AI Stupid Level
Help

Support

Help center

Everything you need to know about the AI Smart Router.

Contact supportAPI documentation

Quick links

Create API keyYour aism_ key, for every tool you connectAdd providersKeys for OpenAI, Anthropic, Google, DeepSeek, Kimi or GLMAPI docsEndpoints, parameters, errors and examplesMonitoringPer-key activity, costs, prompt audit and budgetsAnalyticsCost patterns across your requestsPreferencesDefault strategy, fallbacks and latency limits

Getting started

What is the AI Smart Router?

The AI Smart Router is an intelligent API gateway that automatically selects the best AI model for your requests. Instead of manually choosing between Claude, GPT, Gemini, DeepSeek, Kimi or GLM, our system continuously benchmarks all models and routes your requests to the optimal one based on performance, cost, and your preferences.

Quick start (5 minutes)

  1. Create a universal API key

    Go to API keys → Click "Create key" → Copy and save it securely. Your key starts with aism_

  2. Add provider API keys

    Go to Providers → Add keys for OpenAI, Anthropic, Google, DeepSeek, Kimi or GLM. The more providers you add, the better the routing!

  3. Configure your tool

    Use your aism_ key with any OpenAI-compatible tool:

    Tool settings
    Base URL: https://aistupidlevel.info/v1
API Key: aism_your_key_here
Model: auto-coding (or auto, auto-reasoning, etc.)

Help topics

Set up your tool, choose a routing strategy, keep an eye on spend, and fix common problems.

Setting up your IDE / tool

Your aism_ key works with any tool that supports custom OpenAI base URLs. For detailed setup instructions with copy-paste configs, visit the API keys page and expand "How to use your API key".

  • Roo CodeSettings → "OpenAI Compatible" → paste Base URL + Key
  • ClineSettings → "OpenAI Compatible" → paste Base URL + Key
  • Kilo CodeSettings → API Provider "OpenAI Compatible" → paste Base URL + Key
  • ContinueEdit config.yaml → provider: openai + apiBase + apiKey
  • Cursor IDESettings → Override OpenAI Base URL (chat/plan mode only)
  • AiderSet OPENAI_API_BASE + OPENAI_API_KEY env vars
  • Open WebUIAdmin → Connections → OpenAI → Add Connection
  • ChatboxAPI Host: https://aistupidlevel.info (no /v1)
  • TypingMindFull endpoint URL per model + Bearer header
  • Jan.aiOpenAI extension → paste URL + Key
  • LibreChatlibrechat.yaml custom endpoint with fetch: true
  • AnythingLLMGeneric OpenAI provider → Base URL + Key
  • LiteLLMlitellm-config.yaml → openai/auto-coding
Known limitations

Cursor: Only chat/plan mode uses your base URL — Composer, inline edit, and autocomplete are locked to Cursor's backend.

Windsurf: Cascade is locked to Codeium's backend. Install Roo Code or Cline inside Windsurf instead.

Routing strategies (virtual models)

Use these as the model parameter. The router picks the best real model based on live benchmarks.

Model IDStrategyBest for
autoSaved preferenceUses your configured default strategy
auto-codingBest codingCode generation, debugging, refactoring
auto-reasoningBest reasoningComplex analysis, math, logic puzzles
auto-toolingBest tool useAgents and coding assistants that call tools
auto-valueBest valueNear-top quality for the fewest dollars (combined score)
auto-value-codingBest value: codingMost coding points per dollar among near-top models
auto-value-reasoningBest value: reasoningMost reasoning points per dollar among near-top models
auto-value-toolingBest value: tool useMost tool-use points per dollar among near-top models
auto-consistentMost consistentSteadiest near-top model, answers vary least day to day
auto-fastest-qualityFastest good modelFastest model within 5 points of the best
auto-cheapestLowest priceLowest list price per token, no quality bar
auto-fastestLowest latencyInteractive apps, chatbots, quick queries

You can also pin a specific model (any id GET /v1/models lists) to bypass routing entirely.

Every strategy in the API documentation

Monitoring & budgets

The Monitoring page provides per-key activity tracking, cost dashboards, prompt auditing, and budget management.

Key activity
Per-request log with provider, model, cost, latency, and category
Cost dashboard
Daily/weekly/monthly spend breakdown by model and provider
Prompt audit
Encrypted prompt logging with PII scrubbing (opt-in)
Budgets & alerts
Monthly spend limits (soft warnings + hard enforcement) per key
Efficiency report
Success rates, category distribution, avg cost per request
Troubleshooting
401 Unauthorized
Check your aism_ key is correct and not revoked. Include it as Authorization: Bearer aism_... header. Some tools also accept sk-aism_ prefix.
No models in dropdown
Your tool calls GET /v1/models — this requires a valid aism_ key. Check that auth is configured correctly.
Model not found
If using a specific model ID, check spelling. Direct pin routing infers the provider from the model name prefix.
High costs
Set budget limits in Monitoring → Budgets. Use auto-cheapest strategy. Check Analytics for cost patterns.
Slow responses
Use auto-fastest strategy. Set latency limits in Preferences. Enable streaming for better perceived speed.
Provider key validation failed
Re-check the key on the provider's dashboard. Ensure it has active credits and correct permissions.
CORS errors in browser
The /v1/* endpoints allow all origins. If you see CORS errors, check that your tool is using the correct base URL.

Frequently asked questions

Still unsure? The API documentation covers every endpoint, or contact support.

How much does it cost?

You pay the providers directly for tokens, with your own keys; we add nothing to token prices. Your plan includes routed requests each month (Free 1,000, Pro Intelligence 10,000, Developer 100,000, Teams 1,000,000), and prepaid credits cover any beyond that at $5 per 10,000. Only successful requests count.

Is my data secure?

Provider keys are encrypted at rest (AES-256-GCM) and never shown again. Prompt logging is off unless you turn it on; when it is on, API keys, tokens, email addresses and phone numbers are removed before the prompt is encrypted and stored.

What happens if a model fails?

Automatic fallback: if the chosen model fails before answering, the router tries the rest of your traffic split (if you use one), then your own fallback order (if you set one), then up to 2 automatic fallbacks from different providers. A model that keeps failing is skipped without waiting for a timeout. Configure it in Preferences.

Does streaming work?

Yes — set stream: true. Text arrives as the model writes it, for every provider; tool calls arrive once each is complete, then usage if you ask for it with stream_options.include_usage.

How often are benchmarks updated?

Coding runs every 4 hours, reasoning and tool use daily, and a small canary every hour. Rankings refresh after each run, and the router reads the same scores the leaderboard shows.

Can I use a specific model directly?

Yes — send any real model ID (GET /v1/models lists them) instead of an auto-* strategy. The router forwards directly to that provider; your plan allowance and key budgets still apply.

Does tool calling work with agents?

Yes. Send OpenAI-format tools and return results as role "tool" messages with tool_call_id. The router keeps each model's own turn (such as Claude's thinking) between requests, so multi-step agent loops work, and a loop stays on the model that started it. Forcing a call with tool_choice works on every model: where a provider cannot enforce it, the router asks the model in words and checks that the answer really is the tool call.

Can I send images, audio or files?

No — the router routes text only, on purpose. It picks a model from our benchmark results for coding, reasoning and tool use, and none of those measure image, audio or file understanding, so we would be guessing. Such a request gets a clear 400. Call the provider directly for multimodal work.

Which providers are supported?

OpenAI, Anthropic, Google (Gemini), DeepSeek, Kimi (Moonshot) and GLM (Z.AI). Add as many as you like.

Does it work with embeddings?

Yes — POST /v1/embeddings proxies to OpenAI embedding models. Required by Continue, LibreChat, and Open WebUI for RAG.

Still need help?

Contact supportSomething is not working the way it should.Sales & pricingWhich plan fits, volume pricing, annual terms.SecurityVulnerability reports and security questions.Provider statusEvery provider, checked every 10 minutes.

Need more detail? Check the API documentation