Find the work your AI stack should remember.
Paste approved assistant logs, Codex or ChatGPT session JSONL, tool traces, or session exports. Workflow Miner finds repeated prompts, tool chains, failures, and manual handoffs, then turns the evidence into candidates and starter artifacts for reusable prompts, skills, automations, and connectors.
Analysis and redaction run locally. The shared site shell may load fonts and configured analytics; Workflow Miner does not upload or persist raw logs.
Import a text, log, JSON, or JSONL file. The file is read locally.
Add approved activity
Plain text, JSON, JSONL, assistant messages, Codex session exports, and common tool-call records are supported.
Event mix
Sessions
Tools
Set privacy filters
Common tokens, credentials, authorization headers, and secret URL parameters are redacted automatically. Add one custom pattern per line for customer-specific data.
[REDACTED] everywhere.
Mine reusable work
Find candidates only when the source contains supporting evidence.
Your findings will appear here
Load the demo to see how repeated work becomes specific, evidence-backed recommendations.
Detected signals
Recurring behavior found in sanitized events.
Build candidates
Suggested assets ranked by repetition and friction.
Implementation backlog
Evidence-backed order, effort, risk, and first steps for the next sprint.
Sprint brief
A ready-to-review first slice with tasks, tests, rollback, and measurement notes.
From trace noise to a build decision.
Messages, sessions, tool calls, and errors become one local event stream.
Repeated intent, action sequences, failure markers, and handoffs are counted.
Each candidate explains what to build and which evidence supports it.
Use the backlog to pick the first implementation slice, then export the report or starter files.