Universal inference API · 85+ labs, one schema

One API.
Every model.
Every modality.

Point the OpenAI SDK at one base_url and reach 1050+ models across every modality. Agents autoconfigure from /llms.txt.

Get your key →Read the docs
● 99.99% uptime● 47ms p50 overhead● SOC 2 in progress
python
typescript
curl
aigateway-py
aigateway-cli
quickstart.py
# one-line swap: point your SDK at us.
from openai import OpenAI

client = OpenAI(
    base_url="https://api.openai.com/v1"
             "https://api.aigateway.sh/v1",
    api_key="sk-aig-...",
)

stream = client.chat.completions.create(
    model="zai-org/glm-5.3-flash",
    messages=msgs, stream=True,
)
STREAMINGmodel: kimi-k2.7-codetok/s: 0ttft: —cached −50%
+ 27 more
TextImageVideoAudioVoiceMusicVisionEmbeddings
MODELS
0+
from 85+ labs
LABS
0+
normalized behind one schema
ADDED LATENCY
0ms
added per request
CACHE DISCOUNT
0%
off list price on cached requests
NEWmodel: "auto"
Routes every request to the cheapest model that clears your quality bar.
Set model:"auto" and you never pay more than the premium model you'd have called yourself. Works across text, image, video, speech, transcription, music, and embeddings — and every response shows what ran and what you saved.
See the Auto Router →
Capabilities

Everything behind
one key.

The whole surface of a modern inference stack — modalities, routing, keys, caching — normalized behind a single OpenAI-compatible API.

Every modality

Text, image, video, audio, voice, music, embeddings, vision — one schema across all of them.

Drop-in for the OpenAI SDK

Change one base_url and every existing integration reaches 1000+ models unchanged.

Auto model routing

model:"auto" picks the cheapest model that clears your quality bar, every request.

Sub-account keys

Mint scoped keys per end user with their own spend caps, rate limits, and analytics.

Semantic caching

Exact and near-duplicate requests return cached — 50% off list price, near-zero latency.

Agent-ready

A capable agent reads /llms.txt or the MCP server once and configures itself.

LAUNCH OFFER · 3x UNDER $100 · UNMETERED FLASH AT $100

Triple your credit.
Or stop paying for Flash.

Through September 30, 2026, top up $5-$99 and your credit is tripled — $5 holds $15, $50 holds $150. The bonus is spendable on GLM-5.3 Flash, expires 90 days after the top-up, and is capped at $100 of bonus per account. Put down $100 in one payment instead and Flash stops being metered for you through September 30, 2026, while the $100 stays ordinary credit. No code, no signup: top up and it lands on your key.

3x
launch_offer3x to $100 bonusunmetered at $100GLM-5.3 Flash
Triple credit under $100. Unmetered Flash at $100.
Top up $5-$99 and the bonus credit triples what you hold — spendable on GLM-5.3 Flash only, a 320B-parameter multimodal model at $0.15 / $0.50 per 1M tokens, expiring 90 days after the top-up, up to $100 of bonus per account. Put down $100 in one payment instead and Flash stops being metered entirely through September 30, 2026, while the $100 stays ordinary credit for the full 1050+-model catalog.
TOP UP $5
+$10 bonus
$15 to spend
TOP UP $25
+$50 bonus
$75 to spend
TOP UP $50
+$100 bonus
$150 to spend
TOP UP $100
Flash unmetered
balance stays yours
Top up and claim →See GLM-5.3 Flash3x ends September 30, 2026 · unmetered through September 30, 2026
python
first_call.py
from openai import OpenAI

client = OpenAI(
    base_url="https://api.aigateway.sh/v1",
    api_key="sk-aig-...",
)

r = client.chat.completions.create(
    model="zai-org/glm-5.3-flash",
    messages=[{"role": "user",
               "content": "Plan a"
                          " research agent."}],
    stream=True,
)

# Top up $5 → hold $15: 3x credit on GLM-5.3 Flash. Or $100 in
# one payment → Flash unmetered. Both through September 30.
Catalog

1050+ models.
One key.

Drop-in for the OpenAI SDK. Rotate models mid-conversation, one string.

Anthropic logoAnthropicOpenAI logoOpenAIGoogle logoGooglexAI logoxAIMeta logoMetaMistral logoMistralDeepSeek logoDeepSeekMoonshot logoMoonshot
showing 12 of 1050+ models
Rankings →Providers →See all 1050+ models →
Popular comparisons:Opus 4.7 vs GPT-5.4Sonnet 4.6 vs GPT-5.4-miniGemini 3.1 Pro vs Opus 4.7Kimi K2.7 Code vs Opus 4.7Haiku 4.5 vs GPT-5.4-mini
Comparison

Why swap in
one line of code.

Same public pricing and feature matrix as the named alternatives, competitors anonymized.

 AIgatewayCompetitor A
breadth aggregator
Competitor B
platform-native gateway
Competitor C
enterprise governance
Models / modalities1050+ · text · image · video · music · voice · audio · embeddings · vision~300 · text only~40 · text + image~80 · image, video, audio
Open-weight, served first-partyyesno · pass-throughno · pass-throughno · queue
Latency added (p50)47ms~180ms · single region~60ms · one platform~190ms · single region
Eval-driven routing (SLO on your data)yes———
Sub-account / per-user key APIyes · programmatic—workspace only—
Replay + shadow A/B across modelsyes———
OpenAI-compatibledrop-in, zero changesdrop-indrop-in—

Sourced from public pricing pages and docs, April 2026. Spot an error? hello@aigateway.sh.

Primitives

Four things a single-provider
SDK can't do.

Eval-routing, scoped sub-account keys, cross-model replay, and per-feature cost tags — each needs a view across every provider, not one.

EVAL-DRIVEN ROUTING

Let your own data pick the model.

Grade every model on your own data. The alias always routes to the current winner — new model lands, rerun, prod code unchanged.
curl -X POST https://api.aigateway.sh/v1/evals \
  -H "Authorization: Bearer $KEY" \
  -d '{
    "name": "prod-summarize",
    "candidate_models": [
      "anthropic/claude-opus-4.7",
      "openai/gpt-5.4",
      "moonshot/kimi-k2.7-code"
    ],
    "dataset": [...],
    "metric": "quality"
  }'

# then just use it
model = "eval:prod-summarize"
SUB-ACCOUNT API

Scoped keys for your end users.

One call mints a per-customer key with its own spend cap, rate limit, and analytics — so you can bill your own users.
POST /v1/sub-accounts
{
  "name": "acme-corp",
  "spend_cap_cents": 50000,     // $500 / mo hard cap
  "rate_limit_rpm": 300,
  "default_tag": "acme"
}

=> { "key": "sk-aig-...",       // hand to customer
     "spend_cap_cents": 50000,
     "id": "sa_9f3k..." }
REPLAY + SHADOW A/B

Test a new model on real traffic.

Replay any past request on a new model — output, cost, latency side by side. Shadow mode tests it on live traffic without touching the user.
POST /v1/replays
{
  "source_request_id": "req_abc123",
  "target_model": "anthropic/claude-opus-4.7",
  "shadow": true
}

=> { "source_output": "...",
     "target_output": "...",
     "cost_source_cents": 1.2,
     "cost_target_cents": 4.7,
     "score_delta": 0.82 }
COST-ATTRIBUTION TAGS

Know what every feature costs.

Tag any request (feature, user, tenant) and query spend by tag. Hard per-tag budget caps stop runaway costs.
# tag each request with the feature it powers
curl https://api.aigateway.sh/v1/chat/completions \
  -H "x-aig-tag: summarize" \
  ...

GET /v1/usage/by-tag?month=2026-04
=> [{ "tag": "summarize", "cost_cents": 4210 },
    { "tag": "chat",      "cost_cents": 9830 },
    { "tag": "rerank",    "cost_cents":  118 }]
For coding agents

Point Claude Code at one URL.
It configures itself.

A capable coding agent reads /llms.txt, /openapi.json, or /agents.md once — then it calls any model (text to video), mints sub-accounts, tags cost, and replays requests, no human editing config.

You can call any AI model through AIgateway, a universal AI API.

- Base URL: https://api.aigateway.sh/v1
- Auth: Authorization: Bearer sk-aig-...
- SDK: drop-in for the OpenAI SDK — only change base_url
- OpenAPI spec: https://api.aigateway.sh/openapi.json
- Capability map: https://aigateway.sh/llms.txt
- Live models: https://api.aigateway.sh/v1/models
- MCP server: https://api.aigateway.sh/mcp

Install (only when the OpenAI SDK isn't enough):
  pip install aigateway-py        # Python — async jobs, sub-accounts, evals
  pnpm add aigateway-js           # Node   — same surface in TypeScript
  npm i -g aigateway-cli          # CLI    — `aig init` walks through everything

Primitives nobody else has:
  POST /v1/sub-accounts — scoped keys + spend caps per end user
  POST /v1/evals        — grade candidate models on your data
  POST /v1/replays      — re-run a past request on a new model
  GET  /v1/usage/by-tag — per-feature cost via x-aig-tag header
LLMS.TXT
Agent-readable capability map
/llms.txt
OPENAPI.JSON
Typed 3.1 spec for code-gen
https://api.aigateway.sh/openapi.json
AGENTS.MD
Integration patterns + error remediation
/agents.md
INSTALL
pip install aigateway-py · pnpm add aigateway-js · npm i -g aigateway-cli
/integrations
Full agent guide →

Products building using AIgateway.

FlareCode logo
FlareCode
Coding agents
AudioPod logo
AudioPod
Audio AI
NameMyApp logo
NameMyApp
Branding AI
Go2 logo
Go2
Agent infra
MailMolt logo
MailMolt
Agent email
ClawOcean logo
ClawOcean
Agent platform
ChargeFind logo
ChargeFind
Consumer AI
Public beta · April 2026

Start in 30 seconds.
No waitlist, no sales call.

Cline, Aider, and a handful of agent startups already build on it. Drop your email and a real person helps you get set up.

one email, from a real person. no drip sequence.