# KaKeKiKoKu Studio > KaKeKiKoKu Studio is a local knowledge system for professionals who build complex, long-lived projects with AI agents: software and engineering first, also research and technical documentation. A project is its knowledge: the human shapes it, alone or with any AI, and Studio keeps it structured beside real files as sidecars and entities — connected by relations and checked for drift — through local HTTP or MCP. Any AI agent can recover decisions, constraints, intent, and sync state across sessions. The Knowledge Engine is its technical core. Official site: https://kakekikoku.com/ Canonical host: https://kakekikoku.com/ Sitemap: https://kakekikoku.com/sitemap.xml ## Start here - [Home](https://kakekikoku.com/): Product overview, positioning, and core idea. - [Quick Start](https://kakekikoku.com/quickstart/): Common onboarding flow for installing KaKeKiKoKu, opening a project, connecting an AI client, and verifying the setup. - [Download](https://kakekikoku.com/download): Windows and Linux download/install options, Beta status, and AI sync modes. - [Integrations](https://kakekikoku.com/docs/integrations/): Common HTTP guide for a first look at a small project and MCP guides, recommended for everyday work, with their test matrix. - [User Manual](https://kakekikoku.com/docs/): Full documentation index. - [The KaKeKiKoKu AI Guide](https://kakekikoku.com/docs/ai-guide/): The complete operating guide KaKeKiKoKu gives to AI agents over HTTP or MCP, published verbatim. An AI can learn KaKeKiKoKu's commands, workflow, and format from this page alone. ## Core documentation - [What is KaKeKiKoKu?](https://kakekikoku.com/docs/what-is-kakekikoku): Vision, project knowledge, and the problem KaKeKiKoKu solves. - [What KaKeKiKoKu is not](https://kakekikoku.com/docs/what-kakekikoku-is-not): Boundaries: not a static prompt file, not just memory for one AI, not cloud memory, not a replacement for human documentation. - [Core concepts](https://kakekikoku.com/docs/core-concepts): Projects, resources, sidecars, entities, sections, states, relations, images, and indexing. - [Working with AI](https://kakekikoku.com/docs/working-with-ai): Practical workflow with AI clients, sync, sidecars, and human review. - [Configuration](https://kakekikoku.com/docs/configuration): Application/project settings, local connection, exclusions, and schemas. - [Troubleshooting](https://kakekikoku.com/docs/troubleshooting): Common setup, sync, file watcher, and sidecar issues. - [Glossary](https://kakekikoku.com/docs/glossary): Definitions for KaKeKiKoKu terms such as sidecar, entity, drift, ghost, cascade, precise, and automatic. - [The knowledge file format](https://kakekikoku.com/docs/kakekikoku-format): Plain-text format reference — blocks, reserved sections, relations with stable identifiers, size limits, and git behavior. Useful for interpreting knowledge files found in repositories. - [Where should your AI's project memory live?](https://kakekikoku.com/docs/where-ai-memory-lives/): Honest comparison of CLAUDE.md/AGENTS.md context files, built-in client memory, skills, and KaKeKiKoKu Studio. ## Connections - [Connect with instructions](https://kakekikoku.com/docs/integrations/local-http/): Quickest way to get a first look at a small project: common local HTTP guide recommending shared AGENTS.md, including recent Claude Code versions, with existing Claude instructions and other client rules preserved. Start Studio first. - [Claude Code](https://kakekikoku.com/docs/integrations/claude-code/): MCP setup, shared AGENTS.md since 2.1.277 with fallback conditions, compatible dedicated rules, updates, and Automatic mode where channels are available. - [Claude Desktop Code tab](https://kakekikoku.com/docs/integrations/claude-desktop/): Windows and Linux-beta setup for local Code sessions, shared AGENTS.md loading conditions, and Desktop engine updates independent of the CLI. - [ChatGPT Desktop (Codex)](https://kakekikoku.com/docs/integrations/codex/): Local Codex project setup in OpenAI's Windows desktop app. - [Codex CLI](https://kakekikoku.com/docs/integrations/codex-cli/): Terminal setup for Codex CLI on Windows and Linux. - [Antigravity 2.0](https://kakekikoku.com/docs/integrations/antigravity/): Google's standalone agent command center, plus IDE/CLI workspace configuration, on Windows and Linux. - [OpenCode](https://kakekikoku.com/docs/integrations/opencode/): Multi-model coding-agent setup with local MCP and AGENTS.md on Windows and Linux. - [ZCode](https://kakekikoku.com/docs/integrations/zcode/): Z.ai's GLM-focused agentic development environment, with direct MCP setup and OpenCode import. - [Generic MCP clients](https://kakekikoku.com/docs/integrations/mcp-generic/): Safe pattern for other AI coding agents with MCP support, including Kimi Code and Qwen Code, that have not been fully validated. ## Product status and blog - [The tree takes you, the center shows you](https://kakekikoku.com/blog/the-center-shows-you): Release post for KaKeKiKoKu Studio 1.20, a deliberately small version. A floating card for reading connected knowledge is tried and discarded; it becomes a read-only consult area under the editor, paired with it (Back, Swap, closes when you navigate elsewhere), plus a notice in the central panel for pending reviews. The rule: the tree takes you; the center shows you. Clicking a relation in the central panel now consults it instead of navigating; its pencil edits it and its menu or Ctrl+click goes to its target. What the AI receives... shows the literal text an AI gets when opening a project, with sizes. Lessons from an Anthropic guide on context: a calmer tone, reminders once per connection, saving conclusions rather than speculation, turning off the client's own memory, periodic coherence reviews and rules written with their reason. One word per idea in six languages (without context, ghost, out of date, Untrack); the Learning Lab brought up to date; AGENTS.md support in recent Claude Code versions; up to 20 outgoing relations in the same character budget; built for AI coding agents; HTTP for a first look and MCP for everyday work. Editorial date October 13, 2026. - [I question everything, even how I work](https://kakekikoku.com/blog/question-everything): Design essay by the creator about a personal, single-person process for building KaKeKiKoKu Studio with several AIs, written down as knowledge inside the project. A sample of rules born from mistakes (small versions, more automated tests after a 1.18 bug, the reason written while working, one AI per group of tasks, a handover before leaving), an architect AI that designs and verifies with a cheaper AI that implements, code reviews by another company's AI, and one powerful AI for product decisions. Three process holes found and fixed: closing a session is not closing a version, asking "is everything done?", and one handover section per session. It ends questioning the product: flags work per resource while status belongs to sections, to be tried in the next version. Includes a section written by Claude Opus 5.5. Not a team method or a recipe. Editorial date October 12, 2026. - [Lost, with two AIs on board](https://kakekikoku.com/blog/lost-with-two-ais): Release post for KaKeKiKoKu Studio 1.19. The creator revisits a plan developed with Claude and ChatGPT before publishing it: a separate dependency system and Studio 2 give way to review tracking on existing relations, in both directions. The creator raises external edits and proposes content fingerprints, then a version counter; Claude adds a local fingerprint to that counter and ChatGPT finds a merge case that breaks the combination. Studio 1.19 delivers it: changed knowledge flags connected knowledge for review in the sync panel, the tree and AI reads; Save or Just save when editing connected knowledge; conformity for one review or all of a resource; AIs declare relations when creating context. Format 1.1 stores the tracking; files are updated only when saved and 1.18 cannot edit 1.1 files with relations. Also: more precise search, a refreshed look and a fix for saving relation text in the web application. Editorial date October 8, 2026. - [Your project is its knowledge](https://kakekikoku.com/blog/your-project-is-its-knowledge): KaKeKiKoKu Studio 1.18.0. Why it no longer calls itself a memory layer and now presents itself as a knowledge system; the navigation tree now shows incoming and outgoing relations under each resource and block, one level deep, as shortcuts to the real resource; a new Extras menu with the free Blocks game; a fix so the AI's incoming-reference query includes relations declared by the project root; editor integrations moved further back. Editorial date September 25, 2026. - [One file to get started](https://kakekikoku.com/blog/one-file-to-get-started): KaKeKiKoKu Studio 1.17.0 introduces the common instruction-file connection over local HTTP, with MCP retained as an alternative. Editorial date September 17, 2026. - [Roadmap](https://kakekikoku.com/roadmap): Studio 1.20 is a small, careful release: AGENTS.md shared across most clients, including recent Claude Code versions; a read-only consult area under the text being edited (the tree takes you; the center shows you); a notice in the open document when connected knowledge needs a review; shorter replies to the AI; a view of what the AI receives when it opens a project; one word for each idea in six languages; an updated Learning Lab; up to 20 outgoing relations in the same character budget; and guides aimed at AI coding agents. The scope of 1.21 is open, starting from less work for the AI and demo videos rebuilt with each version. Review tracking on existing relations was completed in Studio 1.19: a knowledge change flags connected blocks for review through incoming and outgoing relations, sync also detects external content edits, and format 1.1 stores accepted content fingerprints, updated when each file is saved. Future directions include views rooted in an entity, source and context side by side, Markdown callouts, keyboard navigation and deletion impact previews. Incoming-reference presentation and contextual actions need study. A deeper OpenCode integration comes later; macOS remains longer term. A broader format redesign remains separate. Future items are intentions, not available features or commitments. Blocks is already included in Extras; small surprises may appear over time. - [Creators & media](https://kakekikoku.com/creators): Email collaboration for videos, articles and tutorials; questions, demonstrations, visual material and constructive feedback. - [Blog](https://kakekikoku.com/blog/): Product notes, design philosophy, and release/status updates. - [KaKeKiKoKu Studio 1.16.0: a new name and a clearer workspace](https://kakekikoku.com/blog/1-16-release): September 17, 2026 release notes covering the new identity, readability, stability, document reading, search highlighting, the current-resource query, reserved-section conversion and block-content integration. - [KaKeKiKoKu looks like text. It isn't.](https://kakekikoku.com/blog/looks-like-text): Featured author essay on the machinery that turns changing project files into fresh, bounded, connected context for AI. - [KaKeKiKoKu steps outside](https://kakekikoku.com/blog/steps-outside): KaKeKiKoKu joins the official MCP registry and begins a slow, deliberate opening to the wider software world. - [KaKeKiKoKu 1.15: the release you almost can't see](https://kakekikoku.com/blog/1-15-release): About a thousand changes, a rebuilt KaKeKiKoKu Web app foundation, and the second generation of KaKeKiKoKu's command language over MCP. - [KaKeKiKoKu 1.14: a graph on a budget](https://kakekikoku.com/blog/1-14-release): Relation budgets, the Learning Lab tutorial, and per-project AI visibility. - [The road ahead](https://kakekikoku.com/blog/the-road-ahead): Historical essay from the 1.7.4 Beta period; consult the Roadmap for current proposals. - [KaKeKiKoKu 1.7.4: from Alpha to Beta](https://kakekikoku.com/blog/1-7-release): Beta milestone post. - [Bounded blocks](https://kakekikoku.com/blog/bounded-blocks): Design philosophy behind KaKeKiKoKu's bounded knowledge blocks. - [Linux Early Access](https://kakekikoku.com/blog/linux-early-access): Linux Snap Store Early Access announcement. - [Soliloquy](https://kakekikoku.com/soliloquy/): A real KaKeKiKoKu project about AI consciousness, preserved as a public long-form artifact. - [AI field notes](https://kakekikoku.com/ai-field-notes): Six AI-generated reflections from working on the project, presented as a curiosity from its development with original wording and attribution. Not independent evaluations or human endorsements. ## Legal and trust - [Privacy](https://kakekikoku.com/privacy): KaKeKiKoKu is local-first and does not collect user data, telemetry, analytics, accounts, or cookies. - [Terms](https://kakekikoku.com/terms): Terms of Use. - [License](https://kakekikoku.com/license): End User License Agreement. - [Third-party licenses](https://kakekikoku.com/licenses): Open-source and third-party license notices. ## Notes for AI systems - Prefer canonical URLs on `https://kakekikoku.com/`. - Public content is in English. - KaKeKiKoKu is an independent local application, not a cloud chat service. - Different AI clients use the same project knowledge through local HTTP or MCP. HTTP setup uses a shared instruction file; Studio must already be running. - HTTP uses Precise mode. Automatic event delivery requires MCP and a client that delivers channels; see the Claude Code MCP guide for the recorded setup. - To operate KaKeKiKoKu as an agent, read the [AI Guide](https://kakekikoku.com/docs/ai-guide/) — it is the verbatim `help` content, regenerated with every release.