Your own fleet of agents, interconnected across all your tasks, solo or as a team.
Stop throttling your delivery.
Solo or as a team, on one project or ten, one secure control tower available 24/7 to put your agents to work.
You run a fleet of agents in parallel: your output is no longer limited by the number of runs one person can monitor. Solo, or across a whole team.
Need 40 GB of RAM for a build your laptop cannot touch? Spin it up. Scale CPU and memory past anything local.
The same cloud dev environment on Mac, Windows or Linux. No more "it works on my machine", no more configuration drift across the team.
Claude Code, Codex, Mistral... Choose the best agent for each task, with no lock-in, on the subscriptions you already have.
Agents run in the cloud, not on a device. Start a run, walk away, pick it up later from any device. It never stopped.
A built-in MCP connects every project and agent in your account. An orchestrator can reach across projects, create agents and collect their results: the whole fleet moves as one, with nothing to install or host.
Drive a single orchestrator: it creates child agents, assigns them tasks, messages them and collects their results. Large jobs are distributed across the whole fleet in parallel.
The internal MCP connects every project in your account, so an agent can coordinate work across several repositories, not just the one where it started.
The MCP is wired into every agent when its runner starts. No server to run or glue code to maintain.
Each task gets its own git worktree, branch and chat history, so you can launch a dozen at once.
Watch your agents think live and steer them without interrupting the run.
Automatic PR status, multi-account support, diffs and commits in-app.
Linear, a built-in browser or your own MCP servers, connected in a few clicks.
Run a prompt on a schedule, recurring checks and reports.
Trigger agent runs from events.
Your work lives on the web and is there as soon as you open a browser. Nothing to install or sync. Use a computer, tablet or phone and pick up the same workspaces.
Open a browser and your projects are already there. No local setup or toolchain.
A complete cloud dev machine in the tab: terminal, files, diffs and live preview. Everything you would open locally, with nothing on your machine.
Computer, tablet and phone all access the same workspaces in real time. Start a run on one and pick it up on another.
Live CPU, memory and disk for your active runners. See what every running agent consumes and catch a heavy build before it stalls.
In API mode you bring your own provider keys and pay per token at cost, so Hecaton shows you exactly where the spend goes: by model, by agent, by workspace. No markup, no surprise at the end of the month.
On a subscription plan you do not pay per token, you work within rolling usage windows. Hecaton tracks every window across your plans, how much is left and when it resets, so no agent stalls on a limit you did not see coming.
PDFs, Markdown, images, screenshots: read everything directly in Hecaton, next to the conversation and the changes. Review a report, proofread a spec or show a screenshot with no download, no friction and no lost context.
Right below the slide you are viewing, prompt the agent to refine it or build the next one. Drop in a screenshot, a chart or a snippet of a document and it becomes a slide. Your deck takes shape one prompt at a time.
Every project runs live at an address you can open and share in one click. Send the link to a client or a teammate, open it on a computer or on a phone: everyone sees the real product. Nothing to install, no deployment to wait for.
Ask for a QA pass: an agent walks through your app the way a customer would, screen by screen. It brings every screenshot back into the conversation, then tells you clearly what works and what gets stuck. You spot problems before your customers do.
Check the new checkout and show me how it looks.
Checkout works. The total adds up and the customer reaches a clear confirmation.
Hosted on our own infrastructure in France. Your data and code stay in the EU, never mixed with another customer's. No infrastructure to manage.
Every environment is fully isolated. Your code and data are never shared or mixed with another customer's.
Connect provider API keys or supported subscriptions. Credentials are encrypted at rest and sent only to the isolated runner that needs them.
Nothing to install or configure. Free compute is enabled as soon as you sign up. Your first explicit agent action starts an isolated runner on demand.
Every account uses the same free plan. An administrator may occasionally gift additional RAM.
A real cloud environment for running your first agents.
LLM usage is not included. Connect your own provider API keys or a supported provider subscription.
Hecaton is a managed cloud that runs AI coding agents in parallel. Each task has its own git worktree and branch, and compute runs in an isolated gVisor runner started by an explicit action. The app is available from any device and agents keep running after you close your laptop.
Locally, agents compete for one machine. Hecaton starts isolated cloud runners for the tasks you launch, lets you steer them live and keeps them running after you close your laptop.
Hecaton is model-agnostic. It runs Claude Code (Anthropic), Codex (OpenAI) and more side by side, so you can pick the best agent for each task, with no lock-in.
You can connect your own API keys, or use supported provider subscriptions such as ChatGPT or Claude. Credentials are encrypted at rest and staged only into the runner that needs them. Hecaton does not add a markup to provider usage.
On infrastructure in France (Scaleway Kubernetes, fr-par). Product data is user-scoped in the managed database; code executes in isolated runners and durable workspace snapshots are scoped to their owner.
Each active workspace runs in its own non-root gVisor Pod with restrictive NetworkPolicies and no product database credential. Product records are scoped server-side to the authenticated user, and snapshots use owner-scoped object paths.
Yes. app.hecaton.io works from desktop or phone with no VPN. Agents run in the cloud, so you can close the laptop and continue later from another device.
Hecaton is currently free and includes 2 GiB of concurrent RAM with no card required. An administrator can gift additional RAM. Each runner can use up to 28 GiB. Provider usage remains billed through the credentials or provider subscription you connect.
Hecaton is a commercial managed service, not an open-source project. The coding agents it runs, such as Claude Code or Codex, are third-party tools you bring with your own keys. You can connect your own MCP servers and skills. Your code and full git history stay yours to export at any time, so there is no lock-in.
No. Hecaton does not use your code or prompts to train models. Prompts are sent to the provider you connect under that provider's terms, from the isolated runner executing your task.
Yes. Hecaton supports two modes. In API mode you bring your own provider keys and pay per token at cost. In subscription mode you connect an existing plan and work within its usage windows. The app tracks usage, cost and remaining limits for both, broken down by model and by agent.
As soon as your account is created, the app is ready. Nothing is prewarmed by simply opening a workspace: your first explicit message or terminal action starts the isolated runner it needs.
