GPT-4o
TextContext: 128K
Iris brings the world’s leading text, image, video, and audio models into one interface. Browse by capability, connect with IrisKEY, and switch models without rewriting your workflow.
Iris is the AI aggregation layer of EASTAR. Instead of juggling a dozen dashboards and SDKs, you connect the models you already trust — OpenAI, Anthropic, Google, Meta, DeepSeek, Mistral, and more — through a single, OpenAI-compatible interface.
With IrisKEY your credentials stay on your device, EASTAR never takes a cut of your AI usage, and you can swap providers at any time. The catalog below links straight to each provider’s official documentation and pricing, so you always see the source of truth.
Browse by modality. Every card links to the provider’s official docs and pricing — EASTAR does not resell tokens or mark up model prices. Connect any model with IrisKEY inside Iris.
Context: 128K
Context: 128K
Context: 200K
Context: 200K
Context: 200K
Context: 200K
Context: 2M
Context: 1M
Context: 128K
Context: 128K
Context: 128K
Context: 128K
Context: 128K
Context: 256K
Context: 128K
Iris is the aggregation layer inside a wider family of EASTAR products. Each surface plugs into the same models and the same IrisKEY credentials.
One OpenAI-compatible interface for every leading AI model. Bring your own keys with IrisKEY and switch providers at any time.
MessagingEnterprise instant messaging on WhatsApp and LINE — one team inbox with an AI copilot behind every reply.
MarketplaceCurated skills and MCP servers that teach Iris new capabilities and connect it to your tools, repos, and data.
IrisKEY is how Iris handles credentials: you bring your own provider key, it stays on your device, and you pay each lab directly.
You contract directly with each model provider and pay them at their published rates. EASTAR is not a reseller: we never see your billing, never add a margin, and never proxy your credentials through our servers.
Because keys live on your device, you keep full control over access, spending limits, and revocation. Remove a key and the connection stops immediately — no waiting on a support ticket.
Provider keys are stored on your device inside Iris and are never uploaded to EASTAR servers.
Pay each provider at their published rates. EASTAR never resells tokens or marks up model prices.
Set spending limits, swap providers, or revoke a key at any time. Removing a key stops calls immediately.
Add a key per provider and switch models freely through one OpenAI-compatible endpoint.
Iris speaks the OpenAI-compatible API, so it drops straight into the tools developers already use — SDKs, editors, agents, and MCP clients.
Point the OpenAI Python or JavaScript SDK at your Iris endpoint. Change the base URL and model name; keep the rest of your code.
Connect Iris to Qoder, WorkBuddy, and other MCP clients so agents can read repos, run tests, and ship changes.
Call the OpenAI-compatible endpoints directly over HTTPS — ideal for scripts, CI pipelines, and serverless functions.
Add Iris as a custom OpenAI-compatible provider in your editor or agent, and switch models per task without reconfiguring.
Drop-in settings for MCP clients and OpenAI-compatible IDEs and coding agents. Swap in your own endpoint and key.
{
"mcpServers": {
"pai": {
"url": "https://your-pai-endpoint/mcp",
"headers": {
"Authorization": "Bearer YOUR_PROVIDER_KEY"
}
}
}
}{
"provider": "openai-compatible",
"baseUrl": "https://your-pai-endpoint/v1",
"apiKey": "YOUR_PROVIDER_KEY",
"model": "claude-3-5-sonnet"
}One aggregation layer powers many jobs. These are the patterns Iris customers run every day — each billed to your own provider account.
Route WhatsApp, Telegram, and LINE conversations to an AI agent that answers in your brand voice, escalates to a human when needed, and logs every interaction.
Draft with a text model, illustrate with an image model, and cut a teaser with a video model — all from the same interface.
Feed long documents to a large-context model, extract structured data with JSON schema output, and summarize findings across languages.
Translate and localize customer conversations, product listings, and support docs with multilingual models, keeping tone consistent across markets.
Transcribe calls and meetings with speech-to-text, then generate natural voiceovers with text-to-speech — across the audio models in the catalog.
Iris speaks the OpenAI-compatible API. Point your existing SDK at Iris, add your provider key with IrisKEY, and choose a model — no rewrite required.
Paste your own API key from OpenAI, Anthropic, Google, or any supported provider. With IrisKEY, keys are stored locally on your device and never uploaded to EASTAR servers.
Choose any model from the catalog by name. Switch between providers and modalities at any time — your prompts and workflows stay the same.
Use the OpenAI-compatible chat, images, audio, and video endpoints. Existing OpenAI SDK code works by changing the base URL and model name.
from openai import OpenAI
# Iris is OpenAI-compatible. Point the SDK at the endpoint shown
# in Iris and use your own provider key (IrisKEY).
client = OpenAI(
base_url="https://your-pai-endpoint/v1",
api_key="YOUR_PROVIDER_KEY",
)
resp = client.chat.completions.create(
model="gpt-4o", # any model from the catalog
messages=[{"role": "user", "content": "Hello, Iris!"}],
)
print(resp.choices[0].message.content)import OpenAI from "openai";
// Iris is OpenAI-compatible. Point the SDK at your Iris endpoint
// and use your own provider key (IrisKEY).
const client = new OpenAI({
baseURL: "https://your-pai-endpoint/v1",
apiKey: "YOUR_PROVIDER_KEY",
});
const resp = await client.chat.completions.create({
model: "gpt-4o", // any model from the catalog
messages: [{ role: "user", content: "Hello, Iris!" }],
});
console.log(resp.choices[0].message.content);# Iris is OpenAI-compatible. Call the endpoint with your own
# provider key (IrisKEY). Your keys stay on your device.
curl https://your-pai-endpoint/v1/chat/completions \
-H "Authorization: Bearer YOUR_PROVIDER_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-4o",
"messages": [{"role": "user", "content": "Hello, Iris!"}]
}'How credentials travel — and how they never reach us.
Every request carries your own provider key in the standard Authorization: Bearer header. Iris issues no shared keys and does not proxy your credentials through EASTAR servers: the key you paste into Iris stays on your device and is sent only to the provider endpoint you configured.
Because authentication is per-provider, you can hold keys for several labs at once and switch models freely. Revoking a key at the provider immediately stops that model; nothing needs to change on our side.
Treat keys like passwords: load them from environment variables or your OS keychain, never commit them to source control, and rotate them on a schedule. The examples below read the key from an environment variable.
Read the key from the environment instead of hard-coding it.
import os
from openai import OpenAI
client = OpenAI(
base_url=os.environ["PAI_BASE_URL"],
api_key=os.environ["PAI_PROVIDER_KEY"],
)import OpenAI from "openai";
const client = new OpenAI({
baseURL: process.env.PAI_BASE_URL,
apiKey: process.env.PAI_PROVIDER_KEY,
});curl "$PAI_BASE_URL/chat/completions" \
-H "Authorization: Bearer $PAI_PROVIDER_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"gpt-4o","messages":[{"role":"user","content":"Hi"}]}'OpenAI-compatible paths. Replace the host with the endpoint shown in Iris.
| Method | Path | Purpose |
|---|---|---|
| POST | /v1/chat/completions | Text & multimodal chat, reasoning, tool calling |
| POST | /v1/embeddings | Vector embeddings for search & RAG |
| POST | /v1/images/generations | Image generation from a prompt |
| POST | /v1/audio/transcriptions | Speech-to-text for calls & meetings |
| POST | /v1/audio/speech | Text-to-speech voiceovers |
| GET | /v1/models | List the models available to your key |
The common subset accepted across providers; provider-specific extras pass through.
| Parameter | Type | Default | Description |
|---|---|---|---|
| model | string | — | Catalog model name, e.g. gpt-4o or claude-3-5-sonnet |
| messages | array | — | Ordered role/content turns (system, user, assistant, tool) |
| temperature | number | 1 | Sampling randomness; lower = more deterministic |
| top_p | number | 1 | Nucleus sampling; tune instead of temperature, not both |
| max_tokens | integer | provider | Upper bound on generated tokens |
| stream | boolean | false | Emit server-sent events as tokens arrive |
| stop | string|array | null | Sequences that end generation |
| tools | array | null | Function/tool schemas for tool calling |
Set stream=true and consume server-sent events for low perceived latency.
stream = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Explain Iris in one line."}],
stream=True,
)
for chunk in stream:
delta = chunk.choices[0].delta.content
if delta:
print(delta, end="", flush=True)const stream = await client.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: "Explain Iris in one line." }],
stream: true,
});
for await (const chunk of stream) {
process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}curl -N "$PAI_BASE_URL/chat/completions" \
-H "Authorization: Bearer $PAI_PROVIDER_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"gpt-4o","stream":true,
"messages":[{"role":"user","content":"Explain Iris in one line."}]}'Standard HTTP semantics; retry only what is safe to retry.
| HTTP | Meaning | What to do |
|---|---|---|
| 400 | Bad request / invalid parameter | Fix the payload; do not retry unchanged |
| 401 | Missing or invalid key | Check the IrisKEY value and provider; re-paste the key |
| 403 | Key lacks access to this model | Enable the model at the provider or pick another |
| 404 | Unknown model or path | Check the catalog name and endpoint path |
| 429 | Rate or quota limit | Back off exponentially; respect Retry-After |
| 500 | Provider internal error | Retry with backoff; switch model if persistent |
| 503 | Provider overloaded / unavailable | Retry later or route to a fallback model |
Limits are set by each provider against your own account, not by EASTAR. Iris adds no reseller quota and no markup, so the throughput you get is the throughput your provider plan grants.
On 429, back off exponentially (for example 1s, 2s, 4s, 8s) and honour any Retry-After header. Keep concurrency modest and batch embeddings rather than firing one request per document.
For production traffic, configure a fallback model in the same modality so a single provider outage degrades gracefully instead of failing the whole workflow.
Pick by job, not by brand: a reasoning model for analysis and planning, a fast inexpensive model for high-volume classification and drafting, a large-context model for long documents, and dedicated image/audio models for media.
The catalog lists each model's capabilities and context window and links to the provider's official documentation and pricing, so you can verify limits and cost before you commit.
Because the interface is uniform, switching is a one-line change of the model name. Keep prompts and evaluation sets stable so you can compare models fairly.
The same client handles non-chat modalities.
emb = client.embeddings.create(
model="text-embedding-3-large",
input=["Iris aggregates models", "IrisKEY keeps keys local"],
)
print(len(emb.data[0].embedding))
img = client.images.generate(
model="gpt-image-1",
prompt="A red seal stamp on white paper, minimal",
)
print(img.data[0].url)curl "$PAI_BASE_URL/embeddings" \
-H "Authorization: Bearer $PAI_PROVIDER_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"text-embedding-3-large",
"input":["Iris aggregates models"]}'IrisKEY means your credentials never leave your device except to reach the provider you configured. EASTAR cannot read, store, or reuse your keys, and never sees your billing.
If a key leaks, revoke it at the provider immediately. Because access is per-provider, revocation is instant and does not affect your other models.
Any OpenAI-compatible SDK works: set the base URL and your key, then call chat, embeddings, images, or audio as usual. The official OpenAI Python and JavaScript SDKs are the shortest path.
For agent workflows, connect Iris through MCP tool servers from the Skill Marketplace so your assistant can call models and tools over a standard protocol.
In IDEs and coding agents, add Iris as a custom OpenAI-compatible provider and select a coding model per task; the configuration example in Coding tools shows the exact fields.
How Iris aggregation, IrisKEY, and limits work.
No. Iris is an aggregation layer. It connects you to models from OpenAI, Anthropic, Google, Meta, DeepSeek, Mistral, and others through one OpenAI-compatible interface. The models are owned and operated by their respective labs.
Usually not. Iris exposes OpenAI-compatible endpoints, so existing OpenAI SDK code works by changing the base URL and the model name. Provider-specific features may need small adjustments.
The catalog lists publicly available models and links to each provider’s official documentation, which is always the source of truth for the latest versions, context windows, and capabilities.
IrisKEY is EASTAR’s bring-your-own-key model. You connect your own API key from each provider, contract directly with them, and pay their published rates. Your keys stay on your device — EASTAR never resells tokens, adds a margin, or proxies your credentials through our servers.
On your device, inside Iris. Keys are not uploaded to EASTAR servers. Removing a key stops the connection immediately.
Yes. Add a key per provider and switch models freely. Each call is billed to the corresponding provider account you own.
Each model provider sets its own rate limits, quotas, and availability under your account with them. Iris does not impose an additional usage cap on top of your provider plan.
Because you can switch models at any time, you can route work to another provider while one is unavailable. Iris surfaces provider errors so you can react quickly.
Your prompts go to the model provider you choose, under your key and their data terms. EASTAR does not proxy your credentials through our servers. Review each provider’s data policy for how they handle inputs and outputs.
Yes. Remove the key in Iris, or revoke it in the provider’s dashboard. Either action stops further calls immediately.
Email [email protected] or use the contact page. Tell us which models and providers your team needs and we will prioritize them in the catalog.
Iris is free during the public beta. You still pay each model provider directly for the AI usage under your own key. See the Terms of Use for the beta license.
Go deeper on the models, the aggregation layer, and the ecosystem.
One interface for every AI model — connect with IrisKEY and switch providers without rewriting your workflow.
Browse the model catalog →Curated skills and MCP servers that teach Iris new capabilities and connect it to your tools and data.
Browse the marketplace →Every model card links to its lab’s official docs and pricing. Start with the OpenAI platform docs for the compatible API surface.
Open provider docs →Tell us which models and providers your team needs — we will help you connect them with IrisKEY inside Iris.
Iris aggregates third-party AI models. EASTAR ASSET MANAGEMENT LIMITED is not affiliated with OpenAI, Anthropic, Google, Meta, DeepSeek, Mistral AI, Alibaba Cloud, Stability AI, Black Forest Labs, Runway, Luma AI, Kuaishou, or ElevenLabs. All model names, trademarks, and pricing belong to their respective owners and are set by each provider.