ackd waveform

The human ACK that lets
your AI agents proceed.

When an agent wants to take a sensitive action, it pauses.
You review. You send the ACK. The agent proceeds.

join the waitlist request early access

open source · MIT license · free to use

the problem

AI agents are taking real actions.
Who's watching?

no approval gate

AI agents can delete, write, and deploy without any human checkpoint. By the time you know what happened, the action is already done.

no audit trail

When something goes wrong, you can't tell which agent took which action, or who approved it. You're debugging in the dark.

no cryptographic proof

"The agent said it was approved" is not something your SOC2 auditor will accept. Assertions without evidence are not controls.

how it works

Three steps. One human decision.

01

agent requests authorization

The AI agent calls the ackd MCP tool with an action and justification. ackd creates a consent request and pauses the agent.

02

human reviews and ACKs

You receive a notification with full context: what the agent wants to do, why, and who's asking. You approve or deny on the consent screen.

03

agent proceeds with a verified identity

On approval, ackd issues a time-limited SPIFFE SVID — a cryptographic identity tied to that specific approval. The agent uses it to act. No SVID, no action.

mcp tool call pauses until human ACKs 
const result = await ackd.request_authorization({
  agent_id:     "deploy-agent",
  action:       "terraform apply --target=prod",
  justification:"hotfix for CVE-2026-0001",
  ttl_minutes:  30
});
// Pauses here until human ACKs

under the hood

One decision. One identity. One audit trail.

the authorization flow

AI agent wants to take a sensitive action request_authorization() ackd gateway pauses agent · creates consent request notification email · slack consent UI browser screen human ACKs human decision approve or deny approved SVID issued agent proceeds denied blocked no action taken

the identity proof layer

SPIFFE TRUST DOMAIN SPIRE server
issues short-lived x.509 SVIDs · tied to the specific approval event
 ackd gateway
validates human decision before requesting SVID from SPIRE
 audit log
every request stored · who approved · when · SVID expiry
 SVID logs SVID identity: spiffe://dev.ackd.io/agent/{agent_id}/{request_id} time-limited · cryptographically bound to one approval · non-transferable no SVID = no action. the SVID is the proof of approval.

for developers

Add a human approval step to any AI agent in 10 minutes. Open source, MIT license. Install via GitHub and get started in minutes.

join the waitlist →

for security teams

Complete audit trail of every agent action — who approved it, when it expired, what identity was issued. Built on SPIFFE. Compliance-ready from day one.

learn about compliance →

built on

Open standards. No vendor lock-in.

MCP

Model Context Protocol — the open standard for connecting AI agents to tools. Any MCP-compatible agent works with ackd out of the box.

SPIFFE / SPIRE

CNCF workload identity standard. Human approvals issue time-limited SVIDs — cryptographic proof that ties each action to its authorization.

AWS

Built on Lambda, DynamoDB, Cognito, and API Gateway. The cloud tier is for teams who want it managed. Open source for everyone else.

early access

ackd is in active development.

Join the waitlist for early access, founding customer pricing, and direct input on the roadmap.

I'm interested in ackd for

how did you hear about us? (optional)

no spam. no marketing drip campaigns. just a direct email from tim when ackd is ready.