Flashback
100%
DETERMINISTIC

Same recorded run every time you replay the cassette.

$0
REPLAY COST

Local replay from cassette costs zero tokens.

0
SIDE EFFECTS

Nothing re-runs live on pure replay.

A few lines of code to
record everything.

Record each run as a local cassette. Replay or fork offline for failures, and when you need to audit a run that succeeded.

  • Full cassette every run, success or fail
  • Replay offline for $0, no live side effects
  • Fork from any step to debug or verify a fix
agent.ts
import { createRecorder } from '@flashback/sdk'

const rec = createRecorder({ agentName: 'order-agent' });
const llm = rec.llm(myLlm);
const charge = rec.tool('stripe.charge', chargeFn);

// run your agent with llm + charge …
const cassette = await rec.end();

Zero-cost time travel.

Watch Flashback take a failed production run, replay it from the cassette, and fork from the failure so you only re-run what you need.

Ready
Production Run
$0.000ms
waiting…
Flashback Replay
$0.000ms
waiting…
Cassette replay
$0.00$0.00
No tokens, no live tools
Fork live tail
$0.00$0.00
Only this step
Side effects
3 live callsIsolated
1 live LLM after fork

Audit any agent run.

See the exact prompts, tool calls, and context for any run. Investigate a failure, or audit a successful workflow, by stepping through the cassette.

Example Domains

Massive block trade executed on a hallucinated earnings report.

Your agent scraped Twitter instead of the official SEC filing, hallucinated a positive earnings beat, and triggered a massive buy order. Flashback lets you time-travel to the exact tool call where the bad data was ingested, fix the parsing logic, and replay the decision safely in complete isolation.

Stop guessing why your agents fail.

Get early access to time-travel debugging for production AI agents. Local cassette replay and fork.