SYSTEM
The system executes
Deterministic, reliable, fast. dbt runs, API calls, Lambda functions, or any existing system already doing the work. Full execution record with status, timestamp and outcome.
A new category
The Supervised Autonomy Platform
Supervised autonomy is the operating model for organizations deploying AI agents safely in consequential work.
Traffic Lights is that platform — for work done by humans, AI agents, and your existing software.
The operating loop
Three steps that turn autonomous agents into a governed system your people can run — and a loop that closes on your terms.
Step 1: Define
Build your process visually or with your AI tool — tasks, dependencies, policies, thresholds. No coding required.
Step 2: Monitor
Red, amber, green health signal per process. SLA tracking. Proactive notifications — or just ask your AI tool.
Step 3: Govern
Agents execute within your thresholds. Exceptions reach the right person with full context. Every decision logged.
Improve · Govern back to Define
Outcomes feed back into policy. Patterns surface. Thresholds adjust.
But only when you say so.
Process Builder
Build your process visually.
Tasks, dependencies, which steps run automatically and which require human sign-off. No code required. Your first process in 15 minutes.
Month-End CloseReadyv14 · 8 tasks · 1 branch
Save changesexceptions found
Custodian + admin fileMachine
Price validationMachine
Exception triageAgent
Approve material adjustmentAwaiting sign-off
NAV sign-offHuman
Distribute statementsMachine
Break investigationAgent
Independent corroborationAgent
Traffic Lights MCPconnected
claude · month-end-closePolicy Engine
Three actors, one governed plane.
You set the thresholds; the thresholds are versioned and logged.
AGENT_SUPERVISED
The agent acts
Inside the threshold you ratified, the agent proceeds on its own. Every action is logged with actor, timestamp and outcome.
AGENT_REVIEW
A human decides
Outside the threshold, work pauses and reaches the accountable person. Approval unlocks only once they have engaged with the context.
The approval surface
Not a rubber stamp. A meaningful decision.
The reviewer sees what the agent did and why. Approve stays locked until they have engaged with the context.
Green · healthyYellow · attentionRed · breach
The feedback loop
Outcomes feed back into policy.
Thresholds adjust as patterns emerge. The system surfaces where automation is safe to expand and where oversight needs tightening — and a human ratifies every material change.
Approval surface — mobile
Client comms · draft 4Yellow · attention
Agent drafted an email to a Tier-1 account referencing a fee change outside the ratified threshold.
- actor:
- agent.comms.v3
- policy:
- AGENT_REVIEW · threshold 0.82
- reviewer:
- r.okafor · engaged 41s
ApproveDelegateReject
Actions unlocked after full context reviewed
Audit trail
When someone asks, you have an answer.
Every process, task and policy is recorded bitemporally: what was true, from when, until when, and who changed it. Nothing is overwritten. Ask what the rules were on any date and get the exact answer.
AuditAs ofmm/dd/yyyy
Search 25 rows…
More rows exist beyond this page. The page count shown is what has been confirmed so far, not a total.
| Task ID | Task | Type | Policy | Recorded from | Recorded until | Changed by |
|---|---|---|---|---|---|---|
| 10000512 | Approve material adjustment | Human | Delegate Approval | 25 Aug 2026, 09:43:58 PDT | current | r.okafor |
| 10000512 | Approve material adjustment | Human | Escalate · 30 min | 22 Aug 2026, 14:02:11 PDT | 25 Aug 2026, 09:43:58 PDT | m.tanaka |
| 10000509 | Exception triage | Agent | AGENT_SUPERVISED | 22 Aug 2026, 13:58:40 PDT | current | m.tanaka |
| 10000509 | Exception triage | Agent | AGENT_REVIEW | 19 Aug 2026, 10:12:07 PDT | 22 Aug 2026, 13:58:40 PDT | agent.recon.v2 |
| 10000501 | Price validation | Machine | Auto Retry · 3 | 19 Aug 2026, 10:11:52 PDT | current | m.tanaka |
Integrations
Any stack.
Any cloud.
One event.
Connects to everything your team already uses.
No rip-and-replace. No new infrastructure. If it can send an HTTP request, send an email, or post to Slack — it connects.
curl -X POST https://api.traffic-lights.io/api/v1/events \ -H "X-Api-Key: $TL_API_KEY" \ -H "Idempotency-Key: $(uuidgen)" \ -H "Content-Type: application/json" \ -d '{"source": "sai.tl.billing", "detail-type": "FLOW_CALLBACK", "detail": {"runID": "…", "taskID": "…", "status": "SUCCESS"}}'# → 202 {"eventId": …} · no SDK requiredDeveloper APIs
Python SDK · Node SDK · HTTP API · REST webhooks
One line to emit an event from any codebase, any cloud.
No-code Automation
Zapier · Make · Power Automate · n8n
Connect any tool in your stack without writing a line of code. If your tool has a Zapier action, it can emit to Traffic Lights.
AI agents over MCP
Works with Claude · ChatGPT (Plus and up) · Gemini Enterprise · Microsoft 365 Copilot (via Copilot Studio)
Connect your AI assistant to Traffic Lights over MCP. It can invoke Traffic Lights directly — emit events, trigger approvals, escalate tasks — all governed by your policies.
Chat and messaging
Slack · Microsoft Teams · Email
Trigger a governed process from a message, a channel post, or a forwarded email. Type a command. Send a status. Traffic Lights handles the rest.
/tl emit FLOW_CALLBACK · task-abc@in.traffic-lights.io
Security posture
We never reach into your infrastructure.
- Traffic Lights does not connect to your databases, applications, or anything inside your firewall. You emit events to us. We govern what happens to them.
- The Event SDK schema is publicly documented. You control which fields are included. No PII required. No proprietary data required.
- Your data is yours. Traffic Lights uses anonymised, aggregated patterns — escalation rates, threshold distributions, approval outcomes — only to surface improvement suggestions within the platform. You can opt out at any time.
Read our privacy policy and terms of service
Early access
Define your first governed process in 15 minutes.
No infrastructure to deploy. No seat licenses for approvers. Approvers tap from their phone.