Skip to content

Tool catalog

What the platform can generate. Each tool is a queue worker; the agent orchestrates them as sub-tools over core-mcp.

The conversational agent. Model tiers route through litellm:

Tier Model family
standard Anthropic Claude
pro Anthropic Claude (larger)
flash Kimi

Corrects the old guide — earlier docs listed DeepSeek; the agent runs Claude (standard/pro) + Kimi (flash) today.

The agent calls leaf tools during a turn (e.g. lyric writing, song naming, album art, timed lyrics via transcription, and a wait-for-task primitive). Some sub-tools are hidden from the schema-driven UI via AGENT_SKIP_TOOLS.

Worker Tool(s) Provider / model Status
core-tool-sansaarm song_gen_v5 Sansaarm Deployed
core-tool-stitch stitch / studio composition, montage (beat-sync) ffmpeg-based Deployed
core-tool-media album art Replicate (flux-schnell) Deployed
core-tool-gpu stem separation (htdemucs), section analysis (allin1) on-box CPU inference
core-conversion format conversion ffmpeg / Pillow Deployed staging
core-tool-dna voice / DNA models In progress
core-tool-allin1 section analysis allin1 In progress
core-tool-copyright copyright checks In progress
core-tool-distribution distribution / publishing In progress