Together AI

Lightweight, dependency-free, in-memory Together AI API fake. Together AI is OpenAI-compatible, so this works with the official together-ai / openai SDKs. All output is deterministic (hash-derived) and SSE streaming is supported.

Default port: 4863

Quick start

import { TogetherAiServer } from "./services/together-ai/src/server.js";

const server = new TogetherAiServer(4863);
await server.start();
// ... run your app/tests ...
await server.stop();

Point the openai SDK at it:

import OpenAI from "openai";

const client = new OpenAI({ apiKey: "parlel-together", baseURL: "http://127.0.0.1:4863/v1" });
const res = await client.chat.completions.create({
  model: "meta-llama/Llama-3.3-70B-Instruct-Turbo",
  messages: [{ role: "user", content: "hello" }],
});

Access via MCP / preview URL

Base URL: http://127.0.0.1:4863/v1
Health: GET /health → { "status": "ok" }
Auth: Authorization: Bearer <key> (any non-empty token).

Implemented operations

POST /v1/chat/completions — OpenAI-compatible chat. Supports stream: true (SSE, data: [DONE]).
POST /v1/completions — legacy text completion.
POST /v1/embeddings — deterministic 768-dim embeddings.
POST /v1/images/generations — deterministic base64 PNG images.
GET /v1/models — model catalog (vendor-prefixed ids).

Service & inspection operations (parlel extensions)

GET / / GET /health / POST /__parlel/reset / GET /__parlel/requests.

Surface coverage

This emulator faithfully replicates the API surface most application code and agents exercise. Anything below the supported lines is either an intentional design choice for a fast, zero-cost local emulator (✓ By design) or a candidate for a future release (⟳ Roadmap) — never a silent inaccuracy.

Legend: ✅ fully supported · ◐ accepted (stored, not strictly enforced) · ✓ by design · ⟳ on the roadmap.

Feature	Status
`chat.completions` (non-stream + SSE)	✅ Supported
`completions` (legacy)	✅ Supported
`embeddings`	✅ Supported
`images/generations`	✅ Supported
`models` list	✅ Supported
Deterministic, reproducible output	✅ Supported
Real model inference	✓ By design — Deterministic stub output — repeatable assertions, no API spend
Fine-tuning / files / batch jobs	⟳ Roadmap
Tool/function calling, JSON mode	◐ Accepted, not specially handled

Error codes & shapes

Errors use the OpenAI envelope: { "error": { "message", "type", "code" } }.

Status	When
`401`	missing/invalid `Authorization`
`400`	missing `model`/`messages`/`prompt`/`input` or bad JSON
`404`	unknown endpoint

Manifest

See services/together-ai/manifest.json:

name: together-ai, port: 4863, protocol: http, healthcheck: /health, startup ≈ 100ms
env: TOGETHER_API_KEY, TOGETHER_BASE_URL

Configuration — `test.env`

Copy these into your test.env (used by the bridge sidecar flow). Tokens are Parlel's seeded test credentials — any non-empty value is accepted by the emulator, so you rarely need to change them. Swap in real credentials only when pointing at the live service in prod.env.

TOGETHER_API_KEY=parlel-together
TOGETHER_BASE_URL=http://parlel-bridge:4863/v1