Axtary
A checkpoint that sits between your AI agent and the real world, so risky actions (a big code change, a message to a client) need a human's okay before they actually happen.
🔗 Visit Axtary
Description
Giving an AI agent access to your code, your messages, or your cloud accounts is powerful but scary — one bad decision and it's already sent the email or deleted the file. Axtary is a permission layer for AI agents: routine, low-risk actions run automatically under rules you set, while higher-risk actions (a big diff, an external message, a destructive cloud operation) pause and wait for a human to approve them first.
Technically, Axtary uses "payload-bound" authorization tokens tied to the exact content hash of an action, so an approved action can't be silently swapped for a different one before execution — a tamper-detection mechanism that rejects mismatched payloads even after approval. It connects to MCP servers and native integrations (GitHub, Slack, Linear, Jira, AWS, GCP, and others), runs policy checks locally via a lightweight runtime coordinated by a hosted control plane, and keeps a verifiable, append-only audit ledger of every decision. Policy rules are compatible with the Cedar and OPA (Rego) engines, and SDKs are available for action approval, policy definition, proxying, ledger access, and MCP/CLI integration across Anthropic, OpenAI, and Cursor agent runtimes. It's Apache 2.0 licensed (v0.5.0).
💬 Our review
The short version: if you're giving an AI agent real access to your systems and the idea of it acting unsupervised on something destructive keeps you up at night, Axtary's human-in-the-loop approval gate — with tamper-proof, content-hash-bound tokens — is a concrete answer rather than a vague promise of "safety."
The free local tier ($0, no account) is a genuinely low-risk way to try policy enforcement before paying anything, and the tamper-detection design (rejecting payloads that don't match what was approved) is a meaningfully stronger guarantee than a simple yes/no approval prompt. Once you want hosted approval inboxes, a team dashboard, and the verified ledger, the jump to $499/month (after a 30-day pilot) is steep for a small team, positioning this squarely at funded startups or enterprises rather than solo builders — for a hobby project, the free local tier or simply writing your own guard rails is more realistic. Compared to building this yourself with custom middleware and manual review scripts, Axtary buys you a maintained policy engine (Cedar/OPA compatible), pre-built connectors, and an audit trail out of the box, which is real time saved once an incident review actually matters.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Local : gratuit (sans compte) · Founding Team : 499$/mois après pilote 30 jours (setup guidé, approbations hébergées, dashboard équipe, ledger vérifié) · Enterprise : tarif sur devis
Pros
Tokens d'autorisation liés au hash exact du contenu (anti-sabotage)
Palier gratuit local sans carte bancaire
Ledger d'audit vérifiable et infalsifiable
Compatible moteurs de policy établis (Cedar, OPA/Rego)
Connecteurs natifs GitHub, Slack, Linear, Jira, AWS, GCP
Cons
499$/mois est un saut de prix important après le pilote gratuit
Palier gratuit local, sans les fonctions d'approbation hébergées les plus utiles en équipe
Produit jeune, peu de retours d'usage en production à grande échelle