Witness
Voice-anchored evidence recorder for AI coding agents
Witness records product feedback the way AI coding agents need it. One CLI launches a managed browser that captures speech transcription, clicks, network and console activity, plus rrweb DOM snapshots into a local .witness/ evidence bundle. Agents then query that bundle through 12 MCP tools — so "that thing that broke a second ago" becomes a queryable evidence chain instead of a screenshot pasted into chat. Local-first: nothing leaves the machine unless you say so.
Features
Launches a managed Chromium recording session in one command; Captures voice with live transcription anchored to clicks and navigation; Records network requests, console errors and rrweb DOM snapshots; Stores everything locally under .witness/ as a queryable evidence bundle; Ships an MCP server with 12 tools for agents to query recordings; Works with Claude Code, Cline, Cursor and any MCP-capable agent
Use Cases
Report a bug by talking and clicking instead of writing repro steps: your agent reads the exact click, console error and request that broke; Give an AI coding agent ground-truth context on UI regressions — it queries the evidence bundle instead of guessing from screenshots; Capture a user interview or dogfood session and let the agent extract issues with timestamps; Turn "it flickered once and I lost it" into a replayable DOM + network timeline
Comments
Built Witness after one too many "it broke for a second, what did you see?" moments working with coding agents. Screenshots and chat descriptions lose the actual evidence — the click, the console error, the request that failed. Witness records voice + clicks + network + DOM into a local bundle your agent can query through MCP, so "that thing that just broke" becomes data instead of a guess. Fully local-first, MIT licensed. Would love to hear how you capture context for agents.
Premium Products
Sponsors
Buy
Comments
Built Witness after one too many "it broke for a second, what did you see?" moments working with coding agents. Screenshots and chat descriptions lose the actual evidence — the click, the console error, the request that failed. Witness records voice + clicks + network + DOM into a local bundle your agent can query through MCP, so "that thing that just broke" becomes data instead of a guess. Fully local-first, MIT licensed. Would love to hear how you capture context for agents.
Premium Products
New to Fazier?
Find your next favorite product or submit your own. Made by @FalakDigital.
Copyright ©2026. All Rights Reserved