Swap your agent's brain in two minutes.
Point Claude Code, Codex, Cursor, Gemini CLI, OpenClaw, Hermes or OpenCode at EvoLink and run GPT-5.5, Claude Fable 5, GLM-5.2 or DeepSeek — one key, one balance, automatically routed to the lowest price.
GPT-5.5 · Claude Fable 5 · GLM-5.2 · DeepSeek · Gemini — one balance
20–76%
cheaper via smart routing
99.9%
uptime with auto failover
1 key
OpenAI · Anthropic · Google · Z.ai · DeepSeek
Three ways in
Pick how your agent meets EvoLink.
CLI · swap the brain
Run our LLMs as the agent
Point Claude Code, Codex or any CLI agent at EvoLink and it thinks with GPT, Claude, GLM or DeepSeek — your choice.
$ ANTHROPIC_BASE_URL=https://direct.evolink.ai
See the configs ↓
MCP · give it tools
Let it generate media
One MCP server hands your agent 60+ image, video and music models it can call mid-task.
$ npx -y @evolinkai/evolink-media@latest
Install the server ↓
Skills · teach it one model
Install a model as a skill
One npx command teaches your agent a single model end to end — prompts, params, retries included.
$ npx evolink-seedance
Browse skills ↓
Get connected
Pick your agent. Paste the config.
Every agent below talks to EvoLink through its own native config — no proxy, no code changes. Create a key once and it works across all of them.
Claude Code
Full guide & troubleshootingClaude Code reads its provider from ~/.claude/settings.json. Point the base URL at EvoLink, keep the whole Claude toolchain.
npm install -g @anthropic-ai/claude-code
Create your EvoLink API key first — it takes about 30 seconds — then paste this into the file:
{
"env": {
"ANTHROPIC_AUTH_TOKEN": "YOUR_EVOLINK_API_KEY",
"ANTHROPIC_BASE_URL": "https://direct.evolink.ai",
"CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
}
}claude "Who are you" # ✓ any normal reply — no 401 / 403 — means you’re on EvoLink
MCP server
Your agent writes code. Now it makes media too.
The EvoLink Media MCP server gives any MCP-capable agent direct access to 60+ generation models — Sora, Kling, Veo, Seedance, GPT Image, Suno — without leaving the editor. Ask for a hero video mid-task; the agent picks the tool, submits the job and polls it to completion.
- Tools for image, video and music generation plus model discovery and task tracking.
- One EVOLINK_API_KEY powers every tool — same billing as direct API calls.
- Failed generations surface structured errors your agent can react to, not dead ends.
Works with Claude Code, Claude Desktop, Cursor, and any MCP-capable client.
npx -y @evolinkai/evolink-media@latest
{
"mcpServers": {
"evolink-media": {
"command": "npx",
"args": ["-y", "@evolinkai/evolink-media@latest"],
"env": { "EVOLINK_API_KEY": "YOUR_EVOLINK_API_KEY" }
}
}
}Skills
Agents already know npx. One command, one new skill.
Each skill package teaches your agent a single EvoLink model end to end — when to use it, how to prompt it, every parameter, and how to recover from errors. Install in OpenClaw, Claude Code or OpenCode.
Seedance 2.0 · video
Text-to-video and reference-to-video generation, up to 4K.
npx evolink-seedance
GPT Image 2 · image
OpenAI's latest image generation and editing.
npx evolink-gpt-image
Nanobanana 2 Lite · image
Fast, low-cost image generation on Gemini.
npx evolink-nanobanana-2-lite
Models
Brains worth swapping in.
Every language model on EvoLink works with the configs above — switch by changing one model id.
| Model | Best for | Context | |
|---|---|---|---|
| GLM-5.2glm-5.2 | 1M-context coding flagship at a fraction of frontier pricing. | 1M | Pricing ↗ |
| Claude Fable 5claude-fable-5 | Anthropic's newest top-tier model with 1M context. | 1M | Pricing ↗ |
| GPT-5.5gpt-5.5 | OpenAI's current flagship for agentic coding. | — | Pricing ↗ |
| DeepSeekdeepseek-chat | Best value for high-volume reasoning workloads. | — | Pricing ↗ |
| GPT-5.6gpt-5.6 | OpenAI's next flagship — the page is live for updates. | — | Coming soon |
Lowest routed price
Smart routing picks the cheapest provider for every call — savings of 20–76% versus going direct.
One balance, every vendor
OpenAI, Anthropic, Google, Z.ai and DeepSeek models on a single pay-as-you-go balance. No subscriptions to juggle.
Failover built in
99.9% uptime with automatic failover between providers, so long agent sessions don't die mid-task.
FAQ
- Config, MCP or skill — which should I use?
- Change the config when you want EvoLink LLMs to BE the agent's brain. Add the MCP server when your agent should call media models as tools mid-task. Install a skill when you want it to master one specific model. They stack — one API key covers all three.
- Which base URL do I use?
- Anthropic-style clients use https://direct.evolink.ai, OpenAI-compatible clients use https://direct.evolink.ai/v1, and Gemini-style clients use https://direct.evolink.ai/v1beta. The snippets above already include the right one for each agent.
- Will this break my existing setup?
- No. Each agent keeps speaking its native protocol — you only change the base URL and the key. To switch back, remove those lines and restart.
- My agent still uses the old provider. What now?
- Restart your terminal so new env vars load, check for environment variables overriding the config file, and in Claude Code run /logout if you were previously signed in with a subscription.
- Which models can I use?
- Any language model in the EvoLink catalog — GPT series, Claude series, GLM, DeepSeek, Gemini and more. Switch by changing the model id in your agent's config.
- How does billing work?
- Pay-as-you-go credits billed per token, with one balance covering every vendor. The dashboard shows usage and spend per API key.
- Is this only for coding agents?
- No. Anything that speaks the OpenAI, Anthropic or Gemini API works — including chat gateways like OpenClaw, workflow agents and your own apps.
Give your agent a better brain today.
Sign up, create a key, paste one config — done.
Sign up & get your key