{ code }

Highly specialized auxiliary coding models designed to make AI codegen products faster and more reliable.

Edit Snippet
Response
TTFT: 541 ms
TPS: 2504 tok/s
Models

Custom SOTA models for coding workflows

Relace models are trained in-house to outperform frontier LLMs on specific steps in autonomous codegen. Our retrieval and code merging models are already state-of-the-art, and they continue to improve with each training run.

Enqueued
Startup
Execution
Status
09:34:22 AM
0.0s
-
 
09:34:19 AM
0.0s
0.0s
 
09:34:15 AM
0.0s
2.2s
 
09:34:14 AM
0.0s
1.9s
 
09:34:12 AM
0.0s
1.7s
 
09:34:10 AM
0.0s
2.1s
 
Deployments

Optimized deployments for fast inference

Each model is hosted with a custom inference engine, tuned to achieve nearly instantaneous output. Our code merging model runs at >2500 tok/s, and our retrieval models can search an entire codebase in <1s.

Integration

A competitive edge starts with only a few extra lines of code

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
import requests

url = "https://instantapply.endpoint.relace.run/v1/code/apply"
headers = {
    "Content-Type": "application/json", 
    "Authorization": "Bearer [YOUR_API_KEY]"
}

initial_code = """def hello_world():
    # TODO: Implement Hello World
    pass"""

edit_snippet = """// ... keep existing code
    print("Hello, world!")
// ... keep existing code"""

data = {
    "initialCode": initial_code,
    "editSnippet": edit_snippet
}
from relace import OpenAI client = OpenAI( api_key= "OPENAI_API_KEY", relace_api_key= "RELACE_API_KEY", ) messages = [ { "role": "system", "content": "..." }, {"role": "user", "content": "..." } ] completion = client.chat.completions.create( model="gpt-4o", messages= messages, )
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
import requests

url = "https://ranker.endpoint.relace.run/v1/code/rank"
headers = { "Content-Type": "application/json", "Authorization": "Bearer [YOUR_API_KEY]" }

query = "Change hello world to goodbye world."
codebase = [
  { "filename": "greeting.py", "code": "print('hello world')" },
  ...
]

data = { "query": query, "codebase": codebase, "token_limit": 128000 }

response = requests.post(url, headers=headers, json=data)
ranked_code = response.json()
import OpenAI from 'relace'; const client = new OpenAI({ apiKey: "OPENAI_API_KEY", relaceApiKey: "RELACE_API_KEY" }); const messages = [ { role: "system", content: "..." }, { role: "user", content: "..." } ]; const completion = await client.chat.completions.create({ model: "gpt-4o", messages: messages });
OUR INVESTORS
Y Combinator logoPioneer Fund LogoAmino capital logoL2 Ventures logo
OUR INVESTORS
Testimonials

Real Feedback, Real Results

Just wanted to say that Relace’s fast rewriting model has been a big boon to our product. It’s made edits much more reliable with hardly any visible downsides. A whole class of bugs is gone for us now.

James Grugett
Co-Founder & CEO of Codebuff

Relace has been a critical tool for us to create custom, design-focused AI models and allowed us to fine tune on our own data and continuously create better and better models.

Teddy Ni
Co-Founder of MagicPatterns

Do you guys ever sleep?

Anton Osika
Co-Founder at Lovable

this is definetely a lot faster!

Junior Garcia
Founder & CEO @hero_ui

Get Started in Minutes

Try out our playground, and start with our free tier to test Relace models in your application.