Your work agent. Not on your laptop.

No index. When you ask, it searches your tools and reads only what the answer needs. What it reads is cached, encrypted, and purged after 30 days unused. Chats stay until you delete them.

1,886 tools, from files, mail and calendar, and chat to CRM, finance, legal, design and media, and travel and property. Add any remote MCP server, or your own computers through the open-source Local Agent, for the rest. The full list is on the integrations page.

No, it runs on our servers, so there's nothing to install and it keeps going when your laptop is closed. If you want it to work with files on your computer, pair it with the open-source Local Agent: it reads only the folders you share, and changes files or runs commands only in folders you allow on the computer, each one approved.

Yes. Ask it to brief you every weekday at 8, or add a schedule in Settings. Each run is a new chat as you, any change it wants still waits for your approval, and you're told when one does.

Yes. It runs code in a sandbox on our servers, with no network and only the chat's files, to calculate and to build decks, documents, and spreadsheets on your company's templates. On your own computer, it can run commands like your tests in folders you allow, each one approved.

It reads instructions you write for yourself, your projects, and your organization, and saves short facts about you that you can see and forget in Settings.

Yes, for teams that want an agent at work without installing, hosting, and securing one themselves. It runs on our servers in the EU, each person's tools stay their own, and every change to your tools waits for approval. OpenClaw fits people who want full control of an open-source agent on their own machine.

Yes, on most connectors. Connect a workspace per client, or a work and a personal account, and name each one: the assistant knows which is which and searches each one separately. Each account keeps its own access level, so one can stay at Read while another can make changes, each with your approval.

Yes. Put a chat in a project and your teammates on it read the answers live, as they're written, and can ask the next question with their own tools. Guests read only the chats you show them. Each person connects their own accounts, so what your tools return stays with you: teammates see the answer, never your access.

Yes. Add any remote HTTPS server in Settings. It starts at Read, offering only its read-only tools. What the server does with your requests is up to whoever runs it.

Only if you let it. Every tool starts at Read. Switch one to Read and change, and each change waits for your approval, shown exactly as it will run. Deleting and sharing always ask. Admins can turn changes off for everyone.

The app runs on Hetzner in Germany. The models run in the EU: Gemini on Google Cloud, GPT on Azure's EU Data Zone, Claude on AWS Bedrock, and the free model on Hetzner. Or self-host it.

Yes, per chat: Gemini, GPT, Claude, or the free Qwen model, all in the EU. Self-hosted installs can use any provider, or local models through Ollama or GPUStack.

Yes, if you want it to. Web search is off by default: an admin allows it for the organization, then you turn it on in a chat. The searches go to Google for Gemini or Bing for GPT, outside the EU, and each costs 1.4 credits.

No. We don't train models on your data, and the hosted app uses model providers whose business terms say customer data is not used to train their models.

Team is €25, €125, or €250 per person a month, plus VAT, with 900, 4,500, or 9,000 credits each, shared across the team, and a year costs ten months. Guests are free. Each organization pays for itself, so company chats never spend anyone's personal credits. Team billing opens soon; until then everyone gets 30 free answers a day.

Yes, on infrastructure you control, with local models if you like. Email hello@chatwithwork.com with what you know.

The CASA Tier 2 assessment is complete, Google OAuth App Verification is approved for the read-only Drive, Docs, Sheets, and Slides scopes, and the SOC 2 Type I audit process is underway. The security page explains the main controls in plain language.