AI Support & Knowledge Ops Platform (KnowOps AI)

An AI-powered technical support backend for DevOps/SRE teams. It answers support questions from a runbook knowledge base with citations (RAG), and runs an agent that can search the knowledge base, create, check and triage incident tickets, and escalate to a human.

Progress: Phases 1–8 of 15 complete. Live status of every phase: AGENTS.md. New to the code? Start with CODE_TOUR.md.

Stack (free tier)

Layer	Choice
Hosting	Vercel (Hobby)
Backend	Python + FastAPI
Database	Neon Postgres + pgvector
LLM + embeddings	Google Gemini (gemini-embedding-001, 768 dims)
Auth	Self-rolled JWT
CRM	Mocked in Postgres
Automation	Vercel Cron + local n8n

Quick start

1. python -m venv .venv and activate it
2. pip install -r requirements.txt
3. Copy .env.example to .env and fill in real values (never commit .env)
4. Run scripts/init_db.sql against your Neon database (SQL editor or psql)
5. python scripts/ingest.py to load data/docs/ into the knowledge base
6. uvicorn app.main:app --reload, then open /health, /docs (Swagger) or / (web console)

Tests: pytest

Project layout

app/        FastAPI app: routes, config, DB, auth, RAG, LLM, agent, CRM
api/        Vercel entrypoint (sprint-2 branch)
scripts/    DB schema, ingestion
data/docs/  runbook knowledge base
static/     web console
n8n/        n8n workflow export
tests/      pytest suite + RAG eval

Deployment

Deployed on Vercel through the project knowops-ai-, which is Git-connected to this repo, so pushes deploy automatically. Environment variables are set in the Vercel dashboard. Details in AGENTS.md.

Understanding the code

• FLOW.md — entry points, startup order, and which function calls which for every route
• DECISION.md — every meaningful decision and the reason behind it (why this library, why this tradeoff)

Team workflow

The build follows a shared 15-phase build guide (ask a teammate for the link). The rules for AI agents live in AGENTS.md:

• Agents log decisions in DECISION.md and keep FLOW.md up to date with every code change.
• Before a code change is accepted, the agent quizzes you with 3 questions about it. You have to get all 3 right for the change to count as accepted.
• After finishing a phase, run /phase-done <phase> in Antigravity or Claude Code. It updates the status table, phase log and known issues in AGENTS.md and the progress line above, so everyone sees the same status after a pull.