CLI

dokkai srcs — Semantic Retrieval Codebase System

Plug dokkai's retrieval into the coding agent you already use, in one command.

SRCS is how you wire dokkai's retrieval into your workflow. With --model claude or --model codex it registers dokkai as an MCP server for Claude Code / Codex and launches it interactively. With --model ollama:<name> it runs a terminal chat or agentic loop against a local Ollama model over that same retrieval.

Which mode?

I want to…	Use
Keep using Claude Code / Codex, with dokkai as one more tool it can call	`dokkai srcs --model claude` (or `codex`)
Chat about a codebase from a terminal, no coding-agent CLI installed	`dokkai srcs --model ollama:<name> --project <name>`
Run a fully local agentic loop with no API server at all	`dokkai srcs --model ollama:<name> --project <name> --agent`

Recipes

bash

# Claude Code, with the dokkai MCP server (re-)registered
dokkai srcs --model claude

# Codex, same registration
dokkai srcs --model codex

# Local Ollama model, terminal chat loop over dokkai retrieval
dokkai srcs --model ollama:qwen2.5-coder:latest --project your-repo

# Local Ollama model, agentic MCP loop — no API server needed
dokkai srcs --model ollama:qwen2.5-coder:latest --project your-repo --agent

Flags

--project <name> — with ollama:<name>, the project to chat about; auto-detected when exactly one project is ingested.
--agent — with ollama:<name> only: agentic MCP loop instead of the /chat REPL. The CLI spawns the dokkai MCP server itself and navigates locally with MCP tools — requires only Weaviate + Ollama (no FastAPI).

Measured live with the API stopped: the canonical question answered via one search call, ≈679 tokens.

← Previousdokkai graph Next →dokkai watch