Composable, MCP-native AI toolkits for text, audio, image, and video. Build pipelines. Not monoliths.
reverberage is a hub-and-satellite ecosystem of independent Python packages. Each satellite does exactly one thing — audio in, text out; claim in, verdict out — and can be used standalone or composed into multimodal pipelines.
Built on Alibaba Cloud's Qwen model family. Running on 103M tokens of free quota.
| Satellite | Modality | What it does |
|---|---|---|
| hub | meta | Docs, roadmap, scaffold scripts, orchestration |
| n3rverberage | engine | Provider abstraction, MCP servers, A2A hub, memory |
| transcriber | audio → text | AI-powered audio/video transcription |
| verify | text → text | Claim verification engine |
| Phase | Satellite | Model | Status |
|---|---|---|---|
| Tier 1 | transform | qwen3-coder-plus | planned |
| Tier 1 | see | qwen3.7-plus | planned |
| Tier 1 | hear | qwen3.5-omni-plus | planned |
| Tier 1 | speak | cosyvoice-v3.5-plus | planned |
| Tier 2 | summarize | qwen3.6-flash | planned |
| Tier 2 | translate | qwen-mt-plus | planned |
| Tier 2 | extract | qwen3-coder-plus | planned |
| Tier 2 | reason | qwq-plus | planned |
See the full roadmap.
pip install rvrb-transcriber
rvrb-transcriber audio.mp3Each satellite follows the Satellite Protocol v2 — hatchling build, typer CLI, Pydantic v2 models, MCP-native, Python 3.11+.
See CONTRIBUTING.md.
Apache-2.0