurring feedback into a reusable prevention check.

Sources to verify the market timing: Docker on AI coding agent security risks, TechRadar on enterprise agent security pressure, and current MCP adoption coverage.

Claude Code · Claude Desktop · Claude Extension

Configure ThumbGate for Claude Code or Claude Desktop.

ThumbGate ships a published Claude Desktop extension bundle (.mcpb) you can install today. Claude Code users can also add the repo marketplace plugin immediately. No waiting for directory approval.

1. Install for Claude Code

Run npx thumbgate init --agent claude-code or add via claude mcp add thumbgate -- npx --yes --package thumbgate thumbgate serve

2. Or install the Claude Extension

Download the .mcpb bundle for Claude Desktop, or use the repo marketplace: /plugin marketplace add IgorGanapolsky/ThumbGate

3. Give feedback, checks auto-generate

Type a concrete thumbs down when Claude makes a mistake. Accepted feedback can distill a lesson from up to 8 prior entries, and recurring failures can become rules evaluated by the configured PreToolUse hook. Detected secret exfiltration is denied by default; strict mode also denies matching warning-mode checks.

Download Claude Extension (.mcpb) Claude Desktop setup guide Claude plugin docs

Claude Code Skill: Type /thumbgate in any Claude Code session. Auto-triggers on “check”, “feedback”, “block mistake”. Free skill on top of the same local gateway.

ChatGPT Entry Point · Live ThumbGate GPT for ChatGPT

Use the GPT as a preflight desk for risky commands, refunds, deploys, and PR actions.

The live GPT is a ChatGPT entrypoint for preflighting a proposed action and finding setup guidance. It does not create a local execution boundary by itself. Local allow, warn, or deny decisions require npx thumbgate init plus the documented integration for the agent that executes the tool call.

1. Open the live GPT

Paste a proposed command, file edit, merge, deploy, refund, invoice, or API call and ask whether to allow, block, or checkpoint it.

2. Save the typed signal

Reply in chat with thumbs up: or thumbs down: plus one concrete sentence. Do not rely on ChatGPT's native rating buttons for ThumbGate memory.

3. Enforce locally

Run npx thumbgate init in the repo and configure the documented agent integration so Pre-Action Checks evaluate proposed tool calls before execution. The default policy denies detected secret exfiltration plus commands that kill the gate process or enable its bypass variable.

Open ThumbGate GPT ChatGPT Actions setup Why ChatGPT ads need checks

Find it fast: if the direct link does not open, go to Explore GPTs, search ThumbGate, and choose the GPT by Igor Ganapolsky in Programming. Plain English rule: ChatGPT is the discovery and checkpointing surface. Accepted typed feedback can become a local lesson; recurring failures can become rules. The Reliability Gateway still runs in the configured local agent or CI lane.

See It In Action

See the enforcement before you buy anything

🔍 Live Dashboard Demo

Search lessons, inspect checks, mark a review checkpoint, and see only what changed since the last pass. No signup or install required.

Open Dashboard Demo →

⛔ Check Decisions

Inspect the matched gate, allow/warn/deny decision, and reason returned by the local evaluator.

📊 Local Reports

Generate local dashboard and report views from the operator's available gate and feedback data. A hosted org dashboard is not generally available.

🧱 Isolated Execution Lanes

Workflow Sentinel can recommend Docker Sandboxes for high-risk local actions. The package includes a signed sandbox-dispatch mechanism, but each deployment must configure and verify its isolation boundary.

Data Processing Boundaries

Local enforcement data stays in the operator's ThumbGate feedback directory by default. Checkout, intake, newsletter, and product analytics are separate hosted processing surfaces. Hosted team lesson sync is not generally available.

🧪 Thompson Sampling

Confidence tiers (none/low/medium/high) are available inputs to adaptive rule selection; they do not guarantee a correct decision.

🪞 History-Aware Lessons

When the current Claude auto-capture hook only gets a vague thumbs-down, ThumbGate can reuse up to 8 prior recorded entries plus the failed tool call, then keep a linked 60-second feedback session open for later corrections instead of creating a dead-end note.

Positioning

Enforcement is the missing layer in AI orchestration.

Big orchestration suites unify data, routes, and decisions. ThumbGate sits closer to the moment of execution: the point where an agent is about to run a shell command, ship a PR, approve a release, or repeat a mistake you already corrected. That is where workflow trust is won or lost.

Broad orchestration platforms

Good at customer journeys, routing, and cross-system context. Weak when you need a coding agent or automation to stop before a destructive or low-trust action actually runs.

ThumbGate

Turns accepted operator feedback into local lessons and recurring failures into Pre-Action Checks. Supported agent integrations evaluate matching rules at the tool-call boundary. Detected secret exfiltration and gate kill/bypass commands deny by default; strict mode also denies matching warning-mode checks.

The stack that makes sense

Use orchestration to decide what should happen next. Use ThumbGate to decide what is allowed to execute. That is the control layer enterprises actually need once agents touch repos, terminals, CI, or production workflows.

Compare orchestration vs enforcement → AI deployment readiness → Platform team rollout → Regulated workflow pattern → Built for the Agent Manager →

Self-improving enforcement

Three steps to catch repeated AI failures

Self-improving — but for safety, not capability: human-in-the-loop accepted feedback becomes local lessons, recurring failures can become prevention rules, and regression tests help detect false-positive rule matches.

Feedback

Give 👍 or a concrete 👎 on your AI agent's actions. Accepted feedback is stored in a local SQLite+FTS5 lesson DB. In the current Claude auto-capture hook, a vague thumbs-down can distill from up to 8 prior recorded entries and the failed tool call before promotion, then stay linked to a 60-second feedback session. Recurring accepted failures can auto-promote into a prevention rule.

Distill + Rules

Repeated failures auto-promote into prevention rules. Thompson Sampling adapts which rules fire, and the reflector lane can propose a reusable rule from the same transcript so high-risk patterns get stricter enforcement while low-risk ones stay relaxed.

Checks

Rules become Pre-Action Checks that evaluate proposed tool calls before execution. Detected secret exfiltration and gate kill/bypass commands deny by default; matching force-push, destructive-delete, fetch-and-run, or learned checks warn unless strict enforcement turns them into denies.

Enforcement

Configured checks evaluate proposed tool calls before execution.

Don't trust — verify

Gate results report the decision, matched gate, and available reason or evidence context.

Real tools, not wishes

Configured checks run at the tool-call boundary. Detected secret exfiltration and gate kill/bypass commands deny by default; set THUMBGATE_STRICT_ENFORCEMENT=1 to deny matching warning-mode checks.

Force models to show work

Decision reasons and Thompson Sampling confidence tiers are available where the evaluator produces them.

Review decisions and recurring failures

Recurring accepted failures can auto-promote to checks. Pro includes a personal local dashboard; a hosted org dashboard is not generally available.

Keep risky runs off the host

When Workflow Sentinel predicts a risky local action, ThumbGate can recommend Docker Sandboxes before the agent touches the host filesystem or broader credentials.

Ship with versioned proof

Changesets, SemVer, sync checks, and verification evidence make new package releases inspectable before a buyer trusts the next rollout.

Install for Your Agent

Claude Code

npx thumbgate init --agent claude-code

Wires PreToolUse hooks automatically

Cursor

npx thumbgate init --agent cursor

4 skills: feedback, rules, search, recall

Codex

npx thumbgate init --agent codex

6 skills including adversarial review

Gemini CLI

npx thumbgate init --agent gemini

Gemini CLI integration

Amp

npx thumbgate init --agent amp

Amp agent integration

Any MCP Client

npx thumbgate serve

MCP stdio server for any compatible client

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "thumbgate": {
      "command": "npx",
      "args": ["--yes", "--package", "thumbgate", "thumbgate", "serve"]
    }
  }
}

Official directory review is separate. Claude Code users can install immediately with /plugin marketplace add IgorGanapolsky/ThumbGate and /plugin install thumbgate@thumbgate-marketplace.

Pricing

Stop paying for agent mistakes you already fixed.

	Free $0 forever	Pro $19/mo or $149/yr	Enterprise Custom — scoped after intake
Local PreToolUse enforcement (configured integrations)	✅	✅	Scoped after intake
Feedback captures	2/day (10 total)	Unlimited	Scoped after intake
Active auto-promoted prevention rules	3	Unlimited	Scoped after intake
Lesson recall + search across sessions	—	✅	Scoped after intake
Personal dashboard + DPO export	—	✅	Scoped after intake
Hosted team lesson sync	—	—	Not GA
Hosted org dashboard	—	—	Not GA
Approval boundaries + rollout design	—	—	Scoped after intake
SSO, SIEM + compliance packaging	—	—	Not GA
Best for	One developer	Solo operators	Teams & regulated orgs

Free

Catch repeated mistakes daily. Forever free for solo devs.

2 captures/day, 3 active rules. Enough to see the value — upgrade when you need more.

2 feedback captures/day — 10 total on Free, then Pro for unlimited
Up to 3 active auto-promoted prevention rules
No recall or lesson search
No exports (DPO, Databricks, HuggingFace)
All MCP integrations (Claude Code, Cursor, Codex, Gemini, Amp, any MCP agent)
PreToolUse hook with built-in safety checks — hard-blocks secret exfiltration by default; warns on rm -rf, force-push, and destructive SQL (blocks them under THUMBGATE_STRICT_ENFORCEMENT=1)
Setup guide for all agents →

$ npx thumbgate init click to copy

Install Free

Pro

$19/mo

Individual local governance with higher limits. The free CLI runs configured gates locally. Pro removes the individual caps and adds the personal dashboard, recall, search, and exports. Hosted team lesson sync is not generally available.

Higher local limits — unlimited feedback captures and active auto-promoted prevention rules (Free caps at 3 active rules)
Recall and search accepted lessons across sessions in the configured local scope
Inspect which check fired — personal visual check debugger for available local gate history
Use supported integrations — configure Claude Code, Cursor, Codex, Gemini, Amp, Cline, or OpenCode through its documented setup path
Export training data — DPO preference-pair export for a separate fine-tuning workflow
Review local enforcement evidence — personal dashboard for the individual operator with available gate history
Prepare model-hardening data — Model Hardening Advisor plus HuggingFace dataset export for a separate training and evaluation workflow
Hand a PR with proof — review-ready workflow support and exportable lesson bundles
Hand off without re-onboarding — Lesson export/import for handoff or migration

What your Pro dashboard looks like

checks: local gate history
feedback: unlimited
exports: DPO + lessons

PAY-NOW PRO

Upgrade to Pro — $19/mo

Billed today · cancel anytime.

Compare plans at a glance
Capability	Free	Pro	Enterprise
Best for	One developer evaluating local checks	One operator who wants personal recall, dashboard proof, managed adapters, and exports	Teams & regulated organizations scoping one governed workflow
Price	$0	$19/mo or $149/yr	Custom — scoped after intake
Feedback captures	2/day, 10 total	Unlimited	Scoped after intake; hosted team sync is not GA
Active prevention rules	3 active rules	Unlimited personal rules	Approval boundaries and rollout requirements scoped after intake
Dashboard and proof	Local CLI evidence	Personal dashboard, check debugger, DPO/HF exports	Hosted org dashboard, SIEM, and compliance packaging are not GA
How to start	`npx thumbgate init`	Self-serve Stripe checkout	Send one repeated workflow failure first — intake before checkout

Free provides capped local evaluation. Pro removes individual limits and adds recall, search, a personal dashboard, and exports. Enterprise starts through intake to scope one workflow; it does not imply generally available hosted team features.

Enterprise

Engineering teams · Banking · Insurance · Healthcare · Public sector

Scope one governed workflow — custom pricing after intake

Enterprise is an intake-led service path for teams that need to define approval boundaries, rollback plans, evidence requirements, and rollout ownership around one workflow. The current public runtime remains local per configured operator; hosted team lesson sync and a hosted org dashboard are not generally available.

One-workflow discovery — identify the repeated failure, runtime, owner, and decision boundary
Approval and rollback design — document who can approve, what evidence is required, and how the workflow recovers
Verification plan — define the local tests and proof required before a wider rollout claim
Deployment requirements — capture data-boundary, identity, hosting, and integration constraints before making delivery commitments
Current availability boundary — hosted team sync, a hosted org dashboard, SSO, SIEM, and compliance packaging are not general-availability features

Start enterprise intake → Read the build-vs-buy thesis first

Pricing and deliverables are scoped after intake; no hosted-team capability is promised before verification.

FAQ

Common questions

Does ThumbGate support model fine-tuning?

ThumbGate Pro includes a Model Hardening Advisor plus LoRA JSONL and DPO preference-pair exports for use in a separate fine-tuning workflow. An export does not guarantee that a trained model will avoid a mistake; that outcome depends on the dataset, training process, model, and evaluation.

How is ThumbGate different from model-training feedback loops?

ThumbGate's runtime keeps lessons and rules outside the model. A configured hook evaluates proposed tool calls before execution. Detected secret exfiltration plus commands that kill the gate process or enable its bypass variable deny by default; matching destructive deletes, force-pushes, and fetch-and-run commands warn by default. Strict mode (THUMBGATE_STRICT_ENFORCEMENT=1) turns matching warning-mode checks into denies.

How is ThumbGate different from persistent agent skills?

Persistent agent skills are reusable instructions. ThumbGate adds a local enforcement loop: accepted feedback becomes lessons, recurring failures can become rules, and configured Pre-Action Checks return allow, warn, or deny before tool execution. Pro adds a personal dashboard and operator-managed exports; hosted team sync and a hosted org dashboard are not generally available.

What's the tech stack?

SQLite+FTS5 lesson DB for fast full-text search. MemAlign-inspired dual recall (principle-based rules + episodic context). Thompson Sampling for adaptive check sensitivity per failure domain. LanceDB + Apache Arrow for local vector search with Hugging Face embeddings. ContextFS for context assembly. Bayesian belief updates on each memory. Configured PreToolUse hooks evaluate known-bad actions before execution; detected secret exfiltration and gate kill/bypass commands deny by default, while strict mode also denies matching warning-mode checks. The local path requires no cloud account.

What AI agents and editors does this work with?

ThumbGate includes documented integration paths for Claude Code, Cursor, Codex, Gemini CLI, Amp, Cline, and OpenCode. Each agent must be configured through its hook or MCP setup; MCP compatibility alone does not install enforcement automatically. The Cursor runtime setup is npx thumbgate init --agent cursor, while the in-app Marketplace listing remains separate. VS Code can host an MCP-compatible agent, but this repo does not ship a standalone VS Code extension.

Do I have to chat inside the ThumbGate GPT for enforcement?

No. The ThumbGate GPT is a ChatGPT entrypoint for checking proposed actions and getting setup help. Local allow, warn, or deny enforcement runs where the coding agent executes and requires npx thumbgate init plus the documented integration setup.

When should I use Pro versus the Workflow Hardening Sprint?

Start with the setup guide if you only need the local path. Choose Pro when an individual operator wants higher limits, personal recall, a local dashboard, DPO export, and operator-managed lesson bundles. Choose the Workflow Hardening Sprint when one workflow owner needs to scope approval boundaries, rollback safety, and evidence requirements before a wider rollout claim.

How do we keep high-risk autonomous runs off the host?

ThumbGate combines pre-action checks with execution guidance. Workflow Sentinel can recommend Docker Sandboxes for risky local actions. The package also contains a signed sandbox-dispatch mechanism, but each deployment must configure and verify that isolation boundary; it is not hosted team lesson sync or a generally available hosted org service.

How do we trust a new package release?

ThumbGate does not rely on vibes, making it the safety net for vibe coding. Release-relevant PRs need a Changeset, SemVer rules keep version bumps honest, sync checks keep manifests aligned, proof lanes run before merge, and the exact main-branch merge commit is verified before the work is called done.

Do I need a cloud account?

No. Free keeps local enforcement on your machine with 2 feedback captures/day, up to 3 active auto-promoted prevention rules, and built-in checks. Recall, lesson search, unlimited rules and captures, a personal dashboard, and operator-managed exports open up on Pro. Enterprise is custom and scoped after intake around one workflow. Hosted team lesson sync and a hosted org dashboard are not general-availability runtime features.

What if my thumbs-down is vague?

For the current Claude auto-capture hook, ThumbGate can reuse up to 8 prior recorded entries and the failed tool call for a vague thumbs-down, then keep a linked 60-second feedback session open for later clarification. The timer resets when more context arrives, so the lesson stays attached to one feedback record instead of fragmenting into duplicates.

How are pre-action checks different from prompt rules?

Prompt rules are a starting point, not a finish line. ThumbGate adds a human-feedback loop plus configured Pre-Action Checks at the tool-call boundary. The default policy denies detected secret exfiltration and gate kill/bypass commands while other matching high-risk checks warn; strict mode also denies matching warning-mode checks.

How does ThumbGate fit the agentic development cycle?

Agentic development is not just generation. Teams need Guide, Generate, Verify, and Solve working as a repeatable loop. ThumbGate can serve as the pre-action execution gate: guidance and accepted feedback can become rules, generated tool calls are evaluated before execution, rules can require verification evidence, and recurring solved failures can become reusable prevention gates.

What does Pro cost?

Pro is $19/mo or $149/yr for individual operators and bills through Stripe. Enterprise is custom pricing, scoped after intake around one workflow's approval boundaries, rollback plan, evidence requirements, and rollout support. Hosted team features are not implied by intake.

Does ThumbGate support enterprise Google Cloud / Vertex AI?

The package ships a npx thumbgate setup-vertex setup helper for operators who choose the optional Vertex adapter. Whether evaluations stay within a particular VPC depends on the operator's Google Cloud configuration and must be verified in that deployment. Dialogflow CX governance is an intake-scoped integration requirement, not a generally available hosted deployment claim.

How does ThumbGate contain enterprise API costs?

FrontierBudget provides a local client-side token ledger and configurable routing or cost-containment limits. Reported token savings are estimates based on configured assumptions, not provider-billing measurements or a spend guarantee; verify actual costs in the provider's billing telemetry.

Catch AI agents before they spend money, leak secrets, or break production.

Run your first pre-action check locally.

“Why not just write my own PreToolUse hooks?”

The market moved from “agents can code” to “who is allowed to let agents act?”

MCP and tool lockdown

Managed agents need receipts

Tokenmaxxing backlash

Production code by AI

Guide, Generate, Verify, Solve still needs an execution gate.

Guide

Generate

Verify

Solve

Native thumbs are a black box. ThumbGate is the inspectable control layer.

Black-box memory

Inspectable ThumbGate memory

Rules before execution

Why this matters now

See the footer before you ship the next repeat.

Configure ThumbGate for Claude Code or Claude Desktop.

Use the GPT as a preflight desk for risky commands, refunds, deploys, and PR actions.

One gateway across the agent surfaces you already use

🧩 Claude Desktop Extension

⚡ Claude Code Skill

🤖 AI CLIs

☤ Hermes Agent guardrails

vLLM serving guardrails

Context and tool governance

☁️ Google Data Agent Kit

🧩 Codex plugin

🎯 Cursor integration

✏️ Editor workflows

🗂️ MCP Server Directory

💬 ChatGPT GPT Actions

See the enforcement before you buy anything

🔍 Live Dashboard Demo

⛔ Check Decisions

📊 Local Reports

🧱 Isolated Execution Lanes

Data Processing Boundaries

🧪 Thompson Sampling

🪞 History-Aware Lessons

Logs describe the damage. ThumbGate evaluates the proposed action before it runs.

Pre-action enforcement

Reviewable decision trail

Enterprise governance mode

Enforcement is the missing layer in AI orchestration.

Broad orchestration platforms

ThumbGate

The stack that makes sense

Stop self-improving coding loops from hacking the benchmark.

Guard the metric

Preserve proof trails

Ship into CI

Reusable instructions are the new baseline. Enforcement is the moat.

Portable skill memory

Instructions plus teeth

Proof for teams