Claude Cowork is Anthropic\'s agentic layer inside Claude Desktop: you grant folder access, Claude works on it. Useful — but inference still happens in the cloud. Kintari inverts that: the same agentic workflow, but in the encrypted vault, with model choice per task. Your data is yours.
Cowork is a well-thought-through agentic product: explicit folder grants, clear read/write permissions, multi-step autonomous tasks, scheduled routines. Anthropic has taken its time and built a UX that doesn\'t leave the sandboxing question to the user.
Anyone looking to automate routine workflows — recurring research, table analyses, small operations chores — gets a polished tool with Cowork without needing to touch the terminal. Connectors to Slack, Google Workspace, Microsoft 365, and apps like QuickBooks, HubSpot, or DocuSign are available (Cowork plus "Claude for Small Business").
What Cowork doesn\'t solve: whether your files are allowed in the cloud at all. The agent operates on the local folder, but inference and tool decisions still happen on Anthropic\'s servers.
In Cowork, the model unavoidably runs in the cloud — otherwise it cannot read your files at all. In Kintari, it\'s a choice per task. Pick a local model and the agent sees your content, the box at home computes the answer, nothing leaves the house.
For professions under confidentiality obligation, that is often the only viable path. Cowork has a clean sandbox architecture, but the model still sees the content — just in cloud memory rather than on disk.
Cowork operates on a regular folder of your choosing — fair enough, but every file inside is open to the agent. Kintari has the encrypted workspace as the central data home: knowledge index, configuration, audit log live in an encrypted file. The agent sees what the workspace has indexed — not arbitrary files on disk.
Practical difference: a stolen laptop reveals the contents of the Cowork sandbox folder. With Kintari, it reveals nothing as long as the vault passphrase isn\'t with it.
Cowork is Anthropic\'s agent — it does what it does excellently, but it is one agent. Kintari is the hub: the built-in Kintari agent plus bridges to Claude Code, Gemini CLI, OpenCode, Aider, and more. One session, one approval queue — regardless of which agent is doing the work.
For research use the built-in agent, for code refactors Claude Code, for GCP topics the Gemini CLI — and all of them share the same workspace. In Cowork you are locked into Anthropic\'s single agent.
| Axis | Claude Cowork | Kintari |
|---|---|---|
| Inference | Anthropic cloud | cloud or local, per task |
| Data location | sandbox folder (cleartext) | encrypted vault |
| Model choice | Anthropic only | Anthropic + Google + Ollama + OpenAI format |
| Platforms | macOS, Windows | macOS, Windows, Linux, mobile (beta) |
| Multi-agent | Claude only | built-in agent + Claude Code + Gemini CLI + … |
| Scheduled tasks | yes (built-in) | yes (Routines with Skills) |
| Cross-app connectors | curated suite (Slack, QuickBooks …) | 1,500+ via MCP & Composio |
| Audit log | in Anthropic\'s infrastructure | on your disk |
| Price | included in Claude subscription | €99 once · cloud tokens separate |
Cowork runs on the strongest available model (Claude Opus 4.7) and offers a curated connector suite that fits Anthropic\'s target audience — office/operations/finance workflows in cloud-OK organisations — very precisely. If you belong there, Cowork is a better fit than Kintari.
The polished UX and the ability to start agentic tasks without configuring local models are real strengths we don\'t reproduce 1:1.
Kintari is the right choice when your content may not leave the machine — or when you want the same agentic working style, but with real model choice and an encrypted workspace as home.
If the Cowork idea resonates but the cloud requirement is in the way, a 30-minute call usually clarifies whether Kintari hits the architecture you need.