How umair777/founderai works
Founder AI is an evidence-first assistant: a founder asks a question through the UI or an API/CLI entry point; the service plans which sources to query, fetches and normalizes context, detects updates, retrieves evidence, gates generation on evidence availability, validates citations, and returns an answer with evidence and trace data. The main orchestration and storage behavior is supported by sampled source. API and CLI are documented product interfaces but their implementations were not sampled, so their wiring is shown only at the service boundary.
Main entry points: service.py Assistant service · context_store.py Context store · connectors.py Gmail connector · llm.py LLM providers
Components 20
Question orchestration
- Assistant service
founder_ai/service.py - Query planner
founder_ai/planner.py - Context and trace models
founder_ai/models.py
Context sources
- Gmail connector
founder_ai/connectors.py - Calendar connector
founder_ai/connectors.py - Linear connector
founder_ai/connectors.py
Evidence and state
- Context store
founder_ai/context_store.py - SQLite database
founder_ai.db
Grounded answers
- LLM providers
founder_ai/llm.py - Citation validator
founder_ai/validation.py - Answer and trace
founder_ai/models.py
Other components
- Founder
- Google services
- Linear service
- SentenceTransformers
- OpenAI / OpenRouter
- TinyLlama
Connections 27
- Founder → Gradio UI: asks via
- Gradio UI → Assistant service: calls ask
- FastAPI → Assistant service: calls service
- CLI → Assistant service: calls service
- Assistant service → Query planner: plans query
- Query planner → Assistant service: returns plan
- Assistant service → Gmail connector: dispatches plan
- Assistant service → Calendar connector: dispatches plan
- Assistant service → Linear connector: dispatches plan
- Gmail connector → Google services: fetches mail
- Calendar connector → Google services: fetches events
- Linear connector → Linear service: fetches context
- Gmail connector → Assistant service: returns items
- Calendar connector → Assistant service: returns items
- Linear connector → Assistant service: returns items
- Assistant service → Context store: ingests items
- Assistant service → Context store: searches evidence
- Context store → SQLite database: reads and writes
- Context store → SentenceTransformers: encodes text
- Context store → Assistant service: returns evidence and updates
- Assistant service → LLM providers: generates from evidence
- LLM providers → OpenAI / OpenRouter: sends prompt
- LLM providers → TinyLlama: runs local model
- Assistant service → Citation validator: validates citations
- Assistant service → Context and trace models: builds trace and response
- Assistant service → Answer and trace: returns result
- Assistant service → Gradio UI: returns answer and evidence