Prompt compression
Saves you between 2-20% of input tokens on agentic traffic. Nothing is summarized, nothing is dropped, and prompt caches keep hitting.
DocumentationPricing is set on a short discovery call. Tell us how you run GoModel - traffic, teams, and the features you need - and get a quote that fits your deployment.
Book a 30-minute discovery callWith the founder who wrote the code, not a sales team.Pro is licensed under the Commercial License Agreement.
Almost every row below is identical, and that is the point. The open-source gateway is not a demo of the paid one.
| Capability | GoModel (MIT) | GoModel Pro |
|---|---|---|
| The gateway itself | ||
| Providers, model aliases, virtual models | All of them | All of them |
| OpenAI and Anthropic compatible endpoints | Included | Included |
| Caching, failover, load balancing | Included | Included |
| Budgets, rate limits, per-user API keys | Included | Included |
| Audit logs, usage and cost tracking | Included | Included |
| Admin dashboard, MCP gateway, guardrails | Included | Included |
| OpenTelemetry traces and metrics | Included | Included |
| Requests, seats, nodes, environments | Uncapped | Uncapped |
| License | MIT, forever | Commercial, on top of the same MIT core |
| What the license adds | ||
| Prompt compression | Not included | 2-20% fewer input tokens on agentic traffic |
| OIDC single sign-on | API keys only | Authorization Code + PKCE, group-gated, audited |
| Per-child quota templates | One shared subtree budget | One rule, an independent budget per child path |
| Intelligent routing | Round robin and cheapest-cost | Easy/hard tiering + adaptive selection (beta) |
| Egress proxy pools | One proxy per provider | Health-checked pools, failover, rules across providers (beta) |
| Support | Discord and GitHub issues | Founder-level support (early offer) |
Without a license the Pro binary starts as the open-source gateway and keeps serving traffic. Three features are the exception: single sign-on, egress proxy pools, and per-child quota templates. If you explicitly enable one and the license does not cover it, startup fails rather than quietly dropping a login boundary, sending traffic from the wrong IP, or changing how limits are enforced.
The full list is in the table above. Intelligent routing and egress proxy pools are in beta; routing is described further down, without the marketing voice.
Saves you between 2-20% of input tokens on agentic traffic. Nothing is summarized, nothing is dropped, and prompt caches keep hitting.
DocumentationOkta, Entra ID, Google, Auth0, or Cognito in front of the dashboard. Deploy the gateway org-wide instead of key-by-key.
DocumentationA fixed egress IP that survives a proxy outage. Named pools of HTTP or SOCKS5 proxies with health checks, failover, and rules that cover a whole fleet of providers. Beta.
DocumentationDirect access to the person who wrote the code.
Compression removes 2% to 20% of input tokens depending on the traffic: long agent sessions that replay files and tool output sit at the top, short chat at the bottom. Requests below the thresholds pass through unchanged.
Input tokens removed, by traffic shape. Output tokens are untouched.

Every rewrite is visible in response headers, audit revisions, metrics, and the dashboard's own savings figure. And if Pro is not for you, there is the 30-day money-back guarantee.
A deterministic tier router scores each request on signals like reasoning parameters, code fences, and context size, then sends it to an easy or hard model tier. An adaptive selector picks the healthiest, cheapest target inside that tier from live latency and error data. Every verdict is explainable in response headers and the audit trail.
Included in the license and working, but beta: routing state is per node rather than shared across a cluster, and streaming latency is measured to stream establishment. Turn it on first for a workload where a wrong tier is cheap.
Routing documentationNothing counts your requests, seats, nodes, or spend. License validation is an offline signature check - it works air-gapped.
Stop paying and the binary keeps running as the open-source gateway, with every feature it has today. Switch off SSO, egress pools, and per-child quota templates before the next restart.
One 30-minute call with the founder sets the price. No sales team, no demo sequence, no procurement maze.
Run the open-source gateway first - it is a single binary and it takes a minute. When compression, single sign-on, or per-tenant budgets become the thing standing between you and shipping, come back.