Omakiten

The local source of truth for your AI agents.

Omakiten is a local checkpoint system for AI-assisted development. No cloud. No account. No telemetry. Just you, your projects, your agents, and a SQLite database on your machine.

21 languages | 100% local | Open source | Zero telemetry

The Problem

AI agents lose context between sessions. Different tools see different parts of the work. One agent rediscovers the bug another agent already fixed. You repeat the same explanation. They repeat the same mistake.

Omakiten gives your agents a shared local source of truth before they continue: tasks, dependencies, workflow state, decisions, errors, solutions, handoff notes, and plans in one place they can read and write through MCP.

How It Works

You keep working with whichever agent you prefer. Omakiten sits underneath them as the shared checkpoint for every project and every supported MCP harness.

flowchart TB
    user([You])

    subgraph projectA[Project A]
        claudeA[Claude Code]
        opencodeA[OpenCode]
    end

    subgraph projectB[Project B]
        claudeB[Claude Code]
    end

    user --> claudeA
    user --> opencodeA
    user --> claudeB

    claudeA <-->|MCP| omakiten
    opencodeA <-->|MCP| omakiten
    claudeB <-->|MCP| omakiten

    cli[okt CLI] <-->|same state| omakiten
    tui[okt tui] <-->|same state| omakiten

    omakiten[(Omakiten<br/>local source of truth)]
    db[(SQLite<br/>projects, tasks, dependencies<br/>context, errors, solutions<br/>workflow guards, plans)]

    omakiten <--> db

The important part is the direction of control: you interact with the agents, and the agents coordinate through Omakiten. Claude Code in Project A, Claude Code in Project B, and OpenCode in Project A can all read from the same local state without inventing their own memory.

What It Solves

The Agent Workflow

Omakiten ships canonical MCP prompts for the common development loop. The happy path is intentionally simple: discover the work, create the task, implement with guardrails, then review the diff.

flowchart LR
    imagine["/okt-imagine<br/>Discover the work"]
    create["/okt-create<br/>Shape and register the task"]
    implement["/okt-implement<br/>Execute with guardrails"]
    review["/okt-review<br/>Inspect the diff"]

    imagine --> create --> implement --> review

Daily Command Highlights

Natural language works too:

See the full MCP surface: 44 tools, 2 resources, and 11 prompts.

Install

Linux / macOS / WSL:

curl -fsSL https://raw.githubusercontent.com/This-Is-NPC/omakiten/master/install.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/This-Is-NPC/omakiten/master/install.ps1 | iex

The installer opens an interactive picker: CLI/TUI language, agent-output language, workflow preset, and MCP harnesses. For headless installs in CI, Dockerfiles, or dotfiles, provide environment variables:

OKT_CLI_LANG=en OKT_AGENT_LANG="English" OKT_PRESET=omakase OKT_HARNESSES=claude-code,opencode \
  bash <(curl -fsSL https://raw.githubusercontent.com/This-Is-NPC/omakiten/master/install.sh)

Supported MCP harnesses include claude-code, claude-desktop, codex, crush, github-copilot, and opencode.

Your First Project

# Register the current project
okt init --name MyProject --slug my-project

# Inspect the current state
okt list

# Open the visual interface
okt tui

Per-project configuration is supported. Put .omakiten/config/omakase.yaml in the repository root and Omakiten loads that workflow whenever you are inside the project tree. Guards, bucket rules, hooks, workflow snapshots, task state, and migration paths stay isolated per project.

Four Work Disciplines

Each preset is a process discipline, not an architecture prescription. Omakiten does not force DDD, Clean Architecture, or any framework-specific structure.

Every preset runs okt-imagine through 5W2H before work is filed, so you understand what you are building before planning code.

What Changes Day To Day

Before: You explain the project from scratch to every agent. They suggest changes that violate the team's process. The same error comes back because the previous fix lived only in a chat transcript.

After: Each agent reads the checkpoint before acting. Workflow rules are enforced at the shared state layer. Search crosses projects. Handoff notes survive sessions. You spend less time repeating context and more time building.

TUI

okt tui opens a terminal UI with three zones: Tasks (board, table, graph, plans), Stats (per-model benchmark, logs), and Settings (runtime info, entity browser). Task descriptions, comments, and entity files render as styled markdown; press M to toggle raw mode.

Outside a project, it opens a multi-project home. Pick a project, work on it, and the shell cds into that project's folder when you exit.

Read the full TUI guide

Update And Uninstall

okt update --check    # check without changing anything
okt update --yes      # download and atomically swap the binary

okt uninstall --yes             # remove binary and wrapper, keep data
okt uninstall --yes --purge     # remove everything, including data and config

Both commands fall back to an interactive picker when called without flags in a TTY.

Documentation

Master index: .docs/README.md

Project

• Changelog
• Contributing
• License

Install now  ·  Read the manifesto