Free Tier Opens: untactit Brings Version Control and Approval to AI Agent Instructions


Posted August 11, 2026 by untactit

untactit opened public access to its agent asset control plane: a free tier, a public REST API, signed webhooks and a five-language site. It indexes the skills, rules and memory AI agents run on across vendors.

 
TOKYO, Japan - August 12, 2026 - untactit today opened public access to its agent asset control plane, the system of record for the skills, instructions and memory files that AI agents actually run on. The release includes a free tier, a public REST API, signed outbound webhooks, and a website rendered in five languages.

The pricing model is deliberately inverted. Reading, loading, searching and deploying an asset are never billed. Only creation actions - extracting a skill, promoting fragments, running a sandbox test - consume credits. "If distribution cost money, teams would ration the thing that makes the product valuable," the company states on its pricing page.

THE PROBLEM

Companies now run Claude, ChatGPT, Gemini, Cursor, Copilot and other agents side by side. The assets those agents execute - prompt files, rule sets, memory - live in personal folders and vendor settings panels. There is no inventory, no approval step, and no record of who changed what.

untactit lists the symptoms plainly: the same prompt exists in four versions across three laptops and none is canonical; a rule that took a week to get right leaves with the person who wrote it; someone edits a customer-facing instruction and it ships instantly without review; legal asks what data the agents can see and the honest answer is that you would have to go ask around.

"Version control was solved for code thirty years ago," the company writes. "For the instructions running agents against production systems, most teams are still on shared folders and hope."

THREE JOBS

Inventory - untactit reads the folders and vendor accounts a team already uses and builds one searchable index across 17 asset types, with duplicate detection that works across vendors rather than inside a single one. Connecting a platform is read-only by default and changes nothing that is already there.

Governance - personal drafts stay personal. Nothing reaches the company layer without an approver acting on it. Every change is attributed to a person rather than an API key and written to an append-only audit log that is never rewritten. Four roles - admin, approver, editor, viewer - are enforced on the server, so the interface cannot permit an action the backend rejects.

Deployment - untactit decides which assets load into which tool, for whom, and pushes that decision to every machine at once. A local copy that diverges from the approved version is flagged as drift. A deployment can be withdrawn without touching a single machine.

WHAT SHIPPED IN THIS RELEASE

- Public REST API - eleven resource groups behind a versioned path, cursor pagination, and an OpenAPI document served from the same build
- Outbound webhooks - seven event types signed with HMAC-SHA256, retried with backoff, with a delivery log showing exactly what failed
- Multi-tenant store - every read and write scoped by organisation, with delta sync so a client catches up without refetching the whole library
- Stateless authentication - tokens carry the organisation and the role; a client cannot widen its own scope by asking
- Website - live product demos replaced static screenshots, and every page now renders in five languages

PRICING

The free tier covers one person, three skills, five memory files, one connector and 50 credits per month. Team plans are USD 60 per editor per month with unlimited free viewers and a three-layer structure of personal, team and company. Enterprise adds deployment into the customer's own cloud account.

VENDOR NEUTRALITY

untactit stores assets in a neutral format and translates on the way out, covering Claude, Cursor, GitHub Copilot, ChatGPT, Gemini, Claude Code, Windsurf, MCP clients and the AGENTS.md convention read by 20+ agents across 60,000+ projects. "The moment we take a side, we stop being infrastructure and become a bet on someone else's roadmap," the company states.

WHERE THE COMPANY IS

untactit describes itself as pre-launch and says so on its own site: the product works and is in daily use, the customer list is short, and the security page names the certifications it does not yet hold.

AVAILABILITY

untactit is available now at https://untactit.com. Connecting the first workspace takes about ten minutes and requires no credit card.

ABOUT UNTACTIT

untactit is an agent asset control plane for teams running AI agents in production. It inventories agent skills, instructions and memory across vendors, adds an approval layer between personal drafts and company assets, and deploys approved versions to every machine at once.
--- END ---
Contact Email [email protected]
Issued By untactit
Phone +81-3-3527-7891
Business Address 15F Toranomon Hills Business Tower, 1-17-1 Toranomon
Minato-ku, Tokyo 105-6415
Country Japan
Categories Computers
Tags ai agent governance , prompt management , agentsmd , agent assets , audit log , deployment , saas , vendor neutral
Last Updated August 11, 2026