Live — Available Now

Code Embeddings That Understand Code.

PyckLM is trained on code structure — not general text. Better retrieval, better similarity, better results for any codebase application.

View API

Not a generic embedding model.

Most embedding APIs treat code like prose. PyckLM is fine-tuned on real codebases — it understands function signatures, call patterns, docstrings, and the semantic relationships between code units.

The result: more relevant retrieval, better similarity scoring, and context that actually makes sense to the AI reading it.

Trained on code structure, not just text tokens
512-dimensional vectors, drop-in compatible
Hybrid BM25 + semantic fusion built in
Usage-based — pay only for what you call
Retrieval accuracy PyckLM vs generic
PyckLM (code-tuned) 91%
Generic text model 63%

Top-5 retrieval accuracy on held-out code benchmarks.

Simple REST API

One endpoint. Accepts text or code. Returns a PyckLM embedding vector.

Python cURL Node.js
import requests

response = requests.post(
    "https://api.pyckle.co/v1/embed",
    headers={
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json",
    },
    json={
        "text": "authentication middleware for JWT tokens",
        "model": "pycklelm-v2",
    }
)

data = response.json()
embedding = data["embedding"]   # list[float], 512 dimensions
tokens    = data["tokens_used"]  # billed per token

512

dimensions

<80ms

typical latency

REST

any stack, any language

What you can build

PyckLM embeddings are the foundation for any application that needs to understand code at semantic depth.

Semantic Code Search

Build search that finds code by meaning — "find where we handle auth errors" returns the right files even when the words don't match exactly.

Similarity Scoring

Score how similar two code snippets are — find duplicates, detect refactoring candidates, or surface related implementations across repos.

RAG Pipelines

Power retrieval-augmented generation with code context that AI models actually understand — precise chunk retrieval, fewer hallucinations.

Simple, tiered pricing

Start free — no credit card required.

Monthly
Yearly Save 17%

Free

$0 forever
  • 100 requests / month
  • 10 requests / minute
  • 512-dim L2-normalized vectors
  • No credit card required
Get Free Key on MCPize
Recommended

Pro

$100 /year

~$8.33/month

  • 50,000 requests / month
  • 60 requests / minute
  • 512-dim L2-normalized vectors
  • Priority support
or subscribe via MCPize

Endpoint: https://api.pyckle.co/mcp/embeddings · Hosted on Fly.io

Start building today

Get your API key instantly. Free tier requires no credit card.

Free Key on MCPize

Questions? Email [email protected]