Workflows from $50/month

The open-source version is free forever — self-host the whole toolchain, no usage gates, no platform fee. The paid tiers exist because running the managed infrastructure (workflow engine + runtime fleet + AI gateway + git storage + observability) has real per-customer cost. A free SaaS tier subsidised by paying customers isn't sustainable for a small team; we'd rather price honestly and offer a 30-day money-back guarantee on every paid plan.

A token is the same unit the underlying AI providers (Anthropic, OpenAI, Google, etc.) bill on — roughly ¾ of a word in English. Forge meters every persona's LLM call across all your workflows. Real usage varies: a typical research workflow runs ~3M tokens; a typical SDLC workflow ~10M; a quick content workflow ~1M. You can see live token usage per workflow in the dashboard.

Overage kicks in at the per-tier rate (Starter $0.040/Ktok, Pro $0.030, Business $0.025, Enterprise $0.020 or pass-through). No throttling, no surprise stops — your workflows keep running and the overage is billed at the end of the cycle. You can set an overage budget cap in the dashboard if you want a hard ceiling.

Yes, on every paid tier. With BYOK, AI inference is billed by the provider directly to you, and Forge charges only the platform fee — no per-token markup on top of your contract. Useful if you've negotiated provider discounts or have compliance requirements about where inference happens. Recommended for Enterprise.

If Forge doesn't work for you within the first 30 days of your paid plan, email[email protected]and we refund the platform fee. AI inference costs (if you used pooled provider keys) are pass-through and not refundable, since we already paid the provider. No hoops, no "talk to retention."

Yes. Upgrades take effect immediately and are prorated. Downgrades take effect at the next billing cycle. If you hit an included-token cap regularly, the overage math usually means upgrading saves money — the dashboard tells you when.

forge-cli, forge-hook, the conductor, the workflow engine, the multi-agent SDLC pipeline, LLMOS, and the AI gateway are on GitHub under MIT or Apache 2.0. No usage restrictions, no telemetry gates, no phone-home. Self-hosting is fully supported for indefinite use.

Yes. The same workflows, characters, and agent configs work in both modes. Migrating is a config change, not a rewrite. Same the other direction — you can take your data out of SaaS and self-host whenever you want.

On-prem deployment (Enterprise tier) supports any infrastructure you own — your VPC, your bare-metal, fully air-gapped if needed. We're on the SOC 2 Type I path on the global track, plus regional compliance roadmaps for specific regulated industries. Talk to us about specific requirements.