Live demo · recorded production runs

Watch an agent complete real work

Four recorded runs from production workspaces, replayed in your browser — the plan, every tool call, every finding, the final deliverable, and what each step cost. No account required.

Replay timing is compressed; original wall-clock per step is preserved in the plan and cost panels. Customer identifiers are replaced.

Code Review Recorded production run · replay compressed
idle
Brief

Select a scenario to load its recorded brief.

Substituted into the recorded run
00:00.0
fo_agent ready · press Run to replay the recorded execution
waiting for run
The deliverable appears here when the run completes.

What you’re watching

Nothing on this page is mocked up for marketing. These are real runs, with the boring parts sped up.

⏺

Recorded, not generated

Each scenario is a production run recorded in May–June 2026. The plan, tool calls, findings, and deliverable are exactly what the agent produced — the replay never improvises.

⏭

Compressed replay

Long-running steps are condensed so a full run plays in about a minute. Original wall-clock per step is preserved in the plan timings and the cost panel — the due-diligence run took 4.1h end to end.

⌨

Your parameter, substituted

The field above each run is substituted into the recorded transcript — brief, tool arguments, deliverable — so you can see how runs are parameterized. It stays in your browser and is never sent anywhere.

🔒

Zero retention, demo included

The platform keeps request data only for the lifecycle of a run, and this page holds itself to the same standard: no demo telemetry, no account, no cookies. Details on the security page.

📊

Cost attribution per step

The Cost tab shows the per-step model routing, token counts, and spend from the original run — the same attribution the June 2026 analytics dashboard gives every production workspace.

🧭

Model-agnostic routing

Steps route across frontier models — Claude Opus 4.8, Claude Sonnet 4.6, Claude Haiku 4.5, GPT-5.1, Gemini 3 Pro — picked per step by the orchestrator. The routing policy is covered in how it works.

248ms
Avg orchestration latency
91%
Deliverables accepted without major revision
4.2h
Avg due-diligence cycle
99.99%
Uptime SLA

Demo questions

Short answers about what this page is and isn’t.

Is the demo calling a model live?

No. Each scenario is a recorded production run from May–June 2026, replayed in your browser. The plan, tool calls, findings, deliverable, and cost table are exactly what the agent produced in the original run — nothing is generated on the fly, so what you see is what the platform actually did.

Why is the replay so fast?

Replay timing is compressed. The original wall-clock per step is preserved in the plan timings and the cost panel — the due-diligence run, for example, took 4.1 hours end to end. The replay condenses idle and long-running stretches so you can watch a full run in about a minute, or faster at 2× and 4×.

What does the parameter field change?

The parameter you type is substituted into the recorded transcript — the brief, tool arguments, and deliverable — so you can see how a run is parameterized. It does not re-execute the run; outputs stay those of the original recording.

Is anything I type stored?

No. The demo runs entirely in your browser — the parameter field never leaves the page, and no analytics are attached to it. The platform itself follows the same principle: zero data retention, with request data living only for the lifecycle of the run. The zero-retention post covers how that works in production.

Can I run this on my own repository or data room?

Yes — that is the product. The Developer tier is free and includes 20 runs per month; create a run with one call to POST /v1/runs or from the dashboard. See pricing for tiers and the docs for the five-minute quickstart.

Run it on your own workflows

The free Developer tier includes 20 runs a month — enough to point an agent at a real PR or a real document set and judge the output yourself.

Start Building See pricing