Agent safety · open beta

Ship agents you actually trust.

Chronoly records every agent action, enforces OS-level boundaries, and stops runaway loops before they burn budget.

Join early access How it works

60s setup·self-host·open source

chronoly · agent tracelive
$ chronoly watch 'claude-code'
→ Blackbox: tracing enabled
● Read src/server.ts
● Write src/middleware/auth.ts
⚠ Shield: blocked rm -rf node_modules
◉ Cost Guard: $0.07 · 3 turns · no loops
✓ Auth added. 0 safety violations.

Works with

Claude CodeCursorHermesOpenAI CodexClineAider

Platform

Three tools. One runtime.

Blackbox records. Shield enforces. Cost Guard kills loops. Use one or all three.

Blackbox tracing

Every prompt, tool call, file change, and command recorded as a replay-grade timeline — not chat history.

→Prompt + response log
→Tool-call chain
→File diff + rollback
→Session replay

eventtool.call
nameWrite
pathsrc/middleware/auth.ts
tracesess_7f2a · 3 turns

OS-level shield

Boundaries at the OS and network layer. Rules agents cannot prompt-bypass.

→Per-path write guard
→Network filter
→Command sandbox
→Per-tool policies

actiondeny
rulefs.write.block
cmdrm -rf node_modules
layeros · not prompt

Cost guard

Loop detection, per-task token budget, and a circuit breaker when a session runs away.

→Loop detect
→Token budget / task
→Cost-per-success
→Auto circuit breaker

spend$0.07
turns3
loops0
stateok

How it works

Install. Run. Inspect.

01

Install

One env var. Claude Code, Cursor, Hermes, OpenAI Codex, and OpenAI-compatible agents.

02

Run

Actions recorded and scored in real time. No SDK changes required.

03

Inspect

Traces, costs, and blocked actions. Replay a session when something goes wrong.

Pricing

Start free. Founder pricing while beta lasts.

Early adopters keep the rate they join on.

Hobby

/forever

Tinkering and side projects.

→3 agents
→7-day trace retention
→Cost Guard basic
→Community Discord

Start free

Most popular

Pro

$29

/month

Shipping agents daily.

→Unlimited agents
→90-day trace retention
→Cost Guard + Shield
→Priority email
→Self-host or cloud

Join early access

Team

Custom

Teams in production.

→SSO + audit log
→SLA + dedicated support
→On-prem deploy
→Custom integrations
→Everything in Pro

Contact sales

FAQ

Straight answers.

No accordion. Read what matters, then ship.

Does it slow down my agent?

Under 2ms overhead per tool call. Async logging, non-blocking checks.

What agents does it work with?

Claude Code, Cursor, Hermes, OpenAI Codex, and any OpenAI-compatible proxy that calls tools, shell, or files.

Can I self-host?

Yes. Local or your cloud. No data leaves the machine unless you opt into cloud features.

Is this just another dashboard?

Dashboards show what happened. Chronoly also blocks what should not happen — at the OS layer, not the prompt layer.

Early access

Stop agents from burning the runway.

Join the waitlist. Lock founder pricing. Ship agents you can audit.

→ self-host ok→ no spam→ founder rate locked

Email only when something ships.