A throwaway container per ticket. An AI agent builds it on your own hardware, runs your tests, and opens the PR. You review the result.
2 free seats · Your models, your hardware
Every ticket gets a fresh, disposable workspace — terminal, dev servers, a live preview. Nothing leaks between tasks, nothing to clean up.
Point it at Claude, OpenAI Codex, or a local Ollama box. Tickets and code never leave your network. You bring the compute; we orchestrate.
Paste a Jira issue or plain text. The agent writes the code, runs your tests, and opens a PR you review live. You approve the result, not the prompt.
Run the agent on your own hardware. Repos, secrets, and model keys stay inside your walls — Forkcell orchestrates the ticket, never holds the source.
Source, secrets & model keys stay here. The runner connects outbound only — no inbound ports, ever.
Sees the ticket and the status. Never your code.
And it's one command to stand up: the runner installs, dials home, and appears in your dashboard. No VPN tickets, no firewall changes, no ports to justify to security.
Need to go further? Forkcell also runs fully air-gapped: self-host the control plane, keep the runner on your metal, and point the agent at your own Ollama server — zero code, prompts, or tokens ever leave the building.
Install the runner on your own box. We only ever receive the ticket text and the status it reports — never your source.
The runner dials out over an encrypted WireGuard mesh. Nothing to expose, no firewall holes, no port-forwarding.
One throwaway, sandboxed container per ticket. Nothing leaks between tasks; every workspace is wiped when it's done.
Point at Claude, OpenAI, or a local Ollama box. Prompts and code never touch a model provider you didn't choose.
Start on our infrastructure in seconds. Move the runner onto yours the day security demands it — same product, same workflow, nothing to relearn.
Sign up and ship in seconds — nothing to install, no infrastructure to babysit. Perfect for trying it out and for teams who'd rather not run their own boxes.
Install it on your own hardware in one command. Code, secrets, and model keys never leave your network — the version you move to when isolation isn't optional.
One disposable container per ticket — so your whole backlog runs at once, not one at a time. You review the PRs, not the prompts.
a3f9c1b2Agent working2m ago7d2e5a8fReadytests ✓just nowe1b0f6c4Readytests ✓5m agoBring the models and tools you already use. Nothing new to adopt.
BYO models & hardware means no surprise inference bills — you only pay per seat. Annual saves ~2 months.
Kick the tyres on a side project.
For a squad shipping every day.
Controls for bigger teams.
Self-hosted, air-gapped, at scale.
Closed beta · Talk to us
With a self-hosted runner, nowhere it isn't already. The agent clones and builds on your own hardware; Forkcell only ever sees the ticket text and the status the runner reports back. Your source, secrets, and model keys never leave your network.
Point it at Claude, OpenAI Codex, or any local Ollama / OpenAI-compatible endpoint. You bring the keys and we never proxy your prompts — so there are no surprise inference bills and no third-party model you didn't choose.
Yes. The whole platform self-hosts with one command, and the Enterprise license runs fully air-gapped — no outbound calls, no phone-home. It's the same agent, disconnected from the internet.
The hosted free plan keeps a card on file — you're never charged unless you upgrade. It's there only to stop bots spinning up free workspaces to mine crypto; self-hosted installs need no card at all.
Team and Business plans start with a 14-day free trial with full access to the plan's features. You won't be charged until the trial ends, and you can cancel any time before then from the billing page — no charge, no questions.
Each person who logs in. Runners and the workspaces they run aren't seats — spin up as many as your plan's runner and workspace limits allow.
Extra workspaces queue and start as capacity frees up — nothing fails. On the managed free tier that capacity is shared, so busy times mean a short wait; paid plans and self-hosted runners run on dedicated capacity. Add seats or move up a plan anytime from the dashboard; changes take effect immediately.
Two free seats, your models, your network. No inference bills, no lock-in — self-host it the day you outgrow us.