The control layer
between your AI agents
and the actions they take.
Pulse intercepts every agent action, scores it for behavioural risk in real time, and can stop it before it completes — writing each event to a tamper-evident ledger mapped to ASEAN regulation.
Built around the regulation that actually applies in your region
Agents act at machine speed. Oversight doesn't.
When an autonomous agent does something it shouldn't — a prompt injection, an instruction override, a runaway loop — most teams find out after the action has already completed. Monitoring tells you what happened. Pulse can stop it.
Watch, log, and find out later
- ✕ Detects drift after the action has run
- ✕ Scores each call in isolation, missing intent that builds across a session
- ✕ No tamper-evident record built for a regulator
- ✕ Tuned to EU/US frameworks, not ASEAN obligations
Intercept, score, and enforce inline
- ✓ Holds or blocks a risky action before it completes
- ✓ Scores cumulative session-window risk, not just single calls
- ✓ Every event hash-chained into a tamper-evident ledger
- ✓ Mapped natively to PDPA, Singapore MGF and Vietnam AI Law
One request. Six stages of governance.
Pulse sits inline between your agent and the systems it touches. Every action runs the same path.
Intercept
Your agent's input and output route through the Pulse SDK to /intercept — the single governance entry point.
Score
A cumulative belief–desire–intention score builds across the session, combined with entropy on each input.
Gate
The circuit breaker evaluates risk against thresholds and decides: proceed, hold, or trip — with per-agent cooldown.
Persist
Each event is written SHA-256 hash-chained to the one before it. No deletes, no edits — a tamper-evident ledger.
Surface
A live executive dashboard shows the ledger, severity, risk scores, and an investigator view for any event.
Evidence
One click exports an evidence package or a full incident report, pre-mapped to the right regulatory deadline.
Watch Pulse govern a live agent
A short walkthrough of the dashboard — interception, scoring, the circuit breaker, and regulator-ready evidence export.
Enforcement, evidence, and an open core.
An open-source engine developers can adopt freely, with the compliance layer organisations need on top.
Inline circuit breaker
Hold or block an agent action the moment cumulative risk crosses threshold — not a report after the fact.
Tamper-evident ledger
SHA-256 hash-chained, append-only. Deletes and edits revoked at the database level. Built to stand up to a regulator.
Session-window scoring
Risk that accumulates across a conversation, catching intent that single-call drift detection misses.
ASEAN compliance mapping
Incidents mapped to PDPA Malaysia, Singapore's framework and Vietnam's AI Law — with the right reporting deadline.
One-click evidence
Export a full evidence package or incident report in a single action, formatted for the obligation that applies.
Open-source core
The SDK and engine are free to adopt and self-host. Pay only for the managed compliance and dashboard layer.
An honest note on maturity.
Pulse is an early-stage build, in active development. The architecture is live end-to-end — interception, scoring, the circuit breaker, the hash-chained ledger, and evidence export all work today.
The risk thresholds are being calibrated against real agent traffic, and the compliance mappings are being deepened with practitioners. If you run AI systems in the region and this is a problem you recognise, the most useful thing you can do is tell me where it would break. That feedback is what's shaping the roadmap.
Governance that keeps pace
with autonomous AI.
Request early access, or get in touch for a technical walkthrough and the architecture detail.