ThumbGate 1.27.13 is live on npm.
The release matters because the problem it addresses is not abstract anymore: AI coding agents can move quickly, but they also repeat mistakes across sessions unless the workflow has memory and enforcement. A note in AGENTS.md or CLAUDE.md helps, but it is still a request. ThumbGate turns operator feedback into local, enforceable pre-action checks before tool calls run.
The simplest install path is:
npx thumbgate init
The working model is local-first:
- Capture explicit thumbs-up or thumbs-down feedback.
- Promote repeated failure patterns into prevention rules.
- Evaluate actions before they execute, not after damage is done.
- Keep audit evidence tied to the operator's machine unless a hosted workflow is explicitly configured.
That makes ThumbGate useful for teams and solo operators who are letting AI agents touch repositories, browsers, payment flows, deployment scripts, or customer-facing workflows.
The public Pro route is:
https://thumbgate.ai/checkout/pro
Important evidence note: npm reports thumbgate@1.27.13 as the latest package. The public thumbgate.ai health endpoint is healthy, but currently reports deployed version 1.27.8, so this article is a package-release announcement, not a claim that the hosted site is already serving 1.27.13.
ThumbGate is early-stage. The honest commercial motion is straightforward: free local guardrails, Pro for solo operators, and a workflow-hardening sprint for teams that need approval boundaries, auditability, and reliable AI-assisted operations.
If your agent has ever made the same mistake twice, the next version of your workflow should not rely on memory alone. It should enforce the lesson.
Top comments (0)