Dynamous AI Coach
https://github.com/coleam00/dynamous-ai-coach- Category
- AI Agents
- Rank
- No. 2025Tools index
- Pricing
- Open Source
- Type
- AGENT
- Builder
- coleam00
- GitHub
- 55 stars
- Date
About
RAG AI agent that coaches users using Cole's content as its knowledge base.
What it does
It turns YouTube transcripts into searchable, timestamped material, then serves a conversational coach over a streaming web API. A separate command-line pipeline fetches, chunks, embeds, and stores transcripts. The coach searches excerpts first, can retrieve full transcripts, cites video moments, and retains conversation history.
Why it's ranked here
The project connects ingestion, semantic retrieval, citations, authentication, streaming, and conversation storage into a coherent reference implementation. Its provider options and timestamp preservation make the retrieval layer practical. However, the author explicitly says the code is not fully human vetted, and several operational choices still need scrutiny before production use.
What's good
Token-aware chunks preserve timestamps, so answers can point users to relevant moments instead of merely naming a video. Search and full-transcript retrieval are separate, which controls context use. The service also includes JWT verification, configurable rate limiting, generated conversation titles, stored message history, structured logging, strict type checking, and unit and integration test organization.
Tradeoffs
Setup requires Supabase tables, pgvector-backed storage, transcript fetching credentials, embedding credentials, and model configuration. The included coach is an authenticated backend API, not a complete user interface. Full transcripts can consume substantial context and are truncated beyond a configurable limit. Rate-limit checks allow requests through when their database operation fails. Cross-origin access is broadly configured.
How to use it well
Use it as a starting point for a team building a source-linked assistant over a YouTube channel or similar transcript collection. Run ingestion on a chosen channel, verify retrieval quality, then place a client application in front of the authenticated streaming endpoint. It does not supply that client, replace database operations work, or remove the need for production hardening.
Technical notes+
pyproject.toml defines a Python 3.11 package built with Hatchling, managed with uv, and checked with strict mypy, Ruff, and pytest. src/main.py exposes the FastAPI app from src/api/main.py, whose lifespan initializes embedding, Supabase, HTTP, and title-generation clients. The API provides /health and JWT-protected /api/pydantic-agent endpoints, streams responses, stores conversations and requests, and applies database-backed rate limiting. src/agent/agent.py registers semantic search and full-transcript tools with a Pydantic AI agent. src/agent/config.py accepts an OpenAI-compatible model base URL and sets transcript and request limits. src/api/db_utils.py implements persistence and fails open when the rate-limit query errors. src/utils/clients.py requires embedding and Supabase credentials. The provided src/api/main.py is truncated, so its complete streaming and persistence path cannot be confirmed from this text.
Observed
- License
- MIT
- Primary language
- Python
- Python requirement
- Python 3.11 or newer
- Packaging
- Hatchling build backend with a wheel containing the src package; dependencies install through uv sync
- Interfaces
- Command-line transcript ingestion pipeline and FastAPI HTTP service with Server-Sent Events streaming
- Storage and search
- Supabase with pgvector-backed semantic search
- Authentication
- Supabase JWT authentication on the coaching endpoint
- Test structure
- Repository structure lists unit tests for agent, retrieval tools, API, and pipeline, plus integration tests
Read from README.md, pyproject.toml, src/main.py, src/__init__.py, src/api/main.py, src/agent/deps.py, src/agent/agent.py, src/api/__init__.py, src/agent/config.py, src/api/db_utils.py, src/utils/clients.py, src/utils/logging.py, src/agent/__init__.py, src/tools/__init__.py, src/utils/__init__.py.
What it can do
Provide personalized coaching advice
User questions or coaching topics → Coaching recommendations based on Cole's methodology
Answer questions using Cole's knowledge base
Natural language queries → Relevant information and insights from Cole's content
Retrieve relevant coaching content
User's specific situation or challenge → Targeted content from Cole's knowledge base
Generate coaching action plans
User goals and current situation → Step-by-step coaching plans based on Cole's methods
Provide contextual coaching responses
User's coaching session history and current query → Personalized advice that builds on previous interactions
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.