CometAPI

@CometAPI list all models available to my key and show which ones support function calling

Your API key has access to 47 models across 8 providers. Function-calling is supported by 12 models including gpt-4o, claude-3-5-sonnet-20241022, gemini-2.0-flash-exp, and llama-3.3-70b-versatile. The list also includes embedding models (text-embedding-3-large, voyage-3), vision models (gpt-4o-mini, claude-3-5-haiku-20241022), and reasoning models (o1, o3-mini). Each entry shows the provider, protocol type (chat/embeddings/moderation), and any usage restrictions tied to your key tier.

This example demonstrates catalog discovery before committing to a generation workflow. Model availability depends on your CometAPI organization tier and key restrictions. The MCP does not expose pricing or rate-limit details per model — consult the CometAPI dashboard for cost planning before high-volume use.

@CometAPI create a chat completion using claude-3-5-sonnet with the message 'Explain quantum entanglement in one sentence' and return the result

Completion created successfully using claude-3-5-sonnet-20241022. The model returned: 'Quantum entanglement is a phenomenon where two or more particles become correlated such that the quantum state of one particle instantaneously influences the state of the other, regardless of the distance separating them.' This request consumed approximately 180 tokens (24 prompt + 156 completion) and was billed against your organization's metered balance.

This write operation demonstrates cross-provider LLM access through a unified interface. CometAPI meters every generation call — ensure your organization has sufficient funded balance before production use. The MCP does not cache responses or manage conversation history; you must track message threads in your own Project context if building multi-turn workflows.

@CometAPI check my API key usage limits, then create embeddings for these three product descriptions and tell me if any pair is semantically similar

Your API key shows 2.4M tokens remaining in the current quota period (resets in 18 days), with no model restrictions. Embeddings created for all three descriptions using text-embedding-3-small. Comparing cosine similarity: Description A and Description C are highly similar (0.89 similarity score) — both emphasize 'cloud-native scalability' and 'real-time analytics'. Description B is distinct (0.61 to A, 0.58 to C), focusing on compliance and audit workflows instead of performance.

This synthesis example pairs quota awareness with embedding generation and AI-driven comparison. The MCP returns raw vectors; the similarity analysis here is performed by the AI reasoning over those vectors. Embedding operations are metered separately from chat completions — budget accordingly if processing large document sets.

When you need to compare GPT-4, Claude, and Gemini outputs side-by-side

A 3-person growth team runs weekly experiments on landing-page copy. They write one prompt, then fire it at four different models to see which output converts better in A/B tests. CometAPI wins here because List Models shows every provider in one catalog, and Create Chat Completion hits all of them through the same interface—no juggling API keys or rewriting prompts for each vendor's format. The team saves 20 minutes per experiment round by scripting the comparison in a single Switchy thread. Trade-off: if you only ever use one model family, native provider SDKs are simpler. But the moment you want to test across Anthropic, OpenAI, and Google in the same workflow, CometAPI's unified protocol is the faster path.

Filtering user-generated content before it hits your knowledge base

A 6-person support team ingests customer feedback from Slack, email, and a web form into a shared knowledge base. Before any text goes live, they run Moderate Content to flag policy violations or PII leaks. CometAPI's moderation tool uses the OpenAI protocol, so the team writes one filter function that works whether they're moderating a single message or a batch of 50 form submissions. The API key usage check (Get API Key Usage) lets them monitor quota in real time, so they know when to top up balance before the pipeline stalls. This setup works until you're moderating more than 10,000 texts per day—at that scale, you'll want a dedicated moderation service with custom rulesets, not a metered API.

Building a 'find similar' feature without standing up a vector database

A 4-person product team maintains 200 Notion pages of specs, retros, and design decisions. They want a 'find pages like this one' button but don't want to run Pinecone or Weaviate. Create Embeddings generates vectors for every doc, and the team stores them in a simple JSON file or lightweight SQLite table with a cosine-similarity search. CometAPI handles the embedding step through the OpenAI protocol, so switching from text-embedding-3-small to a different provider's model is a one-line config change. This approach breaks down around 5,000 documents—search gets slow, and you'll need a real vector store. But for a few hundred docs and a team that wants semantic search in an afternoon, CometAPI's embedding tool is the pragmatic start.