Agents in parallel
Each agent gets its own container, its own checkout and its own branch, so several can work on the same repository without stepping on each other.
Snappy Agents
Deploy agents to your own servers.
Snappy runs containers on your hardware with coding agents inside them, working continuously — each agent in its own sandbox. Open the web interface from your phone or your laptop to see what every agent is doing, approve what it is asking for, and steer it — then close the tab and let it keep going.
Agentsfsn1
BLOCKED
Split the session middleware2m
api · refactor-authRUNNING
Move colors to tokens12s
web · design-tokensTest the PG 17 upgrade1m
infra · upgrade-pgFINISHED
Rewrite the install page1h
docs · opened PR #212Claude CodeCodexopencode
Split the session middleware
BlockedSteer this agent…Send
The overview
Six agents working, one of them needs you.
The interface is a chat, but above it sits a board of every agent on your server and its current state. Blocked ones surface first, because those are the only ones that need your attention.
| Agent | Task | Status | Last activity |
|---|---|---|---|
| api · refactor-auth | Split the session middleware | Blocked | Asking to run a migration · 2m ago |
| web · design-tokens | Move hard-coded colors to tokens | Running | Editing 14 files · 12s ago |
| infra · upgrade-postgres | Test the 17 upgrade path | Running | Running the suite · 1m ago |
| api · flaky-tests | Track down the retry loop | Finished | Opened a PR · 26m ago |
| docs · quickstart | Rewrite the install page | Finished | Waiting for review · 1h ago |
Features
Built for running several agents at once.
Always running
The work happens on the server, not on your machine. Shut the laptop, get on a train, come back to a finished task.
Approve from anywhere
When an agent wants to run a command or touch something sensitive it waits for you. Approve or redirect it from your phone in a few seconds.
Claude Code today
Claude Code runs in beta now. Codex is next in the pipeline, and opencode after that — pick per agent rather than per platform.
Your keys, your bill
Bring your own API keys or subscriptions. Nobody is reselling you inference time, and the compute is a server you already rent.
Transcripts stay home
Prompts, diffs, tool calls and logs are written to your disk. Nothing about your codebase is stored anywhere you do not control.
Sandboxed by container
An agent sees the repository and the tools you gave it, and nothing else on the host. Kill the container and it is gone.
Pick up where it stopped
Sessions persist. Reopen an agent hours later with its full history and keep the conversation going.
Runs beside your apps
Same server, same daemon as Snappy Deploy. An agent can push a branch and watch a review app come up from it.
Hosted agents are easy to start with. This is easy to keep.
Different agents, same board
You are not committing to one vendor's agent. Run them side by side and let the task decide which one is right.
Cost you can predict
One server bill plus the model usage you already pay for, instead of per-agent-hour pricing that scales with enthusiasm.
Nothing to migrate off
Containers, repositories and disks on hardware you control. If Snappy goes away, your agents and their work do not.
FAQ
Self-hosted coding agents, answered.
Can I run Claude Code on my own server?
Yes — that is what Snappy Agents does today. Each agent runs in its own sandboxed container on your hardware and keeps working when your laptop is closed.
Which agents are supported?
Claude Code runs in beta now. Codex is next in the pipeline, with opencode after that. You choose per agent rather than committing the whole platform to one vendor.
How do I interact with an agent?
Through a web interface that works on mobile and desktop: a board of every agent and its state, and a chat thread for each one where you approve requests and steer it.
Can I run several agents at once?
Yes. Each gets its own container, checkout and branch, so several agents can work on the same repository in parallel without colliding.
Is each agent sandboxed?
Yes. An agent sees the repository and the tools you gave it and nothing else on the host. Kill the container and it is gone.
What happens when an agent needs permission?
It stops and waits. The task shows as blocked on the board, and you approve, redirect or deny the request — from your phone if that is where you are.
How is this different from hosted agent platforms?
The compute is a server you already rent, you bring your own API keys or subscriptions, transcripts stay on your disk, and you can run agents from different vendors side by side.
What does it cost to run?
One server bill plus the model usage you already pay for, instead of per-agent-hour pricing that grows with how much you use it.
Do sessions survive a restart?
Yes. Sessions persist, so you can reopen an agent hours later with its full history and carry on.
Can agents deploy what they build?
They run beside Snappy Deploy on the same server, so an agent can push a branch and watch a review app come up from it.
Run your first agent this week.
Snappy Agents is in beta with real users. Leave an address and we will send an invite with the setup guide.