Execution Authority for AI agents
When your AI agents or automations try a payment, refund, price change or API call, your policy decides before it commits. Start by watching: nothing is blocked until you switch it on, and every decision leaves a signed record you can check yourself.
ALLOWESCALATEBLOCK
Illustrative decision stream
Treasury agent
Today's payment run
Pay supplier invoice
ALLOWsingle-use grant issued
Sweep surplus cash
ALLOWsingle-use grant issued
Reclassify intercompany transfer
ESCALATEIC-0192 · priority normal → low
held for approval
No policy covers this call
Presence asks your treasury lead before it commits
Approved for this exact action single-use grant issued
Decision Dossier signed Ed25519 · DSR-5B21-7E04
Fund payroll account
ALLOWsingle-use grant issued
Decionis products
Pick the product for the actions you want to stop first. AgentSafe checks your agents' API calls, MCP tools and deployments. Commerce Gate checks prices, orders, promotions and refunds on your store and marketplaces. Presence verifies a named human when policy requires approval. Both products use the same policy decisions and signed records.
One policy check before the action commits
Your policy → allow, escalate to a person, or block → a signed record you can check
For your agents, APIs and CI/CD
Stop agent API calls your policy doesn't allow.
Put the gateway in front of your agents. Every write is checked before it is sent: it goes through once, waits for your approval, or is refused.
Agent tool calls, API requests, MCP tools, deployments and infrastructure changes.
Start NowInstall AgentSafeInstall it yourself with Homebrew, Docker or Helm. Buy a plan direct; no marketplace needed. Plans
For your store and marketplace orders
Protect your margin before an order, price or refund commits.
Live on Walmart, Shopify and WooCommerce. Stop the discount, order or refund that breaks your margin rules before it commits.
Price changes, orders, promotions, fulfillment, inventory and refunds.
Start NowSee what Commerce Gate stopsBuy a plan direct, or install from the Shopify App Store, WordPress.org or the Walmart Marketplace App Store. Plans
Presence verifies the person who approves, when your policy asks for one, for that exact action.
Decision Dossier keeps the verdict, the policy in force and the signature, so you can check it without asking us.
AgentSafe is open source. It enforces the verdict of a Decionis authority: the hosted service, or one you run in your own cloud under a commercial licence (Private beta). The open-source gateway never decides policy by itself.
With Decionis, you stop the payments, price changes, orders and API calls your policy doesn't allow before they commit. Anything outside your policy waits for the person you name. Everything else runs with a single-use permit for that exact action, and every decision leaves an Ed25519-signed record you can check yourself, without asking us.
Enforcement applies at the boundaries you connect. Each integration publishes its own coverage and what happens if we can't be reached.
How you start
Your systems keep running the actions, and nothing changes until you switch one to enforce.
Pick what you want to protect, your agents' API calls or your store, and confirm your email. Your workspace opens on the free plan.
Run AgentSafe in front of one agent, or install Commerce Gate on your store. Nothing is blocked until you switch an action on.
Every decision shows what your policy would have done, with a signed record you can verify yourself.
Switch one action to enforce. Anything outside your policy now waits for the person you name, and you can switch back at any time.
Proof you can verify
Each time a payment, change or API call is allowed, held or blocked, you get a signed record of what was decided, under which policy and with whose approval.
The record is made when the action is allowed, held or blocked, so nobody has to write it up afterwards.
It names the action, the policy in force and who approved it, so it can't drift from what actually happened.
Every record is Ed25519-signed. You verify it offline against the published keys with one command.
Try it nowpublic corpus · DOI 10.5281/zenodo.22692903npx -y @decionis/verify
See an example recordTrust
Every Decision Dossier is signed with Ed25519, and the public keys are published. You can verify any dossier offline, without an account and without calling us.
An approved action carries an execution grant bound to that exact action. The grant is claimed once, immediately before the action runs, so a replayed request is refused.
Run the Decionis Authority in your own cloud from the versioned Helm chart, under a commercial license. Private beta.
Every public endpoint accepts TLS 1.2 or 1.3 only. Stored data is encrypted at rest with AES-256.
AgentSafe v0.3.5 ships with a CycloneDX SBOM, signed build provenance and a Sigstore-signed release tag.
Report a vulnerability privately to security@decionis.com. Leave production keys, customer payloads and unredacted dossier contents out of the report.
Choose a product, see a decision and understand what changes when you connect it.
Execution Authority answers one question before an action commits: may this exact action run now, under your policy and the approvals it requires? Your identity system proves who is acting and your guardrails judge content and risk. Execution Authority decides whether the resulting payment, change or API call may commit, and leaves a signed record you can verify.
Pick AgentSafe to stop the agent API calls, MCP tool calls and deployments your policy doesn't allow. Pick Commerce Gate to protect your margin on prices, orders, promotions, fulfillment and refunds. Both use the same policy decisions, the same approval step for the person you name, and the same signed records.
Run a live decision in the sandbox with synthetic data. It needs no account, email or card. Pick an action, see what your policy decides and inspect the record. From there you can start free with AgentSafe or Commerce Gate.
You start in Shadow Mode: every action on the workflow you connect is checked, and nothing is blocked. You see what would have been allowed, held for a person or blocked, adjust your policy, then switch on enforcement for one action. You can put the same integration back into Shadow Mode at any time.
No. Your own history is checked and a policy is drafted from what was found, with every limit citing the findings behind it (Walmart in Beta, Business Central in Private beta). Before anything goes live it is replayed on the same records, so you see what it would have held. Then a person you name approves it: run it with your own team, where a second person approves every change, or have it managed for you and give the one approval that releases it (Private beta). You can also start from a policy pack.
Yes. Every verdict produces an Ed25519-signed Decision Dossier, and you can check the signature against our published JWKS offline — no account, and no need to trust our code for the result to mean something.
Pick AgentSafe for your agents' API calls, or Commerce Gate for your store and marketplace orders. Your workspace opens as soon as you confirm your email.
Shadow mode first · nothing enforced until you say so
