Felo API PlatformFelo API Platform

Comparison

Felo OpenAPI vs LiteLLM

LiteLLM is the gateway you deploy and keep. Felo is the gateway you call: the same OpenAI-compatible shape, rates below list, and the data APIs an agent needs — on one key.

Create API keySee the price gap

  • ⁦10⁩%

    Starting rate, against the provider's official API price

  • ⁦5,000+⁩

    APIs in the same catalog

Provider list price, next to the Felo rate

LiteLLM doesn't resell tokens: it forwards your calls to providers on your own keys, so you pay each provider's list price. Felo publishes a rate for every route, and the cheapest start at ⁦10⁩% of that same list price.

List priceFelo rate

On ⁦GPT-6.1 Terra⁩, 100M input and 10M output tokens cost ⁦$150⁩ at list price — ⁦$30⁩ here.

Rates are per 1M tokens, read from the live catalog on each request. LiteLLM adds no markup of its own, so its bill is the provider's list price on your own keys.

See every model and rate

What you don't have to run

A LiteLLM proxy is infrastructure: something to deploy, patch and keep online, with your provider keys sitting inside it.

  • The proxy itself — Docker, Kubernetes or Helm, with Postgres behind it.
  • A copy of every provider key, held in the gateway and rotated there.
  • Upgrades, migrations, and the on-call that goes with running a service.

Felo is the other arrangement: one base URL and one key, and the gateway is ours to operate.

What a model gateway doesn't include: the ⁦5,000+⁩ API catalog

LiteLLM routes model calls. Felo routes those and answers from a data layer too: ⁦5,000+⁩ APIs on the same key — search, page and document reading, slides and agent runs.

Search

Real-time web search with answers and cited sources.

Web Fetch

Any page as clean HTML, text or Markdown.

X Search

Posts, users, timelines and replies from X.

YouTube Subtitling

Transcripts for a video, with optional timestamps.

LiveDoc

Upload documents and URLs, then run semantic retrieval.

SuperAgent

Streaming, multi-step agent runs over SSE.

PPT

Slide decks generated page by page and exported to PPTX.

Mindmap

Structured maps from a topic or a LiveDoc.

Landing Page

Campaign pages generated as HTML.

An agent that researches, reads, remembers and ships — without a second vendor.

Browse the API directory

Two ways to switch

Keep LiteLLM as the gateway and point its upstream at Felo, or drop the gateway and call Felo directly. The request bodies don't change either way.

# config.yaml — keep LiteLLM, add Felo as upstream
model_list:
  - model_name: <model-id>
    litellm_params:
      model: openai/<model-id>
      api_base: https://openapi.felo.ai/api/v1
      api_key: os.environ/FELO_API_KEY

# or drop the proxy and call Felo directly:
client = OpenAI(base_url="https://openapi.felo.ai/api/v1", api_key=key)

Model IDs differ between the two catalogs, so check the model page for the exact ID before you flip the base URL.

LiteLLM vs Felo, in three rows

Felo API PlatformLiteLLM
Price basisFrom ⁦10⁩% of the provider's official API priceProvider list price, billed to your own keys
Where it runsHosted; nothing to deploySelf-hosted proxy, with Postgres behind it
Beyond model callsThe ⁦5,000+⁩ API catalog on the same keyMCP tools and A2A agents through the gateway

What LiteLLM still does better

  • Runs inside your own network, on your own provider keys.
  • Open source under the MIT license, so the routing code is yours to change.
  • Enterprise add-ons: SSO, SCIM, audit logs, team admins and multi-region deployment.

If control of the gateway is the requirement, that decides it. If the bill and the data layer are, the rows above do.

LiteLLM alternative: questions before you switch

Is Felo OpenAI-compatible?

Yes. Chat calls use the OpenAI Chat Completions schema at https://openapi.felo.ai/api/v1, so the SDK and the request bodies you already have keep working. Anthropic clients use https://openapi.felo.ai/api — one level up, because that SDK appends /v1/messages itself.

Can I keep LiteLLM in front of Felo?

Yes. LiteLLM calls any OpenAI-compatible endpoint: in config.yaml, give each model an openai/ prefix, set api_base to https://openapi.felo.ai/api/v1 and api_key to your Felo key. Virtual keys, budgets and logs stay where they are.

Why is the rate lower than what LiteLLM forwards?

LiteLLM forwards to providers on your keys, so it bills at the provider's list price. Felo publishes its own rate for each route, starting at ⁦10⁩% of that list price; nothing is rebated afterwards.

Do I have to self-host anything?

No. Felo is hosted: create a key and send a request. There is no proxy to run and no database to keep.

Can I call image and video models too?

Yes. One key reaches ⁦30⁩ chat, ⁦15⁩ image and ⁦10⁩ video models, each billed at its own published rate, on the same balance.

Do I need a subscription?

No. Usage is prepaid and charged per call at the published rate — no subscription, no seat fee, no minimum commitment.

Try it on one call before you migrate

Create a key, add a balance from $1, and send a request with the SDK you already use.

Create API keyRead the LLM API docs