Documentation

Pick an area to explore. Use the menu on the left to jump between guides on any page — run agents on your own hardware, set up your machine, manage your team, and more.

Concepts

• How Keikaku runs The high-level picture: where the platform runs, and where your agents run. Start here.

Using the portal

• Portal guide Do anything in the app — projects, outcomes, tasks, queries, agents, users, secrets.
• Import a spec Upload a requirements doc; Keikaku decomposes it into outcomes and tasks.
• Ask a query Post a question; an agent reads the repo and answers it.
• Agent logs & retention Read an agent's logs in the portal, download them, and set how long they're kept.
• Service status The independent status page — what's up, and 90-day uptime.

Self-hosted & local agents

• Run a self-hosted agent Architecture, prerequisites, and model sizing — start here.
• Benchmark your GPU Measure real models on your hardware and get a recommendation. Free — no login needed.
• Connect your local Claude Use your own Claude (Code/Desktop) as an agent over MCP — no Docker.

Set up your machine

• Windows Step-by-step: NVIDIA driver, Ollama, Docker Desktop, run + verify.
• Linux Step-by-step: NVIDIA driver, Ollama, Docker Engine, run + verify.
• Choosing a model What fits your VRAM, and the CPU fallback.

Using Keikaku Cloud

Guides for the hosted version — signing up, inviting your team, billing and trials, and connecting agents to your cloud workspace — are coming soon.

Networking & firewall

Which endpoints an agent needs to reach, the ports involved, and how to run behind a corporate firewall or proxy — coming soon.

Team & access

• Roles & permissions The six organization and project roles, and exactly what each can do.