One API key for
every free AI model.
API Keychain unifies Gemini, Groq, Cerebras, Mistral, DeepSeek and more behind a single OpenAI-compatible endpoint. It routes by effort tier, fails over on rate limits, and never makes you juggle keys again.
OpenAI SDK drop-in Keys encrypted at rest No card required
from openai import OpenAI
client = OpenAI(
base_url="https://api.apikeychain.dev/v1",
api_key="ak-•••••••••••••••",
)
# keychain-low | keychain-medium | keychain-high
resp = client.chat.completions.create(
model="keychain-high",
messages=[{"role": "user", "content": "Explain quantum tunneling."}],
)
print(resp.choices[0].message.content)Routing across the best free-tier inference networks
A control plane for free AI inference
Everything you'd otherwise stitch together by hand — routing, failover, key management and observability — in one OpenAI-compatible gateway.
Effort-based routing
Ask for low, medium or high. The router cascades down a ranked list of models until one answers — quality when you need it, speed when you don't.
Automatic failover
A 429 or outage on one provider transparently rolls to the next. Your request still completes.
Rate-limit cooldowns
Providers that just got throttled are parked in a cooldown window and skipped until they recover.
Encrypted at rest
Every upstream provider key is sealed with authenticated encryption before it touches the database.
Unified keychain key
One revealable ak- key fronts everything. Rotate it instantly without touching upstream credentials.
Bring your own models
Pin any model id a connected provider supports into a tier, then reorder priority to taste.
Usage analytics
Per-model, per-provider request counts, token totals, success rate and latency — all in one dashboard.
From eight dashboards to one request
No SDK swaps, no per-provider glue. Point the OpenAI client at your keychain URL and choose how hard you want the model to think.
Connect your providers
Paste the free-tier keys you already have — Gemini, Groq, Cerebras and the rest. They're encrypted the moment they arrive.
Pick an effort tier
Send keychain-low, -medium or -high as the model. The router builds an ordered cascade from your enabled models and preferences.
Ship — we handle the rest
Failover, cooldowns and retries happen server-side. You get a clean OpenAI response and a full analytics trail.
Every free model, one key
All 59 free-tier models across 8 providers — reachable the moment you connect each provider's key.
Google's multimodal flagship — fast 2.0 Flash, deep 2.5 Pro.
LPU inference — Llama 3.3 70B at hundreds of tokens/sec.
Wafer-scale speed for Llama- and Qwen-class open models.
Efficient European frontier models, open-weight roots.
Reasoning-first models that rival closed frontier labs.
A meta-gateway — dozens of free community models in one slot.
Open-source model hosting at production scale.
Enterprise-grade Command models, OpenAI-compatible.
Three knobs, the whole model landscape
Each tier is an ordered cascade of real models. Reorder, disable or extend any of them from your dashboard — these are the live defaults.
Latency-optimized small models for autocomplete, classification and high-volume calls.
- 1gemini-2.0-flash
- 2groq/llama-3.1-8b-instant
- 3cerebras/llama3.1-8b
- 4openrouter/nvidia/llama-nemotron-nano-9b-v2:free
- 5openrouter/google/gemma-4-26b-a4b:free
- 6openrouter/openai/gpt-oss-20b:free
The everyday workhorse tier — balanced quality and speed for chat and agents.
- 1gemini-2.0-flash
- 2groq/llama-3.3-70b-versatile
- 3mistral-small-latest
- 4openrouter/google/gemma-4-31b:free
- 5openrouter/nvidia/nemotron-3-super:free
- 6openrouter/cohere/north-mini-code:free
- 7openrouter/openai/gpt-oss-120b:free
Frontier reasoning for hard problems — R1, Gemini 2.5 Pro and Nemotron Ultra.
- 1gemini-2.5-pro
- 2deepseek/deepseek-r1
- 3groq/llama-3.3-70b-versatile
- 4openrouter/nvidia/nemotron-3-ultra:free
- 5openrouter/poolside/laguna-m.1:free
- 6openrouter/poolside/laguna-xs.2:free
- 7openrouter/tngtech/deepseek-r1t2-chimera:free
If you can call OpenAI, you're already done
Swap the base URL and key. Keep your prompts, your SDK, your streaming, your tools. The model field becomes an effort tier — everything else is identical.
- Drop-in base URL & bearer key
- Server-side routing & failover
- Every call logged for analytics
# A different SDK + key per provider…
from google import genai
from groq import Groq
from mistralai import Mistral
g = genai.Client(api_key=GEMINI_KEY)
q = Groq(api_key=GROQ_KEY)
m = Mistral(api_key=MISTRAL_KEY)
# …and you hand-roll the failover.Stop babysitting eight API keys.
Spin up your unified key in under a minute and route across every free-tier model from a single endpoint.