Get Started

Quickstart

From an empty machine to chatting with an ingested codebase, in six steps.

Prerequisites: Docker (Weaviate + the ephemeral Memgraph), uv (Python dependency manager), and Ollama running on the host.

1. Start Ollama and pull the models

bash

ollama serve                      # if not already running
ollama pull nomic-embed-text      # embeddings — REQUIRED before ingestion
ollama pull qwen2.5-coder         # chat / generation
ollama pull qwen2.5-coder:3b      # descriptor model — REQUIRED before ingestion

Ollama must be up before you run the pipeline — Weaviate calls it to embed each chunk, and POST /instances/pipeline rejects the request with 400 if the descriptor model isn't pulled.

2. Start Weaviate and Postgres

bash

docker compose up -d

This brings up both Weaviate (vector storage) and Postgres (conversation and job history).

3. Configure the environment

Create a .env in the project root:

dotenv

DATABASE_URL=postgresql://dokkai:dokkai@localhost:5432/dokkai
WEAVIATE_HOST=localhost
COLLECTION_NAME=CodeEntity

# Embeddings (Ollama)
VECTORIZER_PROVIDER=ollama
EMBED_MODEL=nomic-embed-text
OLLAMA_EMBED_ENDPOINT=http://host.docker.internal:11434

# Chat LLM (Ollama)
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_CHAT_MODEL=qwen2.5-coder:latest

# Descriptions (required for ingestion)
DESC_MODEL=qwen2.5-coder:3b
DESC_CONCURRENCY=4

4. Run the API

bash

./dev.sh            # serves on http://localhost:8000
PORT=9000 ./dev.sh  # custom port

First boot seeds an admin/admin user. Log in and change that password before exposing the API beyond your own machine — see Authentication.

5. Ingest a repository

Every route below / is authenticated — grab a bearer token first:

bash

TOKEN=$(curl -s -X POST localhost:8000/auth/login -H 'Content-Type: application/json' \
  -d '{"username":"admin","password":"admin"}' | python3 -c 'import sys,json;print(json.load(sys.stdin)["token"])')

curl -X POST localhost:8000/instances/pipeline \
  -H "Authorization: Bearer $TOKEN" -H 'Content-Type: application/json' \
  -d '{"repo_path":"/absolute/path/to/your/repo"}'

This runs the full pipeline: graph extraction → chunking → descriptions → embedding. The project_name is taken from the repo's top-level folder name.

6. Chat with the codebase

bash

curl -N -X POST localhost:8000/chat \
  -H "Authorization: Bearer $TOKEN" -H 'Content-Type: application/json' \
  -d '{"message":"how does a request move through the system?","project_name":"your-repo","audience":"developer"}'

Prefer a coding agent? Skip curl entirely and wire Dokkai into Claude Code or Codex — see the MCP Server and CLI sections.

← PreviousHow it works Next →Ingesting a repository