Integrations

Use your VibeKey as an OpenAI-compatible endpoint. The production base URL is https://www.usevibekey.com/v1. For local development, use http://localhost:3000/v1.

OpenClaw

Tested manually

Best for: agents that accept OpenAI-compatible env-style configuration.

API key:paste your VibeKey customer key in OpenClaw's custom provider config.

Base URL: https://www.usevibekey.com/v1

Model: auto

Model name vs actual model: Your tool may display a different label, but the usage dashboard shows the real provider and model billed.

Known limitations: tool calls and basic file operations verified.

Compatibility checklist
Area	Status
Status	Tested manually
Last tested	2026-05-05
Chat	Works
Base URL override / custom provider	Works
Model	Optional — defaults to 'auto' for smart routing
Lower-cost lane	Verified
Premium debug/coding lane	Verified
Streaming	Backend supported, API verified via curl SSE. Client-side progressive rendering not confirmed.
Tool calls	Verified (DeepSeek + OpenAI)
Agent / file actions	Tool calls and basic file operations verified; agent mode pending manual test
Known limitation	Initially tested through cloudflared/local tunnel; production URL should be used for normal setup.
Data notice	Different providers have different data handling policies. Use auto-premium or provider controls for premium-only routing. Sensitive Mode keeps sensitive prompts off lower-cost providers while still allowing eligible generic prompts to route cheaper.

Usage note

VibeKey logs and bills only API requests that reach the VibeKey endpoint. OpenClaw may serve responses from local session context or internal cache before making a new API call. Those locally-served responses are not billed by VibeKey and will not appear in Usage History.

Vercel / Next.js

Untested

Best for: AI SDK callers using a custom base URL and API key.

API key: store as VIBEKEY_API_KEY in your env (sk-vibekey_live_...).

Base URL: https://www.usevibekey.com/v1

const client = new OpenAI({
  apiKey: process.env.VIBEKEY_API_KEY,
  baseURL: "https://www.usevibekey.com/v1",
});
const response = await client.chat.completions.create({
  model: "auto",
  messages: [{ role: "user", content: "Hello from VibeKey" }],
});

Store VIBEKEY_API_KEY server-side only. Do not expose it in client bundles.

Compatibility checklist
Area	Status
Last tested	Not yet
Chat	Not verified
Streaming	Not yet
Tool calls	Not yet
Code edits	N/A
Notes	Pass baseURL without duplicating /v1 twice in client wrappers. Status: Untested. Example only.

Hermes

Tested manually

Best for: terminal-based AI coding assistants.

API key:paste your VibeKey customer key in Hermes's config or set as an env variable.

Base URL: https://www.usevibekey.com/v1

Model: auto

export HERMES_API_KEY="sk-vibekey_live_..."
export HERMES_BASE_URL="https://www.usevibekey.com/v1"
export HERMES_MODEL="auto"

# Or in a YAML config:
# provider: openai
# api_key: ${HERMES_API_KEY}
# base_url: ${HERMES_BASE_URL}
# model: ${HERMES_MODEL}

Compatibility checklist
Area	Status
Status	Tested manually
Last tested	2026-05-07
Chat	Works
Streaming	Support confirmed
Tool calls	Via OpenAI-compatible endpoint
Code edits	Verified

Supabase Edge Functions

Untested

Best for: server-side calls from Deno with fetch or OpenAI SDK.

API key: store as a Supabase secret (sk-vibekey_live_...).

Base URL: https://www.usevibekey.com/v1

Store your VibeKey in Supabase secrets. Do not expose it to browsers.

Compatibility checklist
Area	Status
Last tested	Not yet
Chat	Not verified
Streaming	Not yet
Tool calls	Not yet
Code edits	N/A
Notes	Keep provider keys server-side only; VibeKey customer key is ok in Edge secrets. Status: Untested. Example only.

Aider

Untested

Best for: CLI coding agents that support OpenAI-compatible endpoints.

API key: env or flag (sk-vibekey_live_...).

Base URL: https://www.usevibekey.com/v1

export OPENAI_API_BASE=https://www.usevibekey.com/v1
export OPENAI_API_KEY=sk-vibekey_live_...
aider --model auto

Compatibility checklist
Area	Status
Last tested	Not yet
Chat	Not verified
Streaming	Not yet
Tool calls	Not yet
Code edits	Untested
Notes	Many CLI tools assume /v1; avoid /api/v1 in docs. Status: Untested. Example only.

Continue

Untested

Best for: VS Code / JetBrains extension users.

API key: Continue config (sk-vibekey_live_...).

Base URL: https://www.usevibekey.com/v1

models:
  - name: VibeKey Auto Balanced
    provider: openai
    model: auto
    apiBase: https://www.usevibekey.com/v1
    apiKey: sk-vibekey_live_...

Compatibility checklist
Area	Status
Last tested	Not yet
Chat	Not verified
Streaming	Not yet
Tool calls	Not yet
Code edits	Untested
Notes	If Continue defaults to OpenAI host, override with your /v1 base. Status: Untested. Example only.

Claude Code CLI

Live

Claude models are available via VibeKey's chat endpoint. Enable Anthropic in your project settings to use Claude Sonnet, Haiku, and Opus.

Claude models are available through the existing /v1/chat/completions endpoint. Enable Anthropic in project settings, then use the model aliases claude-sonnet, claude-haiku, or claude-opus.

Compatibility checklist
Area	Status
Last tested	2026-05-07
Chat	Yes (via /v1/chat/completions)
Streaming	Yes
Tool calls	Converted: tool_use ↔ tool_calls
Code edits	Yes
Aliases	claude-sonnet, claude-haiku, claude-opus

Model lanes

When you set a model, you are picking a lane — not a specific model. VibeKey routes to the best available provider behind the scenes.

auto-save

Uses the lowest-cost allowed model that fits the request.

May route to: DeepSeek, Gemini Flash, or another lower-cost model as providers are added.

auto-balancedRecommended

Uses lower-cost models for simple prompts and premium models for coding, debug, and sensitive prompts.

May route to: DeepSeek for simple tasks, OpenAI for coding/debug/sensitive tasks, and future approved providers.

auto-premium

Uses stronger models by default for coding, debugging, security, and sensitive work.

May route to: OpenAI premium models, Claude Sonnet, Gemini Pro, or another approved premium provider.

auto-best

Uses the highest-quality available model. This can cost more and may drain balance faster.

May route to: GPT-5.5-class models, Claude Opus/Sonnet-class models, or other top-tier approved models.

Auto routes are not fixed model guarantees. VibeKey may update the underlying models as provider pricing, reliability, availability, and quality change. The dashboard shows the actual provider and model used for each request.

Usage costs vary per request. Actual cost depends on the model, tokens, and routing. Provider pricing may change. Spend caps and prepaid balance are your primary protection against unexpected usage.

Comparable model options

VibeKey is not tied to one provider. Each lane can use comparable models based on your project settings, provider availability, pricing, and safety controls.

Lower-cost lane:

DeepSeek
Gemini Flash / Flash-Lite, opt-in
OpenAI nano/mini-class models, planned

Balanced lane:

DeepSeek, where allowed
Gemini Flash, opt-in
OpenAI mini-class models, planned

Premium lane:

OpenAI premium models
Gemini Pro, opt-in
Claude Sonnet, enable in project settings

Best lane:

GPT-5.5-class models, planned
Claude Opus, enable in project settings
Other top-tier approved models

New providers and newly released models launch as opt-in first. They are not added to auto-routing until production behavior, pricing, billing, and Sensitive Mode behavior are verified.

Your usage dashboard shows the actual provider and model used for every request.

Rate limits by traffic class

VibeKey applies different rate limits based on how traffic is classified. IDE-classified clients (Cursor, OpenClaw, Hermes-compatible tools) get higher burst allowances.

Default traffic

300 req/min

Baseline rate for all API keys. Applies to general-purpose clients, SDK calls, curl, and non-IDE integrations.

IDE-classified traffic

1,200 req/min

Higher burst allowance for IDEs and coding agents (Cursor, OpenClaw, Hermes). Detection happens via the User-Agent header and the optional x-vibekey-source header.

Override default limits via environment variables: VIBEKEY_RATE_LIMIT_DEFAULT_RPM and VIBEKEY_RATE_LIMIT_IDE_RPM. Limits apply per API key and reset on a sliding 60-second window. Rate-limited requests include standard Retry-After and X-RateLimit-* headers.

VibeKey does not store raw prompts by default. Requests are transmitted to third-party model providers according to your selected model and project settings.