One free API. Every model. Auto-routed.
Hamro AI sits in front of many AI providers and exposes one simple endpoint. You never pick a model — the system always picks the best free one for the job, and switches instantly if it fails.
Each peak is one provider — height = its count of online free models.
How it works
You send one request
Describe what you want — text, image, code. No model name needed.
We auto-detect & score
The router detects the task type and ranks every enabled free model by health, latency and load.
Best model answers
It calls the top model. If it errors, rate-limits or 404s, it silently falls to the next one in milliseconds.
You get one clean response
With headers showing which model and provider actually served you — full transparency.
Many providers, one gateway
Hamro AI fronts many upstream providers and exposes them as one.
Built to just work
No model picking
There is no model dropdown anywhere on the public site or API. Auto-routing is always on.
Automatic failover
If a model errors, times out or gets removed upstream, the next candidate answers in milliseconds.
Zero cost
Every routed model is free. No credit card, no signup friction, no surprise bills.
One unified API
OpenAI-compatible. Existing SDKs work unmodified — point them at ai.hamro.site.
Transparent status
A live status page shows real model health, latency and fallback events — never fake numbers.
Millisecond failover
On failure, the next candidate answers instantly. You get both speed and reliability.
Start in 30 seconds
curl https://ai.hamro.site/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer og_YOUR_KEY" \
-d '{
"messages": [
{"role":"user","content":"नमस्ते! Tell me about Nepal."}
]
}'Frequently asked
What does "free" actually mean?
Every model we route is on its provider's free tier. We never charge you. No credit card, no surprise bills.
Do I need a credit card?
No. A free API key is issued instantly with no signup. Use it for as many requests as your tier allows.
What are the rate limits?
Per key: 200 requests/day, 30/min. Enough for most genuine uses; abusive clients are throttled.
What if the model I know is gone?
You never name a model. The system always picks a live one, and swaps to another if it fails or was removed upstream.
What happens on failure?
The next candidate answers in milliseconds. You won't notice unless you check the x-gateway-* response headers.
Your next idea, on a free API.
No signup, no credit card. Start in 30 seconds.