Give your agent abilities

Beyond models, one Flux key gives your agent abilities: search, reading pages, speech, transcription, images, apps and a second opinion. See where each one stands and how to call it.

Your agent already has a brain. Abilities let it do more with the same Flux key: read a page, search the web, speak and listen, make images, act in your apps and have an answer checked. There is no second account, no second key and no second bill.

What an ability is

An ability is something Flux does for your agent beyond answering with a model. Each one runs on the gateway, is billed to the same account as your model calls, and comes back with a receipt showing what it cost.

You reach an ability in one of two ways:

  • As a tool, through the Flux tools server. You add one MCP server to your agent and it sees the Flux tools next to its own. This is the easy path for Claude Code, Codex, Cursor and other agents. See Set up the Flux tools server.
  • Over the API, from your own code. Each ability has its own endpoint on https://api.fluxrouter.ai/v1, with the same Authorization: Bearer key. The endpoints are listed in the API reference.

Apps are the exception: they are reached only as tools.

Where each ability stands

The badges read from the same record as the Abilities pages, so they change the moment an ability does. A part that works today has had a paid call checked end to end. A part that is being switched on is running in production but still waits for that check, so it may not work on every key yet. A part that is on its way is built but not yet reachable with a Flux key: over the API it answers 404, and as a tool it answers ability_disabled.

AbilityOver the APIAs a toolGuide
Flux tools servern/aLiveSet up the tools server
Read a web pageLiveLiveRead a page
Search the webRolling outRolling outSearch the web
Speech to textLiveShippingSpeech to text
Text to speechRolling outShippingText to speech
Voice listLiven/aChoose a voice
Make imagesLiveLiveMake and edit images
Edit imagesLiveNo tool yetMake and edit images
AppsTools onlyShippingUse apps
Second opinionRolling outn/aSecond opinion
Routing servern/aLiveThe routing server
MemoryShippingn/aComing later

Vision, sending images to a model in a chat request, is part of every chat call rather than a separate ability. See Vision and image inputs.

  • Most abilities need a paid plan with a cleared payment, or pay-as-you-go. That covers page reading, search, speech, transcription and images. On the Free plan the call is refused with premium_locked before any work is done, so nothing is charged.
  • Tools stay visible on the Free plan. Your agent still sees a paid tool, and a call returns premium_locked with a link to upgrade. After that refusal the tool drops out of the agent's tool list, so it stops trying. web_fetch is the exception: it stays listed and answers payment_required with the same link.
  • Apps work differently. Accounts without a paid plan or credit can run app actions for free, up to 50 runs a day, then need credit. See Use apps.
  • Your key must be allowed to use the ability. If you limited a key to certain models, add the ability's model names too (for example flux-voice or flux-image), or the call is refused with 401 or 403.

Shared limits

Speech to text, text to speech and search share one rate gate per account: up to 60 requests a minute across all three together, and 36,000 seconds of audio an hour, also across all three. Over it, a call answers 429. Images and page reading are not counted. Apps have their own limits, listed on Use apps.

Where prices live

  • Pricing lists every ability's current price.
  • GET /v1/models carries the price of each speech, transcription and image model, so your code can read it live. See The models list.
  • Every response says what that call cost, in the x-flux-cost-usd header or, for a tool, in the result.

Pay-as-you-go accounts pay a small uplift on top of the list price. The ability pages print each price from the same catalogue as the Pricing page, so they never disagree.

Next steps

  1. Set up the Flux tools server in your agent.
  2. Try an ability: read a page or make an image.
  3. Put abilities together: talk to your agent, research with sources or chain apps.
  4. If something goes wrong, the error reference decodes every ability error.