Multi-project · OpenAI-compatible · Language, audio, image, and embeddings

One governed API for your AI stack.

Sophy puts supported models behind project-scoped keys. Each project brings its own Vercel AI Gateway credential while its team governs prompts, models, schemas, knowledge, logging, and limits from one console.

Read the docs Sign in or create an account

Each key is bound to one primary model; use an API route that matches its capabilities.

from openai import OpenAI

client = OpenAI(
    base_url="https://sophy.in/v1",
    api_key="mw_live_…",
)

resp = client.chat.completions.create(
    model="sophy",  # the key selects the model
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

from openai import OpenAI

client = OpenAI(
    base_url="https://sophy.in/v1",
    api_key="mw_live_…",  # bound to a transcription model
)

with open("meeting.m4a", "rb") as audio:
    resp = client.audio.transcriptions.create(
        model="sophy",
        file=audio,
        language="en",
    )
print(resp.text)

from openai import OpenAI

client = OpenAI(
    base_url="https://sophy.in/v1",
    api_key="mw_live_…",  # bound to an embedding model
)

resp = client.embeddings.create(
    model="sophy",
    input=["First document", "Second document"],
)
print(resp.data[0].embedding)

from openai import OpenAI

client = OpenAI(
    base_url="https://sophy.in/v1",
    api_key="mw_live_…",  # bound to an image model
)

resp = client.images.generate(
    model="sophy",
    prompt="A monsoon city at blue hour, watercolor",
    size="1024x1024",
)
print(resp.data[0].b64_json[:32])

Seven supported API routes

One base URL, purpose-built surfaces

Review compatibility details

POST/v1/chat/completions

Chat + streaming

POST/v1/responses

Responses + tools

POST/v1/audio/transcriptions

Raw or policy-processed text

POST/v1/embeddings

Float or base64 vectors

POST/v1/images/generations

Inline image output

POST/v1/files

Key-scoped uploads

GET/v1/models

The key’s bound model

One identity, isolated projects

Your work starts with the right project

Sophy keeps membership, policy, data, usage, and provider spend inside each project, while giving you one account and a personal default.

Start in My Project

Join through an invitation

Your invited project becomes the default when it is your first Sophy project, with the role its Admin assigned.

Choose your default

Use one identity across projects, switch without signing out, and choose which project opens first.

Bring a Gateway per project

Every project connects its own Vercel AI Gateway credential before its Sophy keys can make paid model calls.

Build and govern

Keep clients simple. Keep control central.

Sophy separates application input from operator policy, while preserving the tool, streaming, multimodal, and structured workflows modern AI products need.

A familiar API surface

Use supported OpenAI SDK methods for Chat Completions, Responses, audio transcription, embeddings, image generation, models, and files.

Policy lives on the key

Pin a primary model and centrally own its prompt, optional transcript processor, parameters, schema, knowledgebase, logging policy, budget, and rate limit.

Tools and agent workflows

Run standard function-tool loops, or opt trusted server-side agents into per-request instructions while the key’s policy stays first.

Portable structured output

Attach a JSON Schema to a key for schema-constrained output, normalized across supported language models and validated on completion.

Managed knowledgebases

Ingest PDF, DOCX, Markdown, text, CSV, and JSON sources once, share them across keys, and ground Chat or Responses automatically.

A searchable model catalog

Compare supported language, transcription, image, and embedding models by provider, capability, context window, and estimated pricing.

Budgets and rate limits

Enforce per-key requests-per-minute limits and monthly cost caps before paid model calls reach the provider.

Key-scoped multimodal files

Stage short-lived images and documents, then reference them in multimodal requests with cross-key reuse blocked inside Sophy.

How it works

From integration to evidence in three steps

Configuration is read from the key on every request, so policy changes and model switches apply without a client release.

Connect your project

A Project Admin adds the project’s Vercel AI Gateway key. Sophy validates it without making a generation call and stores it encrypted.

Configure a Sophy API key

Choose a primary model and set its prompt policy, parameters, limits, logging, schema, optional transcript processor, and optional project knowledgebase.

Integrate and improve

Use the OpenAI SDK with the Sophy key, then trace project usage, evaluate challengers, and update policy without redeploying clients.

Operate and optimize

Choose models with production evidence

Sophy is a control plane, not just a proxy. Observe real workloads, compare a challenger without changing the client response, and attribute client, evaluation, and knowledgebase spend separately.

Production model evaluations

Shadow successful live text requests to a challenger, score outputs with a blinded judge, and compare quality, cost, and latency before switching.

Usage analytics and request tracing

Explore requests, tokens, estimated cost, latency, model mix, errors, and spend by source—with optional content capture where supported.

Project-isolated teams and spend

Give one identity an Admin or Editor role per project. Each project keeps its own members, Sophy keys, knowledgebases, telemetry, and Vercel Gateway credential.

Mature key operations

Rotate secrets in place, revoke immediately, search active and revoked keys, and apply model changes or start evaluations in bulk.

Blind quality judgment

Randomized response order reduces position bias.

Cost and latency per task

Compare operational tradeoffs beside the verdict.

Recommendation, not auto-switch

Operators stay in control of the final model change.

Governed by default, agent-ready by choice

Let trusted agents bring dynamic instructions

Standard keys drop client prompt instructions. Enable Agent mode only for trusted server-side apps, and Sophy appends their leading instructions after the key-owned prompt while model, parameters, limits, and observability remain centralized.

Read the prompt policy

from openai import OpenAI

client = OpenAI(base_url="https://sophy.in/v1", api_key="mw_live_…")

# On a trusted key with Agent mode enabled, these instructions are
# appended after the centrally managed key prompt.
resp = client.responses.create(
    model="sophy",
    instructions="Plan carefully and explain the final decision.",
    input="Compare the two launch options.",
)
print(resp.output_text)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://sophy.in/v1",
  apiKey: "mw_live_…",
});

const resp = await client.responses.create({
  model: "sophy",
  instructions: "Plan carefully and explain the final decision.",
  input: "Compare the two launch options.",
});
console.log(resp.output_text);

Ready to put your AI stack under control?

Read the compatibility guide or sign in to manage keys, knowledge, usage, and evaluations.

Read the docs Sign in or create an account