Any model + fallbacks

Any model you want, and a backup ready before you need it

Pick from live catalogs of Anthropic, OpenAI and 350+ OpenRouter models, with context size, price and tool support shown up front. Stack up to five backups from any provider, and if a model errors or stalls, the next one answers at once. Every attempt, token and cent is on the record.

The primary model is rate limited, so the request instantly fails over to a backup model from another provider, and every attempt is loggedNew messageNeeds a replyclaude-sonnet-5Anthropic · primarygpt-5OpenAI · backup 1llama-4-maverickOpenRouter · backup 2429 rate limited200 OK · 1.8 sYes, we can deliverby Friday. Want meto confirm it?Run · model attempts#1claude-sonnet-5rate_limit (429)0.4 s#2gpt-5ok · reply sent1.8 sno retry waits: SDK retries off
350+models in the picker
5backup models per character
60 sdefault timeout before failover
Per runtokens and cost recorded
Deep dive

A closer look at Any model + fallbacks

01 · Picker

A model picker that shows you what matters

Search live catalogs from Anthropic and OpenAI (with your key) and OpenRouter. Each model shows its context window, price per million tokens and whether it supports tool calling and thinking. Know a model id that isn't listed yet? Type it in.

  • Live catalogs, cached for 30 minutes
  • Context size and price per million tokens
  • Tool-calling and thinking support flags
  • Any model id can be typed in
Searching the model picker shows each model's provider, context size, price tier and tool and thinking supportCharacters › Ava › AI modelclaudeAnthropicOpenAIOpenRouter · 350+MODELPROVIDERCONTEXTPRICETOOLSTHINKclaude-opus-5-5Anthropic1M$$$claude-sonnet-5Anthropic1M$$$claude-haiku-4-5Anthropic200K$$$anthropic/claude-sonnet-5OpenRouter1M$$$anthropic/claude-3.5-haikuOpenRouter200K$$$Selected · any id can be typed in
02 · Failover

Outages become a non-event

Give a character up to five backup models, mixing providers. If a model errors, hits a rate limit, refuses, is missing a key or passes the character's timeout, the next one is tried immediately. SDK retries are switched off so nothing waits twice, and every attempt shows in the debugger.

  • Up to five cross-provider backups
  • Fails over on errors, rate limits, refusals, missing keys and timeouts
  • No double waiting: SDK retries are off
  • Every attempt visible in the debugger
The primary model times out, the first backup has an outage, and the second backup answersFALLBACK CHAIN · UP TO FIVE BACKUPSPRIMARYclaude-sonnet-5AnthropicBACKUP 1gpt-5OpenAIBACKUP 2gemini-3-proOpenRouterBACKUP 3gpt-5-miniOpenAI+ add backuptimeout 60 s503 outageansweredFAILS OVER ONErrors & outagesRate limitsRefusalsMissing keysTimeoutsNext model tried at once, with SDK retries off, so thecontact never waits twice. Every attempt is in the debugger.
03 · Cost

Know what every conversation costs

Each run records its token usage and cost, so you can see exactly what an agent spends per conversation, per booking and per day, and choose the model that gives the best result for the money.

  • Input, output and reasoning tokens per run
  • Cost per run and per agent
  • Adaptive thinking with per-character effort
  • Use your own provider keys or the server defaults
Each run records input, output and reasoning tokens with its cost, rolled up into daily spend and cost per bookingRun #48213 · usageInput tokens3,412Output tokens286Reasoning tokens512Cost$0.0142AI spend · last 7 daysMTWTFSS$0.38cost per booking$0.011avg cost per reply2,914replies this week
Under the hood

The details

Anthropic

Messages API with adaptive thinking, summarised reasoning and per-agent effort.

OpenAI

Responses API with reasoning summaries and encrypted reasoning passed back each turn.

OpenRouter

350+ models through one key, with unified reasoning kept across tool calls.

Your keys

Add your own provider keys in Settings, encrypted at rest with AES-256-GCM.

Timeouts

A per-character timeout (default 60 s) that triggers the next backup.

Health

Paused providers and failing keys are surfaced before they become a problem.

Use cases

Who it's for

Always-on SMS

A fast primary model with a different-provider backup keeps replies flowing through outages.

Cost control

A premium model for complex steps, a cheaper model for routing and confirmations.

Model evaluation

A/B test two characters on different models and compare conversions.

FAQ

Any model + fallbacks: common questions

Can I mix providers in the fallback chain?

Yes. Backups can come from Anthropic, OpenAI and OpenRouter in any order.

Does failover slow replies down?

It's immediate. SDK retries are switched off so a failed model never makes the contact wait twice.

Can I use a model that isn't in the list?

Yes. Type any model id into the picker.

Where do I see what a run cost?

In the debugger for each run and in Analytics across agents.

Get started

Put it to work on your own inbox

Connect an inbox, describe the job, test it in a sandbox and go live with exactly the autonomy you're comfortable with.