Amp
Bridge Amp to LLMRPM through a local CLIProxyAPI proxy.
Configuration
| Setting | Value |
|---|---|
| API shape | Anthropic-compatible |
| Base URL | https://llmrpm.com/api |
| Auth | Authorization: Bearer llmrpm_... |
- Create a CLIProxyAPI
config.yamlwith anllmrpmcustom provider. - Set provider
base-urltohttps://llmrpm.com/api/v1andapi-keyto your LLMRPM key. - Map Amp's requested model names to LLMRPM public model IDs under
model-mappings. - Start the local proxy on
http://localhost:3456, then point Amp'sANTHROPIC_BASE_URLat that proxy.
Example
config.yaml
server:
port: 3456
custom-providers:
- name: "llmrpm"
base-url: "https://llmrpm.com/api/v1"
api-key: "llmrpm_YOUR_KEY"
models:
- "claude-sonnet-5"
- "claude-haiku-4.5"
model-mappings:
gpt-5.4:
provider: llmrpm
model: claude-sonnet-5
gpt-5.6-sol:
provider: llmrpm
model: claude-sonnet-5
glm-5:
provider: llmrpm
model: claude-haiku-4.5
force-model-mappings: trueDocker
docker run -d --name cliproxyapi \
-p 3456:3456 \
-v ./config.yaml:/CLIProxyAPI/config.yaml \
eceasy/cliproxyapiShell
export ANTHROPIC_BASE_URL="http://localhost:3456"
export ANTHROPIC_API_KEY="llmrpm_YOUR_KEY"
ampRecommended models
Any public ID from the model list works. Common picks:
claude-sonnet-5— strong model for reasoning, refactors, and coding agents.claude-haiku-4.5— fast option for autocomplete, small questions, and lightweight tasks.deepseek-v4-pro— alternative based on your active catalog and plan.
Troubleshooting
| Symptom | Cause & fix |
|---|---|
| Proxy not starting | Verify Docker is running with docker ps, then check the container logs for a config parse error. |
| Authentication error | Check that the API key has no extra spaces and has not been revoked in the dashboard. |
| Model rejected | Use a public model ID returned by GET /api/v1/models; do not use display names or aliases. |
| Connection issue | Confirm the configured base URL is exactly the one shown above and that your network allows HTTPS to llmrpm.com. |
Notes
- Amp speaks the Anthropic format, so it reaches LLMRPM through the proxy's model mappings.
- Keep force-model-mappings on so every Amp model name resolves to a known LLMRPM ID.