Sign in
Open your TookEffect account and Workspace.
TookEffect manual
How to use TookEffect from connection to proof.
You do not need to understand the internal architecture first. Connect a supported system, give your AI agent controlled TookEffect access, and require a verified result before important workflows continue.
Checks the real resultOnly touches what you allowStops when unsureKeeps proof
01The AI asksone authorized action on one exact target
02TookEffect runs itthrough the supported connection
03TookEffect checks realityreads the real external system again
04Your workflow gets proofclear result + evidence + receipt
VERIFIED RESULTAPPLIED
Expected outcome confirmedStart here
The five-step TookEffect setup.
Quickstart, Dashboard onboarding and Docs use the same sequence.
Connect a supported system
Authorize the account and exact resource TookEffect is allowed to use.
Connect your AI Agent
Create a separate TookEffect identity for the Agent instead of giving it your provider credential.
Choose what the Agent may do
Allow the action, require human approval, or keep it blocked.
Verify the first real action and inspect proof
Run a supported action, wait for TookEffect to check the real provider state, then inspect Evidence and the signed receipt.
The one rule to remember
The AI can ask for the action. TookEffect checks whether the expected result really happened.What TookEffect does
A success response is not proof.
An external service can accept a request without ending in the state you expected. TookEffect checks the final state independently before it tells the workflow to continue.
AI agentAsks for an actionone exact target and expected result
→TookEffectRuns and checksauthorize · act · read again · decide
→Real systemFinal state existsGitHub · Cloudflare · Vercel · certified integrations
→ProofResult you can keepdecision · evidence · receipt
request → act → check the real system → confirm the result → keep proofDashboard
Where to find everything.
The Dashboard is the control panel. The names below are the product terms you will see there.
Accounts you connect
The external systems TookEffect is authorized to reach.
Exact resources
The repository, Worker, Project, or other exact resource an action may use.
Access for each AI
Create and revoke a separate TookEffect token for every agent or automation.
Rules and approvals
Restrict what an agent can do and require a human where needed.
What really happened
See the request, observed state, decision, reason, and receipt after a run.
Safety state
Keep unavailable or uncertain integration paths disabled instead of pretending they are safe.
Connect systems
Connect, authorize, choose the resource, done.
A Connection is the external account authorization. A Verified Target is the exact resource inside that account that TookEffect may act on and verify.
GitHubActive
Select Connect GitHub, authorize/install the TookEffect GitHub App, choose the repositories TookEffect may access, and let TookEffect verify the installation and resource. Normal users do not copy repository IDs or access tokens.
Connect GitHub →Cloudflare WorkersActive
Select Connect Cloudflare, authorize TookEffect, choose the Worker, and let TookEffect verify it before creating the Connection and Verified Target. Account ID, Worker name and API token remain available only under Advanced · connect with an API token as a fallback.
Connect Cloudflare →VercelActive
Select Connect Vercel, authorize TookEffect, choose the Project, and let TookEffect verify the production Target. Team ID, Project ID and access token remain available only under Advanced · connect with an access token as a fallback.
Connect Vercel →Why exact Targets matter
Authorizing one repository, Worker, or Project does not authorize every resource in the account.Connect an AI agent
Give the AI a TookEffect key, not your provider key.
Create one TookEffect token per agent. You can revoke that token without rotating the credentials for GitHub, Cloudflare, Vercel, or another connected system.
https://tookeffect.com/mcpUse this when your AI agent supports MCP tools.
https://tookeffect.com/api/v1/effects/...Use this for custom agents, servers, scripts, and workflows.
Safe token handling
- Use a different token for every agent or automation.
- Keep the token in the agent's secret manager.
- Do not paste provider credentials into prompts.
- Revoke the TookEffect token when the agent is retired or compromised.
Simple rule for the agent
For supported actions, use TookEffect. Do not continue until TookEffect returns a completed APPLIED result. Treat NOT_APPLIED and AMBIGUOUS as stop-and-review states.Run a verified action
Choose an action TookEffect knows how to prove.
Each supported action defines what should change and what TookEffect must read back from the real system before it can return success.
ACTIVE
Verified Merge
GitHub · merge one exact pull request into one exact destination branch and confirm the result from GitHub.
POST /api/v1/effects/github/merge-pull-requestACTIVE
Verified Deployment
Cloudflare Workers · apply an authorized Worker deployment and verify the final deployment and traffic state.
POST /api/v1/effects/cloudflare/deploy-worker-versionACTIVE
Verified Promotion
Vercel · promote an authorized deployment and verify from Vercel that production aliases map to the requested deployment.
POST /api/v1/effects/vercel/promote-deployment Retry safely
If the same request must be retried, keep the exact same input and idempotency key so TookEffect does not intentionally repeat the logical action.Read the result
Three decisions. No guessing.
Your workflow should decide what to do from the TookEffect verdict, not from the external service's HTTP success response.
TookEffect proved that the expected final result exists.
TookEffect proved that the expected result did not happen.
TookEffect could not establish the truth safely, so it refuses to guess.
For normal use: continue only on a completed APPLIED result.
Keep and verify proof
The evidence remains after the AI is finished.
Every completed verified action keeps the request, observed state, verdict, reason, and receipt. Most users only need to inspect this in the Dashboard; technical systems can verify the receipt independently.
Read the run in the Dashboard
See what was requested, what TookEffect observed, and why it returned its result.
Verify the signed receipt
Production receipts use Ed25519/JWS public-key signing. Use the published verification keys or the dependency-free verifier when independent cryptographic verification is required.
/api/v1/receipt-keysPublished public keys for receipt verification.
/verify-receipt.mjsDependency-free verifier for technical integrations.
Multiple agents
Give every agent its own identity and boundary.
Do not share one permanent token across every AI. Separate access makes it possible to revoke one agent, audit its actions, and apply different rules without disturbing the others.
One token per agent
Use clear labels so you know which AI or automation made each request.
Limit what it may touch
Keep each agent bound to the systems, targets, and actions it actually needs.
Require a human when needed
Consequential actions can be held behind explicit approval instead of granting blanket autonomy.
Keep separate evidence
Evidence remains attributable to the TookEffect identity that requested the action.
Security
Assume uncertainty and fail safely.
TookEffect is designed not to manufacture certainty. If authorization, provider state, or the final result cannot be established safely, the workflow should stop rather than pretend success.
Authorize exact resources
Connections and Targets limit which real systems an action may reach.
Keep provider credentials away from agents
Agents use TookEffect tokens instead of receiving the underlying provider secret.
Uncertainty does not become success
AMBIGUOUS is a stop state, not an invitation to guess.
Keep evidence after execution
Completed results remain inspectable and signed receipts can be independently verified.
Troubleshooting
When something fails, start with the boundary.
Check the Connection
Confirm the external account authorization is still valid.
Check the exact Target
Make sure the requested repository, Worker, Project, or resource is the one you authorized.
Check the agent token
Confirm the token belongs to the intended agent and has not been revoked.
Read the TookEffect verdict and reason
APPLIED, NOT_APPLIED, and AMBIGUOUS tell the workflow what happened from TookEffect's point of view.
Do not bypass uncertainty
If the result is AMBIGUOUS or an integration is gated, stop and investigate instead of calling the provider directly just to make the error disappear.
Still stuck?
Use TookEffect Support with the run ID and a description of what you expected. Do not send provider secrets.Ready
Let the AI act. Make TookEffect check.
Use the Quickstart for the shortest path, or open the Dashboard to connect your first supported system.