Maple Proxy
The maple-openclaw-plugin manages a local OpenAI-compatible proxy server that forwards requests to Maple's TEE (Trusted Execution Environment) backend. All AI inference runs inside secure enclaves.
Setup
1. Add the Maple provider
Add a maple provider to your openclaw.json with your Maple API key and the models you want to use. maple-proxy runs on port 8787 by default.
{
"models": {
"providers": {
"maple": {
"baseUrl": "http://127.0.0.1:8787/v1",
"apiKey": "YOUR_MAPLE_API_KEY",
"api": "openai-completions",
"models": [
{ "id": "kimi-k2-5", "name": "Kimi K2.5 (recommended)" },
{ "id": "deepseek-r1-0528", "name": "DeepSeek R1" },
{ "id": "gpt-oss-120b", "name": "GPT-OSS 120B" },
{ "id": "llama-3.3-70b", "name": "Llama 3.3 70B" },
{ "id": "qwen3-vl-30b", "name": "Qwen3 VL 30B" }
]
}
}
}
}
Use the same Maple API key you configured in the plugin config -- maple-proxy forwards the Authorization: Bearer header to the TEE backend for authentication.
To discover available models, use the maple_proxy_status tool or call GET http://127.0.0.1:8787/v1/models directly.
2. Add models to the allowlist
If you have an agents.defaults.models section in your config, you must add the maple models you want to use. If you don't have this section at all, skip this step -- all models are allowed by default.
Add each model you want to use as maple/<model-id>. Check available models via GET http://127.0.0.1:8787/v1/models or the maple_proxy_status tool.
{
"agents": {
"defaults": {
"models": {
"maple/kimi-k2-5": {},
"maple/deepseek-r1-0528": {},
"maple/gpt-oss-120b": {},
"maple/llama-3.3-70b": {},
"maple/qwen3-vl-30b": {}
}
}
}
}
3. Restart the gateway
Restart the OpenClaw gateway to pick up the new provider and model config.
Using Maple Models
Use maple models by prefixing with maple/:
maple/kimi-k2-5(recommended)maple/deepseek-r1-0528maple/gpt-oss-120bmaple/llama-3.3-70bmaple/qwen3-vl-30b
To spawn a subagent on a Maple model:
Use sessions_spawn with model: "maple/kimi-k2-5" to run tasks on Maple TEE models.
Status Tool
Use the maple_proxy_status tool to check if the proxy is running, which port it is on, its health status, and the available models endpoint.
Embeddings & Memory Search
maple-proxy serves an OpenAI-compatible embeddings endpoint using the nomic-embed-text model. You can use this for OpenClaw's memory search so that embeddings are generated inside the TEE — no cloud embedding provider needed.
1. Enable the memory-core plugin
The memory_search and memory_get tools are provided by OpenClaw's memory-core plugin. It ships as a stock plugin but must be explicitly enabled. Add it to plugins.allow and plugins.entries:
{
"plugins": {
"allow": ["memory-core"],
"entries": {
"memory-core": {
"enabled": true
}
}
}
}
This requires a full gateway restart (not just SIGUSR1) since it's a plugin change.
2. Configure memorySearch to use maple-proxy
Point memorySearch.remote at the local maple-proxy endpoint. Important: the model field must be nomic-embed-text (without a maple/ provider prefix) — the proxy does not strip provider prefixes for embedding requests.
{
"agents": {
"defaults": {
"memorySearch": {
"enabled": true,
"provider": "openai",
"model": "nomic-embed-text",
"remote": {
"baseUrl": "http://127.0.0.1:8787/v1/",
"apiKey": "YOUR_MAPLE_API_KEY"
}
}
}
}
}
Use the same Maple API key you configured in the plugin config. This replaces the need for a separate OpenAI, Gemini, or Voyage API key for embeddings.
Common mistake: Setting the model to
maple/nomic-embed-textwill cause 400 errors from the proxy. Usenomic-embed-text(no prefix).
3. Restart and reindex
After updating the config, do a full gateway restart, then build the vector index:
# Full restart (plugin changes require this)
systemctl restart openclaw.service
# Index memory files and generate embeddings
openclaw memory index --verbose
# Verify everything is working
openclaw memory status --deep
The status output should show:
- Provider:
openai(this is the API format, not the actual provider) - Model:
nomic-embed-text - Embeddings:
available(notunavailable) - Vector:
ready
4. Test with the CLI and tool
Test from the command line first:
openclaw memory search "your query here"
Once that works, the memory_search tool will also be available to the agent in chat. The agent can call memory_search to semantically search across MEMORY.md and memory/*.md files, with results ranked by relevance and cited with source paths.
Troubleshooting
- "memory slot plugin not found" in logs →
memory-coreis not inplugins.alloworplugins.entries, or hasn't been restarted after adding it - Embeddings 400 error → model name includes provider prefix (
maple/nomic-embed-text), change tonomic-embed-text - Embeddings 401 error → wrong API key, or key is a literal string like
${MAPLE_API_KEY}instead of the actual key value - "Batch: disabled" in status → embeddings failed too many times, fix the config and restart to reset the failure counter
- Only 1/7 files indexed → embeddings were failing, fix config, restart, then run
openclaw memory index --verbose
Direct API Access
GET http://127.0.0.1:8787/v1/models- List available modelsPOST http://127.0.0.1:8787/v1/chat/completions- Chat completions (streaming and non-streaming)POST http://127.0.0.1:8787/v1/embeddings- Generate embeddings (model:nomic-embed-text)GET http://127.0.0.1:8787/health- Health check
Port Override
The default port is 8787. To change it:
{
"plugins": {
"entries": {
"maple-openclaw-plugin": {
"config": { "port": 9000 }
}
}
}
}
If you change the port, update your models.providers.maple.baseUrl and memorySearch.remote.baseUrl to match.
Configuration Changes
Plugin config changes (port, API key, backend URL) require a full gateway restart to take effect. Model and provider config changes hot-apply without a restart.
Converted and distributed by TomeVault — claim your Tome and manage your conversions.