CometAPI

Reach CometAPI's multi-provider model catalog through one FlowRunner connection. Agents compare or fail over between models without a second credential or a second connector.

6 actions API key available

CometAPI website ↗ Platform Documentation ↗ Capability data verified 2026-08-01

A campaign brief is approved and hands the agent a list of items to produce

Agent runs Get API Key Usage to read the quota left on the key

Agent reads the per-million-token pricing for the chosen model from List Models

Agent projects the batch against the remaining quota and finds it does not fit

Agent reports the shortfall and names the other live flows that share the key

The operations owner decides to trim the batch, raise the quota, or let it run and accept the pause

What This Integration Enables

Most model gateways make you find out what something cost after you spent it. CometAPI does not, and that is the reason to choose it over a plain passthrough. List Models returns each model's provider, type, capabilities, and per-million-token pricing in the same call, and Get API Key Usage returns quota, usage, expiration, and model limits for the key you are actually running on. Those two operations turn cost from a monthly surprise into a value a flow can read before it acts.

That matters because the key is a shared resource. One CometAPI key typically backs several flows at once, and a batch that consumes the remainder does not fail politely on its own path. It takes down the ticket triage flow and the nightly summarization run with it. An agent that can read the meter can be built to respect it.

The surface is genuinely multimodal on that one meter: chat completions covering GPT, Claude, Gemini, Grok, DeepSeek, and Qwen; embeddings for retrieval; image generation; and text to speech, whose audio lands in FlowRunner file storage and comes back as a URL you can pass straight to the next step. Nothing streams and nothing runs as an async task, so every action returns a finished result and the flow stays readable.

Without FlowRunner

Spend discovered after the fact The first signal that a key ran dry is an unrelated flow failing

Model cost is a lookup somewhere else Choosing between two models means opening a pricing page in another tab

Separate accounts per modality Text on one vendor, images on a second, speech on a third, reconciled by hand

With FlowRunner

Quota is readable inside the flow The agent checks what is left before it commits to a batch

Cost travels with the catalog Per-million-token pricing arrives in the same call that lists the models

One meter for four modalities Chat, embeddings, images, and speech draw on the same balance

Use Case Scenarios

Pre-flight on a bulk product description run

A merchandising team pushes several hundred new SKUs. Before generating a single description, the agent calls Get API Key Usage, reads the per-million-token price for the selected model from List Models, and estimates the run against what is left. If it fits, the agent generates and writes results to Shopify. If it does not, the batch never starts, which is the whole point: a half-finished catalog is worse than an unstarted one.

Localized voice notes for a customer campaign

A lifecycle campaign needs short audio versions of a written message in several languages. The agent generates each translation with Create Chat Completion, converts each one with Create Speech, and receives back a stored file URL for every language. Those URLs get attached to the campaign in Klaviyo. Text and audio come off the same key, so there is no second vendor to reconcile at month end.

Retrieval-backed answers with a cost ceiling

Support articles are embedded with Create Embeddings and stored in Pinecone. Incoming questions are embedded, matched, and answered by a chat model grounded in the retrieved text. The agent tracks usage against the key as the day runs and switches to a cheaper model from the catalog once a defined share of the daily quota is spent, rather than stopping the service entirely. Degrading deliberately beats failing accidentally.

Human-in-Loop Highlight

The gate here is spend against a shared meter, and it exists because exhausting a CometAPI key is not a local failure. When an agent is handed a batch large enough that Get API Key Usage says it will consume most or all of what remains, it does not start and hope. It stops before the first request and posts to the finance and operations channel in Microsoft Teams: "This run needs roughly 70 percent of the quota left on the shared key. Four other live flows draw on the same key, including support triage. Run it, trim it to the top 200 items, or wait for the quota to reset?" The person answering can see both numbers the agent used, because both came from the connector rather than from an estimate. Spend cannot be recalled once the tokens are consumed, and the flows that would break are not the ones asking. This is an automation exception in its most literal form: the agent is capable of proceeding and stops anyway, because the cost lands on somebody who was not in the room.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

6 actions

Text Generation

Create Chat Completion Sends a conversation to any chat model in the catalog through one OpenAI-compatible interface, covering GPT, Claude, Gemini, Grok, DeepSeek, Qwen, and more. Supports tool calling, JSON and JSON Schema structured output, and reasoning effort control. Parameter support varies with the provider behind the model, so the catalog is worth reading before committing a flow to a specific option.

Embeddings

Create Embeddings Turns a batch of strings into vectors in one call, returning one vector per input in the order supplied. Used for semantic search, deduplication, and retrieval grounding. Inputs must stay inside the model's token limit, which is 8,191 tokens for the text-embedding-3 family.

Images

Generate Image Generates an image from a text prompt using an OpenAI-compatible image model such as GPT Image, Gemini image models, or FLUX. Supported sizes and quality values depend on the model chosen, and GPT image models return base64 data rather than a URL. The action returns the finished image rather than a job to poll.

Audio

Create Speech Converts text into spoken audio with a chosen voice and model, stores the result in FlowRunner file storage, and returns its URL. Input is capped at 4,096 characters per request and playback speed is adjustable. Used when a flow needs an audio asset it can hand to the next step without a manual download.

Catalog and Account

List Models Reads the public catalog with each model's provider, type, capabilities, and per-million-token pricing. Filter by capability before selecting a model ID for another action, and re-run it when you need current pricing or newly released models.
Get API Key Usage Returns quota, usage, expiration, and model limits for the configured key. Used as a pre-flight check before a large batch, and as the trigger condition for a low-balance alert that reaches a person before an unrelated flow starts failing.

Frequently Asked Questions

What can FlowRunner do with CometAPI?

FlowRunner agents can run Create Chat Completion, Create Embeddings, and Generate Image in CometAPI, plus 3 more actions.

Does connecting CometAPI to FlowRunner require OAuth?

No. CometAPI connects to FlowRunner with an API key, no OAuth flow required.

Can CometAPI trigger a FlowRunner workflow automatically?

CometAPI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with CometAPI

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us