Under construction — Colorcast is in early access while we finish building the best experience. The download and free tier work today; paid plans open soon.

Docs

Everything you need on the first day, in the order you'll need it.

Set up in five minutes

  1. Output. Audio → pick where narration should play. For stream only: your speakers or a dedicated virtual cable captured by OBS. For Discord too: a virtual cable set as your Discord mic alongside your real mic (VoiceMeeter mixes them).
  2. Persona and game. Session → choose a commentator and a preset. The Naturalist and Booth are free.
  3. Triggers. Each preset has default hotkeys per moment; remap them on the Triggers page. The defaults work while the game has focus.
  4. Mic (optional). Turn on mic callouts and say a moment — "got him", "clutch", "that was lag". Recognition runs on your PC.
  5. OBS. Add a Browser Source at http://127.0.0.1:3847/overlay. On the Audio page, connect OBS ducking (obs-websocket) to lower game audio while the persona speaks. More on the OBS overlay page.

Stream Deck

Install the plugin (details on the Stream Deck page). Four actions: Trigger a moment, Say a custom line, Mute for a while, and Switch persona or preset. Paste the API token from Triggers into the plugin's global settings once.

Game events

Triggers → Game adapters turns on automatic moments for supported games. Setup for each: Counter-Strike 2, League of Legends, Dota 2.

Stingers

Repeated moments (kill, death, victory) are pre-rendered per persona, voice and game, saved on your PC, and never rendered again — switching games all night costs nothing. Settings → Stinger library shows each cached set and lets you delete it.

Local API

GET http://127.0.0.1:3847/trigger?event=kill&token=… fires a moment; POST /transcript sends a callout; /mute, /unmute, /status, /events (SSE) and /overlay round it out. State-changing calls need the bearer token shown on the Triggers page. Streamer.bot, Lioranboard and shell scripts all work.

Privacy, plainly

Mic audio is transcribed on your PC unless you switch to cloud recognition. Hosted plans send the moment name and a few words of context to our relay to write and speak the line; we keep counts, not transcripts.

Troubleshooting