Claude on an OpenAI-compatible endpoint. First call in under ‹N› min.

1. 1

   One OpenAI-compatible key for many models

   What

   point the OpenAI SDK you already use at https://api.nexoai.dev/v1, change the key, and the same chat/completions calls reach Claude and every other model we route. Switching models is one string in the request, not a new integration.

   Why it matters

   going direct means a separate SDK, key, and bill per provider, and glue code to switch between them. One key collapses that to one integration.

   Honest limit

   if you only ever call one model from one provider, this buys you little — go direct. The relay earns its place the moment you use more than one.

2. 2

   No spend-tier ladder to climb for model access

   What

   you get access to the models we route from your first call, on the free credits, with no monthly-spend qualification.

   Why it matters

   going direct, new-model access and higher rate limits are gated behind spend tiers you have to climb — the P1 audience said it plainly:

   "Going direct to the model providers does not work... you get rate limited and are last in the list for new model access"

   and

   "[it] doesn't make you jump through some tiering system."

   A relay pools that access so you're not last in line.

   Honest limit

   ‹N› — NEEDS NUMBER FROM USER: NexoAI's own rate limits / any tiering on the relay side. State them here plainly — if the relay has its own ceilings, this section must name them or it reads as the exact evasion devs distrust.

3. 3

   Automatic fallback when an upstream errors

   What

   when a provider returns an overload or error, the request is retried against another route instead of failing in your app.

   Why it matters

   going direct, an upstream 529/5xx is your outage to catch and handle. A relay can absorb the retry so a transient provider blip doesn't become your incident.

   Honest limit

   ‹N› — NEEDS CONFIRMATION FROM USER: does NexoAI actually do automatic fallback/routing? If not, DELETE this entire reason — it is the single most tempting claim to overstate and the fastest way to lose this audience. Same flag as P1. If fallback is real, state exactly what it falls back to and what the latency cost is; if it's just retry-same-provider, call it that, not "fallback."

4. 4

   One balance, one bill

   What

   usage across every model runs down one prepaid balance you watch decrement in real time, on one invoice, with no card required to start ($5 free credits).

   Why it matters

   going direct with multiple providers means multiple bills, multiple billing dashboards, and separate spend to reconcile. One balance is one number to watch and one place a surprise can't hide.

   Honest limit

   this is convenience, not savings. The relay's fee is where that convenience is priced — ‹N› — NEEDS NUMBER FROM USER: NexoAI's exact fee/markup model — same figure the A2 pricing page needs; the "why a relay" case is incomplete and slightly evasive without it

Go direct if price is your only axis and you're happy managing tiers, keys, and bills yourself. Use NexoAI if one key across many models, no tier ladder, [fallback — pending confirmation], and one balance are worth the fee. That's the whole trade. No lock-in: it's an OpenAI-compatible endpoint, so switching back is changing the base URL back.

Going direct is the cheapest sticker price. It also rate-limits you by monthly spend, gates new models behind tiers, and hands you a separate key and a separate bill for every provider. NexoAI is a relay: one OpenAI-compatible key reaches Claude and the other models we route, with automatic fallback when an upstream provider errors, and one balance you watch decrement in real time. That convenience is not free and we won't pretend it is — ‹N› — NEEDS NUMBER FROM USER: NexoAI's exact fee/markup model — flat % on credits like OpenRouter's 5.5% Stripe / 5% crypto, per-token margin, or per-model rate If your only axis is raw token price and you're fine managing tiers and keys yourself, go direct. If one key, no tier ladder, and fallback across providers is worth the delta below, that's what you're buying.

1. Get your API key

   Create an account and copy your key from the dashboard. No card required for the free tier. ‹N›

   shell Copy

   export NEXOAI_API_KEY="sk-nexo-..."   # paste your key

2. Send a request with curl

   The endpoint speaks the OpenAI chat/completions format.

   shell Copy

   curl https://api.nexoai.dev/v1/chat/completions \
     -H "Authorization: Bearer $NEXOAI_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "model": "claude-sonnet-5",
       "messages": [{"role": "user", "content": "Reply with the word: online"}]
     }'

3. Confirm the response

   You should see a JSON body like this. The content is Claude's reply; usage is what this call cost you in tokens.

   Response 200 OK

   {
     "id": "chatcmpl-...",
     "object": "chat.completion",
     "model": "claude-sonnet-5",
     "choices": [{
       "index": 0,
       "message": {"role": "assistant", "content": "online"},
       "finish_reason": "stop"
     }],
     "usage": {"prompt_tokens": 12, "completion_tokens": 1, "total_tokens": 13}
   }

4. Use it from the OpenAI SDK you already have

   No new library. Point the OpenAI client at NexoAI's base URL.

   Python JavaScript

   python Copy

   from openai import OpenAI
   
   client = OpenAI(
       api_key="YOUR_NEXOAI_KEY",          # or os.environ["NEXOAI_API_KEY"]
       base_url="https://api.nexoai.dev/v1",
   )
   
   resp = client.chat.completions.create(
       model="claude-sonnet-5",
       messages=[{"role": "user", "content": "Reply with the word: online"}],
   )
   print(resp.choices[0].message.content)   # -> online

5. Switch models with one string

   Change model to route to a different model on the same key. See the pricing page for per-model rates. ‹N›

Common errors

401

your key is missing or wrong. Check the Authorization: Bearer header matches the key in your dashboard.

404 model not found

the model string isn't one NexoAI routes. See the model list. ‹N›

429

you hit a rate or spend limit. ‹N›

API and all routed upstreams responding normally. Last checked ‹timestamp›, checks run every ‹N›.

• NexoAI API api.nexoai.dev/v1 Operational
• Dashboard & key management Operational
• Claude upstream Operational
• ‹N› — remaining upstreams, one component each

Uptime, last 90 days

‹N›

p50 / p99 API overhead

‹N›

14:02 UTC — Investigating — Claude (upstream)
What:      Elevated 529 (overloaded) responses from the Claude upstream.
Impact:    Claude calls returning errors or timing out. Other routed models and
           the dashboard are unaffected.
Started:   13:58 UTC.
Next update: by 14:20 UTC.

14:19 UTC — Identified — Claude (upstream)
What:      Confirmed upstream capacity issue at the provider, not on our side.
Impact:    Same as above.
Started:   13:58 UTC.
Next update: by 14:45 UTC.

14:38 UTC — Monitoring — Claude (upstream)
What:      Upstream error rate back to normal. Watching to confirm it holds.
Impact:    Claude calls succeeding again. No action needed on your end.
Started:   13:58 UTC.
Next update: by 15:10 UTC, or a Resolved note if stable.

15:09 UTC — Resolved — Claude (upstream)
What:      Upstream stable for 30 minutes. Incident closed.
Impact:    None ongoing. Total window: 13:58–14:35 UTC (~37 min) of elevated
           Claude errors.
Started:   13:58 UTC.