docs / first-time-setup

First-time setup

What to do right after install, in the order that matters most. Five small steps that turn a fresh install into something useful.

1. Connect Ollama

Soriku auto-detects a local Ollama install on http://localhost:11434. If you already have models pulled they show up in the model list. If not, pull at least one.

ollama pull qwen2.5-coder:7b

2. Run the capability benchmark

Out of the box Soriku uses bootstrap scores so it can route immediately. To get routing tuned to your specific hardware, run the local benchmark.

# quick mode: ~4 minutes per model
python -m benchmarks --quick
# full sweep: ~15 minutes per model
python -m benchmarks

Results land in benchmarks/results/capability_map.json. The router picks them up on the next prompt.

3. Add a remote provider (optional)

Bring your own keys for Claude, OpenAI, Mistral, Groq or Gemini if you want Soriku to consider them during routing. Keys stay in your local keystore unless you explicitly use hosted Soriku Cloud.

Add a remote provider →

4. Pick or build agent personas

Soriku ships with 15 built-in personas: code architect, code reviewer, backend developer, frontend developer, devops engineer, data analyst, researcher, creative writer, editor, and more. You can use them as-is or fork them into your own.

See the built-in personas →

5. Wire up your editor

Soriku speaks Model Context Protocol, so Cursor, VS Code and other MCP-aware tools can use it directly without an extra adapter.

Sanity check
Open the chat UI at http://localhost:8765, ask What model are you? and verify the response footer shows the model that answered and why it was picked. If routing looks off, see Routing not working.