Start free. Pay for orchestration, not tokens.
Every plan is BYOK: your LLM key, your provider contract, your data terms. The subscription covers the run infrastructure, the isolation, and the collaboration features.
Free
Start hereEUR 0
Try the loop on one real repository.
- 1 repository
- 1 seat
- 1 concurrent run
- 100 runs per month (hard stop)
- Runs up to 15 minutes
- gVisor sandbox runtime
- 30 days run history
- All forges, including self-managed
- BYOK, all supported providers
Solo
EUR 29/month
One engineer who wants the agent on real work.
- 300 runs per month (fair use)
- Runs up to 30 minutes
- Codebase context (RAG)
- 90 days run history
- 1 repository and 1 seat, same as Free
- 1 concurrent run
- gVisor sandbox runtime
Team
EUR 99/month
A team that wants parallel runs and stronger isolation.
- 10 repositories
- 10 seats
- 3 concurrent runs
- 1,500 runs per month (fair use)
- Runs up to 60 minutes
- Codebase context (RAG)
- Kata VM sandbox runtime
- 1 year run history
- Need more repos or seats? See Business
Business
EUR 299/month
Several teams or one org on a shared plan.
- 30 repositories
- 30 seats
- 5 concurrent runs
- 5,000 runs per month (fair use)
- Runs up to 120 minutes
- Codebase context (RAG)
- Kata VM sandbox runtime
- 1 year run history
Self-serve billing is on its way; until it lands, one email sets you up. The listed price is the price: emailing to activate does not start a negotiation.
Enterprise
More repositories, seats, and parallel runs than Business offers, sized to your organization, with runs up to 4 hours. Hosting is a choice at this tier: run by us on EU infrastructure, or deployed on your own hardware by agreement. No self-hosted deployment is live today; we scope the first one together rather than promise an installer. We answer your vendor security questionnaire and walk your security team through the architecture.
typically from
EUR 599/month
Prices in EUR, excluding VAT. Cancel anytime from the billing portal; your plan stays active until the end of the paid period. Plan limits follow our server capacity: as it grows and higher limits prove sustainable, we raise them.
The subscription never includes LLM tokens. Model calls spend your own key and land on your provider's bill; the in-app cost dashboard meters planning and indexing calls only, not the agent's in-sandbox calls. A dedicated provider key for FragonForge keeps that spend easy to see, and a spending limit in the provider console caps it.
What you pay us for, and what you don't
You connect your own API key for Anthropic, OpenAI, Google, OpenRouter, Groq, DeepSeek, or any OpenAI-compatible endpoint. The model bills you directly. We never resell tokens, so we never mark them up. The subscription pays for the part you would otherwise build yourself:
Orchestration
Webhooks, queueing, retries, the issue-to-MR loop, and the project orchestrator that turns a goal into a pipeline of issues.
Isolation
Per-run Kubernetes jobs, hardened runtimes, default-deny networking, and a credential-free sandbox. This is most of the product.
Codebase context
AST-aware indexing of your repository so the agent retrieves relevant code instead of guessing (paid plans).
Collaboration
Seats, roles, review gates, run history, and a cost dashboard for planning and indexing spend. The agent's in-sandbox model calls are not metered there; they appear only on your provider's bill, which stays the source of truth.
Quality gate
33 static analysis tools on every change, a baseline-aware gate that flags only new findings, an A to F repository rating over time, and optional auto-fix merge requests.
Frequently asked questions
Does it work with plain gitlab.com or github.com?
Yes, out of the box. Connect a personal or group access token and you are set. A custom API endpoint is only needed for self-managed instances.
What counts as a run?
One run is one agent execution: a labeled issue, a retriggered task, or one step of an orchestrator pipeline. Static analysis on its own does not consume runs. A run that fails early still counts, because the sandbox was provisioned; a run that never starts (for example, over quota) does not.
What counts as a seat?
Every user account in your FragonForge organization uses a seat, regardless of role: someone who only reviews the agent's merge requests or reads run history counts the same as someone who triggers runs. Free and Solo include 1 seat, Team 10, Business 30. The limit is enforced when a user is added or accepts an invitation, so you cannot overrun it by accident.
Will the agent bury my team in merge requests to review?
Only if you ask it to: runs happen when you label an issue, trigger one from the API or CLI, or start a scan, so the volume is yours to control. The quality gate pre-checks every change against 33 tools, so review time goes into substance rather than lint. And review is deliberately the bottleneck, because FragonForge never merges on its own. Start with one repository and widen as your review capacity allows.
What happens when I hit the monthly limit?
On Free, runs stop. The limit is a hard stop at 100; new runs are rejected with a clear error until the month resets or you upgrade. On Solo, Team, and Business, the limit is fair use: runs beyond the quota are parked and requeued instead of rejected. Parked runs start automatically when the new month begins (quotas reset per calendar month, UTC); nothing is silently dropped, and there are no overage charges.
Will the plan limits stay this tight?
Not forever, but they are honestly gated. Every run executes in an isolated sandbox on our own EU hardware, so limits follow real server capacity, not an autoscaling abstraction. As we add capacity and can see that higher limits hold up economically, we will raise them; we would rather run a service that is still here next year than a generous one that is not. No date promised, because we do not publish dates we cannot keep.
Do you charge for LLM tokens?
No. Your key, your provider, your bill. Every model call, planning and indexing included, spends your own key. One honest caveat about visibility: the in-app cost dashboard currently meters the planning and indexing calls, not the agent's own model calls inside the sandbox (those are proxied for isolation and pass through unmetered). Set a spending limit in your provider's console and treat its billing page as the source of truth. A dedicated key just for FragonForge keeps that page easy to read.
Can I use my self-managed GitLab on the Free plan?
Yes. Forge support is not a paid feature. Self-managed GitLab, GitHub Enterprise, Gitea, and Forgejo endpoints work on every plan, including Free.
How does cancellation work?
Self-serve, from the billing portal, no email required. Your plan stays active until the end of the period you paid for, then drops to Free. If you are over the Free limits at that point (repositories or seats), existing data is kept but new runs are blocked until you are back within them.
What happens to my data if I leave?
You can delete your organization from the app. That purges repository metadata, run history, the code index, stored keys and tokens, and your billing link. It is a GDPR-grade deletion, not a soft delete.
Do runs use my hardware?
No. Runs execute on our infrastructure in the EU, in isolated per-run sandboxes. Your forge and your LLM provider are the only external systems a run talks to.
How long can a single run take?
A run has a hard wall-clock limit that depends on the plan: 15 minutes on Free, 30 minutes on Solo, 60 minutes on Team, 120 minutes on Business, and up to 4 hours on Enterprise. A run that hits the limit is stopped and counts like any other run.
Can I switch the connected repository on Free?
Yes. Disconnect the repository and connect a different one; the slot is freed immediately.
Does it work with a forge that is only reachable inside our VPN?
Not today. FragonForge is a hosted service and connects to your forge's API from our infrastructure, so the instance must be reachable from outside your network. If your GitLab or GHE lives strictly inside a VPN, FragonForge cannot connect to it. The exception is Enterprise, where a deployment on your own hardware can be arranged by agreement (none is live yet, and we say so). Tell us if this is your setup; it moves the priority.