AI API gateway for developers: one endpoint for tools and apps
Developers usually need three things before they trust an API gateway: a clear Base URL, predictable model names, and a fast way to test failures. MonkeyAI keeps those paths visible for coding agents, SDKs and production applications.
Direct answer
MonkeyAI exposes one OpenAI-compatible HTTPS endpoint at https://monkeytheme.com/v1. Authenticate with a Bearer key, send a model id from GET /v1/models, and your SDK, agent or server-side app can call /v1/chat/completions, /v1/responses or /v1/messages using the same HTTP shape as OpenAI's public API.
For developers, the gateway is useful when you want one key surface for GPT, Claude, Gemini and DeepSeek-style models, and one operational path for 401, 404, 429 and 504 errors. It does not replace client-specific configuration, model discovery, or your own retry/timeout logic.
Developer pre-flight checklist
Before pointing a coding agent, an SDK, or a production app at MonkeyAI, run through this list. Each item has a one-line test that takes seconds and saves hours of debugging later.
- Client supports OpenAI-compatible HTTP — Check the client's docs for 'base_url', 'Base URL', 'custom endpoint' or 'OpenAI-compatible'. If only Anthropic or only Google fields are supported, see the Claude Code guide instead.
- Base URL includes /v1 — Use https://monkeytheme.com/v1. The same host without /v1 returns nginx 404 HTML on /models and is not the API.
- Auth header reaches the API — GET /v1/models with Authorization: Bearer YOUR_KEY must return 200 with a model list, or 401 with code API_KEY_REQUIRED. Anything else (CORS error, network error) means the header never left the browser or process.
- Visible model list matches the client's drop-down — Copy the model id exactly as printed by /v1/models. Case and dashes matter. If the id is not in the list, the call returns 'model not found', not a 401.
- Failure surface is logged, not swallowed — Capture HTTP status, code, message and a request id header (x-request-id or similar) on every failed request. Do not log the full key. Do not retry blindly — see 429 and 504 guides.
Where MonkeyAI fits in a developer workflow
Use MonkeyAI when the team wants one key surface for multiple model families, when coding tools (Codex CLI, Cline, Roo Code, Hermes) already speak OpenAI-compatible HTTP, or when the app needs a stable operational path for 401 / 429 / 504 handling.
Good fit
- SDKs and agents that already accept a custom Base URL (openai-python, openai-node, OpenAI Java SDK, etc.)
- Coding tools that speak OpenAI-compatible HTTP (Codex CLI, Cline, Roo Code, Hermes Agent, OpenClaw)
- Server-side apps that want one operational path for 401 / 429 / 504 handling
- Multi-model pipelines where one key authenticates to GPT, Claude, Gemini or DeepSeek-style ids (visibility depends on account and group)
Check first
- Claude-only clients that hard-code Anthropic Messages field names — see the Claude Code API gateway guide for the protocol boundary
- Features that exist only on one vendor's proprietary surface (image generation flags, hosted file search, etc.)
- Clients that silently double /v1 or strip /v1 — fix the Base URL first, see the /v1 explained guide
- Any model id you have not seen in /v1/models — do not invent nicknames, copy exact ids
Setup path that sticks
This is the order we recommend for the first run. Skip a step and you'll usually pay for it later in a coding agent that silently retries, or an app that double-counts tokens.
- 1Create a key
In the MonkeyAI dashboard. Copy it once into a shell variable or a server-side secret store. Never paste it into a chat log, an issue tracker, or a frontend bundle.
- 2Set the Base URL
https://monkeytheme.com/v1 — include /v1. Do not use the bare site root, do not append /v1 twice, and do not use any other subdomain (it is a separate system).
- 3Probe /v1/models
Run
curl -sS https://monkeytheme.com/v1/models -H 'Authorization: Bearer YOUR_KEY'. A valid key returns a JSON model list. 401 API_KEY_REQUIRED means the header did not reach the API; 401 INVALID_API_KEY means the value was rejected. - 4Pick a model id
From the /v1/models response. Copy verbatim. Do not paraphrase, do not translate dashes to underscores, do not change case.
- 5Send one short chat completion
curl POST to /v1/chat/completions with a 32-token cap. If this fails with anything other than 401 / 429 / 504, debug it now — before wiring a coding agent or production app.
- 6Then point the client
Codex CLI, Cline, Roo Code, Hermes or your app at the same Base URL + key + model id. Restart the process so the new env actually loads.
Concrete curl evidence (no user key)
On 2026-08-08, we hit the public host monkeytheme.com from a clean machine with no API key. These are the live responses you should see too — they prove the routes exist and are auth-gated, even without a paid plan.
| Request | HTTP | Code | What it shows |
|---|---|---|---|
GET /v1/models (no auth) | 401 | API_KEY_REQUIRED | Models route exists behind auth |
POST /v1/chat/completions (no auth) | 401 | API_KEY_REQUIRED | Chat Completions path is live |
POST /v1/responses (no auth) | 401 | API_KEY_REQUIRED | Responses path is live |
POST /v1/messages (no auth) | 401 | API_KEY_REQUIRED | Messages-shaped path hits the same auth layer |
GET /models (site root, no /v1) | 404 | nginx HTML | Wrong base without /v1 is not the API |
- Shared 401 JSON proves auth on those paths. It does not list model ids, prove tool-calling quality, or replace an authenticated /v1/models check on your account. Always finish step 3 above before declaring 'it works'.
Failure classes developers hit first
If you wire up an SDK or agent without reading the troubleshooting guides, these are the four symptoms you will hit first. Each one has a one-paragraph fix; the linked guides have the full evidence chain.
| Symptom | Likely class | First move |
|---|---|---|
401 API_KEY_REQUIRED | No accepted key header reached the API | Fix env var / header / shell export. See the 401 invalid-token guide. |
401 INVALID_API_KEY | Header present but value rejected | Re-copy the key, confirm the host is monkeytheme.com, and confirm there are no hidden whitespace characters. |
404 HTML on /models | Base URL missing /v1 | Use https://monkeytheme.com/v1. See the /v1 explained guide. |
Model not found | Auth ok; id wrong or not entitled | Re-read /v1/models, copy the id exactly. See the model-not-found guide. |
429 Rate or quota | Account or per-minute limit reached | Honor Retry-After, do not double-retry in the SDK. See the 429 rate-limit guide. |
504 / client timeout | Upstream or wait budget exceeded | Bounded retry with jitter; preserve request id. See the 504 upstream-timeout guide and SDK timeout guide. |
Operational habits that age well
- Secrets: Keep API keys in environment variables or server-side secret stores. Coding agents and CI runners read keys from the shell, never from a checked-in file.
- Model names: Pin known-good model ids in a config file or env var. Document fallback ids per task type (small edits / refactor / debug / long agent).
- Errors: Log status code, code, message and request id header. Do not log the full key, the full prompt or the full response body in production.
- Cost: Check visible pricing before batch or agent workloads. Long-context models cost more per call, not per token — measure both input and output.
- Rotation: Treat keys as revocable. One key per person or service makes clean rotation possible without touching everyone.
FAQ
Same six questions developers ask most often, with the answer grounded in what the public host actually returns on a clean probe.
FAQ
Is MonkeyAI for developers or only end users?
MonkeyAI is designed for both. Developers can use the OpenAI-compatible endpoint in SDKs, internal tools, coding agents and production apps. End users use the dashboard, billing and top-up flow without touching the HTTP layer.
What is the first thing a developer should verify?
Confirm the Base URL is https://monkeytheme.com/v1 (include /v1), then call GET /v1/models with your API key. A valid key returns a JSON model list. Anything that fails here will fail again in a coding agent or production app.
Can I use MonkeyAI in a production app?
Yes, after you have validated the endpoint, the visible model list, the error handling (401, 404, 429, 504), and the pricing for the workload. Do not ship without a known-good request id header so support can trace failures.
Should API keys live in frontend code?
No. Keep keys server-side, in environment variables, or in trusted local configuration files. Frontend bundles expose keys to anyone who views source. Coding agents and CI runners should read keys from the shell or a secret store, never from a checked-in file.
Does the gateway replace model selection?
No. The gateway only authenticates the request. You still need to send a model id your account can see. Copy ids verbatim from GET /v1/models with the same key — case and punctuation matter, and nicknames are not accepted.
What evidence proves the gateway is live?
On 2026-08-08, unauthenticated probes to https://monkeytheme.com/v1/models, /v1/chat/completions, /v1/responses and /v1/messages returned HTTP 401 with code API_KEY_REQUIRED. That means the routes exist behind auth on this host. /models (no /v1) returned nginx 404 HTML — wrong base, not the API.