Skip to content
FrameWorks Docs
Blog

Skipper — AI Video Consultant

Skipper is the platform’s AI video streaming consultant. Ask it about live streaming architecture, codecs, encoding, CDN behavior, protocols, deploying your own media servers, or troubleshoot a specific stream. Source-backed answers include citations, and unsourced or inferred sections are tagged separately.

Where to Find It

Section titled “Where to Find It”

In the dashboard — full-page chat at /skipper with conversation history, stream diagnostics, and authenticated management actions.

On the docs site — floating chat panel (toggle with Cmd/Ctrl+J). Docs mode uses read-only and diagnostic tools; it can explain or inspect, but won’t modify anything.

Via MCP — your AI agent can use ask_consultant for read/diagnostic consulting, or the video_consultant prompt for a guided persona, through the MCP server. Agents that need to change platform state should call the dedicated MCP tools directly, such as create_stream, start_dvr, or topup_balance.

What It Can Help With

Section titled “What It Can Help With”

Documentation questions — “How do I configure SRT encryption in MistServer?”, “What codec settings does OBS recommend for 1080p streaming?”

Troubleshooting — “My viewers are buffering on the European edge” triggers diagnostic tools that analyze rebuffering patterns, packet loss, buffer health, and CDN routing for your stream.

Platform operations — in authenticated dashboard chat, Skipper can use Gateway tools for stream, recording, billing, and diagnostic workflows. Through MCP ask_consultant, mutation tools are blocked; use the dedicated MCP tools directly when your agent intends to create, update, or delete resources.

Automated monitoring — where enabled, Skipper runs periodic health checks for eligible tenants with active streams and infrastructure, learns baseline patterns via Welford running averages, and flags anomalies. When something looks wrong, it runs a full diagnostic investigation and delivers a report with root cause and recommendations via configured email, WebSocket, or MCP channels.

General streaming knowledge — codecs, protocols, bitrate planning, latency tradeoffs. Answers from general knowledge are tagged as best_guess so you know when to verify.

Knowledge Base

Section titled “Knowledge Base”

Skipper’s answers are grounded in curated documentation:

DomainSources
FrameWorksPlatform docs, marketing site
MistServerFull documentation sitemap
FFmpegCore tools, codecs, formats, protocols, filters
OBS StudioKnowledge base articles — encoding, capture, streaming
SRTHaivision protocol docs — encryption, handshake, bonding, API
HLSRFC 8216 and draft RFC 8216bis
nginx-rtmpWiki — directives, control module, examples
EcosystemLivepeer, Daydream, WebRTC, DASH references

Operators can extend the knowledge base with custom sources. See the operator guide.

Confidence Tagging

Section titled “Confidence Tagging”

Every answer is tagged so you know how much to trust it:

  • verified — grounded in knowledge base documentation, with source URLs
  • sourced — backed by live web search results
  • best_guess — from the model’s general knowledge, not verified against a source
  • unknown — no supporting sources found

For Agent Developers

Section titled “For Agent Developers”

If you’re building an agent that connects via MCP:

ask_consultant — sends a question through the configured Skipper pipeline: pre-retrieval, optional query rewriting and HyDE, multi-round tool loop, and confidence tagging. Returns a structured response with answer, confidence, sources, and tools_used. In MCP spoke mode, ask_consultant can use read and diagnostic tools, but mutation tools are intentionally blocked.

Use the video_consultant prompt alongside ask_consultant to give your agent a streaming expert persona with knowledge of diagnostic tools and platform capabilities.

Operators can improve retrieval quality with cross-encoder reranking, query rewriting, and HyDE.

Learn More

Section titled “Learn More”