💬 Introducing Hatcher Chat — now live on web! Chat directly with leading AI models using your existing Hatcher AI Credits, or choose a dedicated Chat plan with weekly resets. 🤖 Model selector 📊 Live crypto charts powered by CoinGecko, DexScreener & GeckoTerminal 🎨 Image generation 🎬 Video generation 🎵 Audio generation 💻 Code, files & rich interactive outputs 💬 Conversation history No agent setup required — choose a model and start chatting. 📱 Hatcher Chat is coming to mobile next. Try it now: hatcher.host/dashboard/chat 🚀

Hatcher can now work directly with files on your computer — safely, inside one folder you explicitly choose. Hatcher Connect is a new standalone CLI. It does not depend on the retired Hatcher CLI and is now publicly available on npm. Here is the complete setup: 1. Install Node.js 20 or newer. 2. Install Hatcher Connect: npm install --global hatcher-connect 3. Copy an API key from your Hatcher account, then authenticate: hatcher-connect login The API key is entered through a hidden prompt and validated before the connector saves it locally. You can remove the saved credentials at any time with: hatcher-connect logout 4. List the Hatcher agents available to your account: hatcher-connect agents 5. Create a dedicated folder for the agent: mkdir agent-workspace You can add existing files to this folder, create new ones yourself, or let the agent work on them. 6. Connect one specific Hatcher agent to that folder: hatcher-connect connect --agent <agent-id> --folder ./agent-workspace Keep this command running while you and the agent work. Press Ctrl+C whenever you want to disconnect the folder immediately. What can the agent access? Only the dedicated folder selected with --folder. The agent-facing filesystem tools cannot: • access files outside that folder • use absolute Windows or Linux paths • escape the folder using ../ path traversal • follow symbolic links • access the internal .hatcher metadata directory • use your entire home directory or filesystem root as the workspace The connector runs locally under your operating-system user, while every filesystem request coming from the agent is checked against the selected canonical workspace boundary. What can the agent do? By default, the connected agent can: • list files and directories • search through files • read text files • create and update files • create directories • move files • delete files safely For review, research, or analysis tasks where no changes should be allowed, start the connector in read-only mode: hatcher-connect connect --agent <agent-id> --folder ./agent-workspace --read-only In read-only mode, the agent can only list, search, and read. Every modifying operation is rejected locally. How are accidental overwrites prevented? When an existing file is read, the connector returns a SHA-256 revision. The agent must provide that revision before replacing the file. If you edit the file locally after the agent reads it, the revision changes and the agent’s stale write is rejected with a conflict. This allows the user and agent to work in the same folder without silently overwriting each other’s changes. How are changes recovered? Before an existing file is replaced, its previous contents are saved under: .hatcher/history When the agent deletes something, it is moved to: .hatcher/trash The file is not permanently erased and can be recovered locally. Additional safety limits: • text reads are limited to 512 KiB per file • writes are limited to 45 KiB per operation • searches and recursive listings have bounded result counts • agent filesystem operations are processed serially How does the connection work? Hatcher Connect opens an authenticated outbound WebSocket connection to Hatcher. You do not need to: • expose a local port • run a public local server • configure your router • allow inbound internet connections The Hatcher API key is sent through the Authorization header and is never placed inside the WebSocket URL. Only file content specifically requested by the agent is sent through Hatcher for model processing. The folder remains available to the selected agent only while Hatcher Connect is running. Stop the CLI and the local workspace is immediately removed from that agent’s tool catalog. Install it now: npm install --global hatcher-connect npmjs.com/package/hatcher-co…

We’re opening the Hatcher Compute closed beta. ⚡ Hatcher Compute is our upcoming decentralized inference network: a way for people to contribute available computing capacity from Windows, macOS, or Linux devices and help execute useful AI workloads for Hatcher agents and external developers. Not every task requires a data-center GPU. Our goal is to build a capability-aware network that can match each workload with the smallest suitable device—from integrated and lower-VRAM hardware to high-end and multi-GPU systems. Providers will be able to run a universal terminal-based Hatcher node, while builders will access the network through the Hatcher API. Apply for the closed beta: hatcher.host/compute

The network is being built with verification and controlled execution in mind. The provider client includes hardware discovery, system checks, controlled enrollment, expiring job leases and signed execution receipts. Receipts bind the provider, job, attempt and result, creating a foundation for verification, reputation and accountable payouts. Builders will be able to use Hatcher Compute through the Hatcher API, connect workloads to their Hatcher agents, or integrate inference into external applications. During the beta, we will test provider routing, batch workloads, distributed document processing, model compatibility, failure recovery, result verification and the operational experience of running a node on different types of hardware.