Loading
Reality check
OpenSmartRoute publishes a lot of figures: catalogue sizes, indexed resources, live counters, benchmark results and worked examples. They are not the same kind of number, so this page keeps them apart. Everything below is computed from the running platform when you load it - nothing here is a static claim.
The numbers, by class
A catalogue count is not an executable count; an index of imported listings is not adoption; a lifetime counter is not a benchmark. The badge on each tells you which it is.
Platform v1.11.0-preview.138 · developer preview of the next release · public hosted deployment · computed 2026-10-07 17:35 UTC, refreshed every minute
Measured saving: 26% lower modelled cost than sending every executed request to llm-onprem, over 762 executed requests on the public deployment - a measurement of this deployment, not a customer result.
Routing SLM: a trained routing model votes in the ensemble; autonomous retraining (the autopilot) is switched off on this deployment, so the model only changes with a release.
Capability by capability
Live means it works on this deployment at this moment. Limited means the code runs but something here narrows it, and the reason is written next to it. Not here means switched off or not part of this edition. Each row links to where you can check it yourself.
| Capability | Status here | Why | Check it |
|---|---|---|---|
| Route and answer | |||
| Smart routingSend a request; the router picks the model, agent, skill, tool or person that fits it best and tells you why. | Live | 91 targets in the catalogue, 19 of them answer on the platform's own providers. | /platform/playground |
| OpenAI-compatible chatPoint any OpenAI client at /v1 with model "auto" and the answer comes from the model the router chose. | Live | ||
Model providers
The SDK ships adapters for every family below - that is what a self-hosted router or your own keys can use. Whether this public deployment has one configured, and whether any requests actually went through it, are separate columns.
| Provider | SDK support | Configured here | Traffic · last 7 days |
|---|---|---|---|
| OpenAI | Yes | Not on this deployment | None |
| Azure OpenAI / Azure AI Foundry | Yes |
Not claimed
Things a buyer might assume from a product site of this kind, stated plainly so nobody has to guess.
None are published yet. The savings figures on this site are the deployment's own measurement and the published benchmark, never a customer's result.
No SOC 2, ISO 27001 or HIPAA certification is held. The trust page says what is targeted and what you can verify today; industry pages describe controls you configure, not compliance you inherit.
The reference catalogue indexes hundreds of hosted models; this deployment executes on the targets and providers in the table above, and a workspace can add its own keys for the rest.
The marketplace index is mostly imported from open registries so one search covers them. Installs and ratings of what is published here are the small numbers, shown with their scope on the marketplace page.
The ROI calculator opens with an example workload, the bill section shows an invented export, the industry pages show reference rule files and the product demos use made-up people and drafts. Each carries an Example badge.
Machine-readable, no key needed, cached for a minute:
GET /api/v1/factsthe numbers and their classesGET /api/v1/servicesthe capability statusesGET /api/v1/modelstargets with providers and trafficReadiness and incidents are on the status page.
| 14 models answer here; streaming and tool calls included. |
| /platform/playground?mode=chat |
| Cost quotes before you sendTokens, price per candidate model and the recommended pick for a request - no account needed. | Live | Free without a key (rate limited per address); metered as an estimate with one. | /estimate |
| Plans: persona, skill, modelOne request becomes a composition - a persona, a skill package and the model to run them - with a step-by-step trace. | Live | Included from the payg plan. | /platform/playground |
| Human hand-offA request that needs a person is queued for one; the answer feeds the learners like any other outcome. | Live | 1 human target in the catalogue. | /platform/dashboard/handoffs |
| MCP serverRoute, estimate, recommend, explain and report outcomes from Claude, Cursor or any MCP client over JSON-RPC. | Live | Served at /mcp with the same key as the API; tool list readable without one. | /platform/dashboard/integrations |
| Learn and improve | |||
|---|---|---|---|
| Learning from outcomesReport whether an answer was good; the strategies re-weight and the next decision gets better. | Live | Included from the free plan. | /platform/dashboard/activity |
| Routing SLM and autopilotA small language model trained on routing outcomes votes in the ensemble; the autopilot retrains it and promotes a better challenger. | Limited | Trained routing model loaded and voting; autonomous retraining is switched off on this deployment (OSR_PLATFORM_AUTOPILOT), so the model only changes with a release. | /platform/dashboard/learning |
| Your own model providersBring your own keys - Azure OpenAI, OpenAI, Bedrock, Vertex, Ollama or any compatible endpoint - and the router executes on them. | Live | Included from the free plan. | /platform/dashboard/providers |
| Operate and govern | |||
|---|---|---|---|
| Traces, events and telemetryEvery decision has a trace you can open: the signals, the strategies' votes, the ranking, the execution and its cost. | Live | Tracing is on; traces are scoped to your workspace. | /platform/dashboard/events |
| Savings ledgerWhat routing saved against always sending to the most expensive model, per request and per month. | Live | Computed from your metered usage; the public status page shows the platform-wide figure. | /platform/dashboard/savings |
| Policy, budgets and guardsDeny lists, allowed regions, monthly budgets, prompt-injection and PII guards - set per workspace, enforced on every request. | Live | Input guard on, PII redaction on, output guard flag. | /platform/dashboard/governance |
| Tenants, audit trail and healthPer-tenant constraints and budgets, a hash-chained audit trail of every decision, circuit breakers per target. | Live | Included from the team plan. | /platform/dashboard/audit |
| Alerts and notificationsBudget, error-rate, quota and provider alerts to e-mail, Slack, Teams or a webhook, plus the in-app inbox. | Live | Evaluated every 60 s. | /platform/dashboard/notifications |
| Usage digests by e-mailA daily, weekly or monthly digest of what your workspace routed, what it cost and what it saved, in your time zone. | Live | Delivered through m365. | /platform/dashboard/account |
| Support desk and assistantAsk the assistant (it answers from the documentation through the router) or open a ticket a person answers. | Live | 1121 documentation passages indexed; the assistant runs in router mode. | /support |
| Catalogue, community and news | |||
|---|---|---|---|
| MarketplaceAgents, skills, personas, prompts and MCP tools you can install into your routing pool - or publish yourself. | Live | 282,426 published listings from 111,731 publishers. | GET /api/v1/registry |
| Model catalogue and rankingsEvery routing target with measured traffic, the reference LLM catalogue with list prices and the published-benchmark leaderboard. | Live | 465 reference models, refreshed 3.0 times a day. | GET /api/v1/rankings |
| BlogDaily notes on new models, LLM releases, agent frameworks and research, written through the router from the sources we follow. | Live | Feeds polled every 3 h; posts publish automatically. | GET /api/v1/blog |
| NewsletterA morning issue (and an evening one on days with more news) or a weekly digest with the new posts and what the platform runs for you; registered people get it with their account. | Live | Arrives at 09:00 and 18:00 in each reader's time zone; new accounts are enrolled once their address is confirmed. | /newsletter |
| Content publishing for your websiteFollow your own RSS feeds or send a brief: posts are written through the router as your workspace's requests and delivered to your website - a signed webhook, WordPress or Ghost - or pulled as RSS; the same tools are on the MCP server. | Live | Feeds polled every 60 min; up to 20 feeds and 10 destinations per workspace. | /platform/dashboard/content |
| Ship it your way | |||
|---|---|---|---|
| Single sign-onSign in with GitHub, Google, Microsoft or GitLab; organizations bring their own IdP with e-mail-domain login and just-in-time membership. | Live | Shared providers: microsoft, google, github. Organization SSO from the enterprise plan. | GET /api/v1/auth/providers |
| Plans and billingStart free, upgrade in the dashboard, pay by card or from a prepaid credits wallet; every metered row is priced. | Live | Card payments through Stripe; plans: enterprise, pro, team. | /pricing |
| CLI, installers and SDKsOne line installs the osr CLI; the Python and TypeScript SDKs speak to this platform with the same key. | Live | curl -LsSf https://opensmartroute.ai/install.sh | sh - then osr login --url https://api.opensmartroute.ai. | /downloads |
| Self-hostingThe same router as a service on your own infrastructure: Docker image, Compose bundle, Helm chart and the Azure Marketplace package. | Live | Zero-dependency Python package; osr serve, the image and the chart ship with every release. | /downloads |
| Browser extension and desktop appSee what a prompt costs before you send it on ChatGPT, Claude, Gemini and the provider consoles; one shortcut opens the router anywhere. | Live | Free and open source; they sign in with your account and count under Apps in the dashboard. | /extension |
| Not on this deployment |
| None |
| Anthropic | Yes | Not on this deployment | None |
| Google Gemini / Vertex AI | Yes | Not on this deployment | None |
| AWS Bedrock | Yes | Not on this deployment | None |
| Mistral | Yes | Not on this deployment | None |
| Ollama (self-hosted open models) | Yes | Yesollama-deepseek-r1-8b, ollama-glm-4-7-flash, ollama-glm-ocr, ollama-gpt-oss-20b, ollama-qwen3-5-4b, ollama-qwen3-embedding-4b, ollama-translategemma-4b | Yes673 requests |
| Any OpenAI-compatible endpoint (vLLM, Groq, Together, OpenRouter, Z.ai ...) | Yes | Yesspeaches | None |
SDK support means the package ships an adapter; a self-hosted router or a workspace's own keys can use any of them. Configured and traffic describe this public deployment only, from GET /api/v1/facts and GET /api/v1/models.