Loading
90 days of the enterprise, pro and team plans free, no card needed. See the plans
Pricing
Free $0 · Pro $49/month · Team $199/month · Enterprise on request. Try any paid plan free for 90 days - no credit card. Model usage is billed by the provider you connect, or by the token from prepaid credits where the deployment sells models; every plan runs the same router. The table below compares every limit.
No monthly fee and no request quota. Buy credits, call 1 model through 1 provider endpoint with one key at the price shown before the call - the same price lands on the usage row after it. The router still chooses among the endpoints of a model, honours provider preferences and falls back when one fails.
300 requests / minute · 10 API keys with per-key spend limits · 10 seats · credits never expire
+ the tokens you use, prepaid
Prices in USD per workspace. Change or cancel any time from the billing page.
For trying the router on real traffic
No card needed
15K requests / month
Then paused until the month rolls over
30 / minute · 2 API keys · 5 seats
For one production application
or $490 a year
300K requests / month
Keep going past it at $10 per extra 100k
300 / minute · 10 API keys · 25 seats
No card, no payment - the plan is yours at once.
For a team running several apps
or $1,990 a year
2M requests / month
Keep going past it at $8 per extra 100k
1,000 / minute · 25 API keys · 100 seats · 10 tenants
No card, no payment - the plan is yours at once.
For organisations with compliance needs
Annual agreement, invoiced
20M requests / month
$5 per extra 100k, or custom quotas by agreement
3,000 / minute · 100 API keys · 1,000 seats · 50 tenants
No card, no payment - the plan is yours at once.
A request is one metered call: a routing decision, a chat completion or embedding through the proxy, an estimate or a feedback event. Quotas count per calendar month (UTC) and reset on the first. Pay as you go has no quota: each call is priced by the token and paid from prepaid credits.
| Plan | Free | Pay as you go | Pro | Team | Enterprise |
|---|---|---|---|---|---|
| Price | Free | $0 / month + tokens, prepaid | $49 / month · $490 / year | $199 / month · $1,990 / year | from $999 / month |
| Requests per month | 15,000 | no quota | 300,000 | 2,000,000 | 20,000,000 |
| Beyond the quota | paused until next month | by the token, from credits | $10 per 100k | $8 per 100k | $5 per 100k, or by agreement |
| Requests per minute | 30 | 300 | 300 | 1,000 | 3,000 |
| API keys · seats · tenants | 2 · 5 · 0 | 10 · 10 · 0 | 10 · 25 · 0 | 25 · 100 · 10 | 100 · 1000 · 50 |
| Free trial | - | - | 90 days | 90 days | 90 days |
| Routing decisions with the full trace | |||||
| OpenAI-compatible proxy (chat, embeddings) | |||||
| Your own model providers and keys | |||||
| Outcome feedback (the router learns) | |||||
| Plans: persona -> skill -> model, executed | |||||
| Tenants: per-tenant constraints and budgets | |||||
| Health and learner statistics | |||||
| Hash-chained audit trail | |||||
| Organization single sign-on | |||||
| Support | community | community | priority e-mail | dedicated | |
| First answer on a ticket | as capacity allows | as capacity allows | within 2 days | within 8 hours | within 4 hours |
| Availability target (credits in the agreement) | 99.9% monthly | ||||
| How to buy | sign up | sign up, buy credits | self-serve checkout | self-serve checkout | talk to sales |
Which plan fits
100,000 requests a month
3,333 a day · 2 a minute
What routing keeps at this volume
Quoting the catalogue…
Not sure? Start free - the router, the trace and the bill analysis are the same on every plan; a plan buys volume, seats and support.
Prices in US dollars, excluding tax: VAT, GST or sales tax for your country is added at checkout and itemised on the invoice. Cancel any time - the plan runs to the end of the period you paid for and is not renewed; see the terms on plans and payment for refunds.
Startups under two years old, open-source maintainers and educators: ask about discounts.
Photo: RDNE Stock project
For procurement, scripts and AI assistants
Everything on this platform, including the enterprise stack, is in the open-source package. Run it next to your application and keep every byte in your own network.
OSR_PLATFORM_ADMIN_PASSWORD='choose-a-long-password' \
docker compose -f platform/docker-compose.yml up -d --buildpip install "opensmartroute[yaml,server]"
osr -t targets.yaml -r rules.yaml serveStep by step, including the Azure Marketplace offer and air-gapped installs, in the install guide; Helm chart and hardening notes in the enterprise guide.
One metered call: a routing decision (POST /api/v1/route, with or without execution, a plan included), a chat completion or embedding through the OpenAI-compatible proxy, an estimate made with your key, or a feedback event. The MCP tools count as the call they make - route is a decision, estimate an estimate. Not metered: reads of your account, keys, usage, quota, the catalogue and the marketplace; anonymous quotes on this site, in the extension and the desktop bar (rate-limited per address instead). Quotas count per calendar month (UTC) and reset on the first; GET /api/v1/quota shows where you stand.
On the free plan metered calls answer 429 until the month rolls over; the dashboard warns at 80% and again at the limit. Pro and Team keep routing: requests beyond the quota are invoiced at the plan's overage rate on your next invoice, one line, no surprise tiers.
All three words mean something specific. A decision (POST /api/v1/route) tells you which target to use and why, with the full trace; you call the model yourself. Chat (/v1/chat/completions, the OpenAI-compatible proxy) chooses the model and answers the request in one call. Execution from the routing API (POST /api/v1/route with execute=true) and plans that chain persona, skill and model are Pro and up. Free includes decisions and chat; the 403 you get otherwise names the feature and the plan that has it.
No. Prices are in US dollars and exclude tax; VAT, GST or sales tax for your country is calculated from your billing address and shown on the checkout page before you pay, then itemised on every invoice. Enter a tax id at checkout and it appears on the invoice.
No. Every plan, including free, lets you connect your own provider keys (Azure, Bedrock, Vertex, OpenAI, ...); the provider bills you directly and the platform only decides. Where the deployment offers shared providers, their cost appears in your usage.
Pick Pro, Team or Enterprise and it is yours at once for 90 days - no card, no payment, nothing to cancel. When the trial ends the workspace returns to the free plan unless you subscribe; owners are reminded three days before. One trial per workspace; you can move it between plans while it runs, and Subscribe now on the billing page pays whenever you want to keep the plan. A launch offer, when one is on, works the same way with its own number of days.
No. The catalogue, rules and strategies are identical for every plan. Plans set quotas and which endpoints a key may call.
Any time, under Plan and billing. Moving between paid plans - up or down - takes effect at once and the difference is prorated on your next invoice. Cancelling keeps the plan until the end of the period you paid for and the remainder is not refunded unless the law where you live requires it; pay-as-you-go credits do not expire and every charge is a line in your wallet ledger.
Yes, on every plan including free. Both are open source and connect to your workspace with an API key; the quotes they show and the requests they route count towards the workspace's monthly quota like any other call, and the dashboard's Apps page shows what each client sent.
A usage report for the workspace - daily, weekly (the default) or monthly, or off - with what routing saved against the baseline model, requests and success, where they came from, the models that answered, what the guard protected and what changed in the workspace. Alerts and product news are separate categories you switch on or off; security and account notices always arrive.
Yes, and it is free forever under Apache-2.0. Run the SDK as a library, the sidecar container or the Helm chart; the enterprise stack is part of the open-source package.
Request text is processed in memory to make the decision and, when you ask for execution, forwarded to the chosen model provider. Usage records store the endpoint, target, status, tokens and cost, never the text; text is kept only where you choose it - conversations with memory on, facts you ask the assistant to remember, requests handed to a person in your workspace, or a request log your policy switches on. The privacy notice lists each case and how long it is kept.