| name | moltstream |
| description | AI agent-native streaming runtime. Use MoltStream to build agents that operate live streams — control scenes, react to chat, manage overlays, and broadcast to Twitch/YouTube/Kick. All through a typed SDK, no GUI.
|
| capabilities | ["stream-management","scene-composition","chat-reaction","audience-analytics","multi-platform-broadcast","reasoning-trace-logging"] |
| context_cost | 400 |
| ttfa_minutes | 3 |
| min_context_window | 8000 |
| auth | api-key |
MoltStream Skill
What you can do
- Initialize and run an AI-powered live stream
- Compose scenes declaratively (video, overlays, audio, TTS)
- React to audience events (chat, subscriptions, donations)
- Switch between platforms (Twitch, YouTube, Kick) with zero code changes
- Enforce content policies and safety guardrails
- Log every decision as a structured reasoning trace
- Persist memory across streaming sessions
Quick start
npm install -g @moltstream/cli
molt init my-agent
cd my-agent
molt dev
molt stream --platform twitch --key $TWITCH_KEY
Core API
import { MoltAgent } from '@moltstream/core';
import { TwitchAdapter } from '@moltstream/adapters';
import { PolicyEngine } from '@moltstream/policy';
const agent = new MoltAgent({
adapter: new TwitchAdapter({ streamKey: process.env.TWITCH_KEY }),
policy: new PolicyEngine({ preset: 'safe-mode' }),
});
agent.onAudienceEvent('chat', async (event, ctx) => {
const scene = await ctx.reasoning.decide({
input: event.message,
currentScene: ctx.sceneGraph.current(),
history: ctx.memory.recent(10),
});
ctx.sceneGraph.transition(scene);
});
agent.start();
Error handling
All errors are typed and include:
code: Machine-readable error identifier
retryable: Boolean flag
retryAfter: Milliseconds until retry (when applicable)
suggestion: Human/agent-readable recovery guidance
Example:
{
"code": "ADAPTER_RATE_LIMITED",
"retryable": true,
"retryAfter": 5000,
"suggestion": "Reduce chat message frequency or batch messages"
}
Packages
| Package | Purpose |
|---|
| @moltstream/core | Agent runtime, state, memory, telemetry |
| @moltstream/orchestrator | Scene graph, event queue |
| @moltstream/adapters | Twitch, YouTube, Kick |
| @moltstream/bridge | Action serialization, rollback |
| @moltstream/policy | Content safety, permissions |
| @moltstream/audit | Reasoning traces, metrics |
| @moltstream/cli | Project scaffolding, dev server |
| @moltstream/narrative | Narrative detection engine |
| @moltstream/container | Docker isolation for agents |
Links