impossibl
inkling is now live

the ai api

one endpoint. every model.

get started
docs

$ ./impossibl

one string for whatever's next

keep your code. swap a string for upgrades.

integration is a one-liner

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["IMPOSSIBL_API_KEY"],
-   base_url="https://api.openai.com/v1",
+   base_url="https://api.impossibl.com/v1",
)
# new models are just strings
resp = client.chat.completions.create(
    model="",
    messages=[{"role": "user",
        "content": ""}],
)

$ impossibl stats --live

unified experience.

unified usage

tokens per second, every provider

anthropicopenaigoogle

automatic fallbacks

providers degrade. you're always up

latency

p50 and p95, every request

p50

212ms

p95

468ms

one bill

every provider, one balance

$128.40

this month

$ impossibl init

60s to first token.

one prompt for your agent. no reasoning tokens for you.

  1. paste this into your agent

    auth.md has everything your agent needs.

    prompt

    read https://impossibl.com/auth.md and set me up.

  2. let it cook

    new account, api keys and test curl made for you. no dashboard required.

  3. claim your dashboard · optional

    everything runs without it. when you want it, ask your agent to claim the account and follow the url. claiming adds $1 in credits.

    the agent's key keeps working.

$ man impossibl

frequently asked questions

what is the ai api?

one openai-compatible api that sits in front of openai, anthropic, google and xai. you pick the model, we route the request and handle the billing. one key for every provider.

why use the ai api instead of going direct?

one api, one key, one balance for every provider. switching models is just editing a string, and if a provider fails before the first token we retry the next route for you. going direct means a separate sdk, key and invoice per provider, plus a migration every time a better model ships.

how is pricing calculated?

usage is charged at the provider's list price with 0% usage markup. prepaid credit purchases include a 5% platform fee. there is no subscription. usage comes off your balance when the request finishes.

what if i already have credits or a contract with a provider?

just bring your key. add your openai, anthropic, google, xai, bedrock or azure key once and matching requests go through it at zero credit cost, billed straight to your provider account. startup credits and negotiated rates keep working, and you still get one api for everything.

how does agent onboarding work?

paste the setup prompt into your agent. it reads auth.md, creates an ai api account, generates an api key and connects your app. claiming is optional: open the complete claim link it gives you to attach the account to your identity, unlock the dashboard and get an extra $1 in usage credits. the agent's key keeps working after the claim.

my agent created an account for me. how do i claim it?

ask your agent for the complete claim link. open it, sign in and confirm the handoff. the link already includes the 6-digit code, so you don't have to copy it. that attaches the account to your identity, unlocks the dashboard and adds $1 in usage credits — once per person. nothing breaks in the meantime: the agent's api key keeps working before, during and after the claim.

will the ai api work with my existing stack?

almost certainly. we speak openai chat completions, openai responses and anthropic messages, so the openai and anthropic sdks work as is, and so does anything built on top of them, like the ai sdk. moving over is a base url swap, not a rewrite.

how do i integrate?

change one line. point your openai or anthropic sdk at api.impossibl.com/v1 and swap the key. chat completions, responses and the messages api all work unchanged. models are provider/model strings, and get /v1/models lists every one with live pricing.

more in the docs →

ship the impossibl.

paste into your agent