embedrock 🪨

OpenAI-compatible embedding proxy for Amazon Bedrock. Drop-in replacement for any tool expecting /v1/embeddings.

Why?

Many AI tools (OpenClaw, LangChain, LlamaIndex, etc.) expect OpenAI's embedding API format. embedrock translates those calls to Amazon Bedrock, so you can use Titan, Cohere, and other Bedrock embedding models without changing your tools.

Zero API keys needed — uses your AWS credentials (instance profile, env vars, or shared config).

Supported Models

Model	ID	Dims	Notes
Titan Embed Text V2	amazon.titan-embed-text-v2:0	1024	Default
Titan Embed G1 Text	amazon.titan-embed-g1-text-02	1536
Cohere Embed English v3	cohere.embed-english-v3	1024
Cohere Embed Multilingual v3	cohere.embed-multilingual-v3	1024
Cohere Embed v4	cohere.embed-v4:0	1536	Latest, best quality

Model family is auto-detected by ID prefix. Titan and Cohere use different Bedrock request/response formats — embedrock handles this transparently.

Single-model design: Each embedrock process serves one model. If a request specifies a different model, it returns HTTP 400 with a clear error. Run multiple instances for multiple models.

Install

From releases:

# Linux arm64 (EC2 Graviton)
curl -fsSL https://github.com/inceptionstack/embedrock/releases/latest/download/embedrock-linux-arm64 -o /usr/local/bin/embedrock
chmod +x /usr/local/bin/embedrock

# Linux amd64
curl -fsSL https://github.com/inceptionstack/embedrock/releases/latest/download/embedrock-linux-amd64 -o /usr/local/bin/embedrock
chmod +x /usr/local/bin/embedrock

# macOS arm64 (Apple Silicon)
curl -fsSL https://github.com/inceptionstack/embedrock/releases/latest/download/embedrock-darwin-arm64 -o /usr/local/bin/embedrock
chmod +x /usr/local/bin/embedrock

# macOS amd64
curl -fsSL https://github.com/inceptionstack/embedrock/releases/latest/download/embedrock-darwin-amd64 -o /usr/local/bin/embedrock
chmod +x /usr/local/bin/embedrock

From source:

go install github.com/inceptionstack/embedrock/cmd/embedrock@latest

Usage

# Default: localhost:8089, us-east-1, Titan Embed v2
embedrock

# Cohere Embed v4 (recommended for best quality)
embedrock --model cohere.embed-v4:0

# Custom config
embedrock --port 9090 --region eu-west-1 --model cohere.embed-english-v3

# Show version
embedrock --version

# All flags
embedrock --help

CLI Reference

Flag	Default	Description
--port	8089	Port to listen on
--host	127.0.0.1	Host to bind to
--region	us-east-1	AWS region for Bedrock
--model	amazon.titan-embed-text-v2:0	Bedrock embedding model ID
--version		Show version and exit

Subcommands

Command	Description
embedrock update	Self-update to the latest release from GitHub
sudo embedrock install-daemon	Install as a systemd service

Self-Update

embedrock can update itself in place:

embedrock update

This will:

1. Check GitHub for the latest release
2. Download the correct binary for your OS/architecture
3. Verify the SHA-256 checksum
4. Replace the current binary atomically
5. Restart the systemd service if running as one (requires root)

If running as a non-root user with a systemd service:

Updated embedrock from v0.3.0 to v0.4.0
Restart embedrock.service manually (requires sudo)

Install as Daemon

Install embedrock as a systemd service with one command:

# Install with defaults (port 8089, us-east-1, Titan v2)
sudo embedrock install-daemon

# Install with custom settings
sudo embedrock --port 9090 --region eu-west-1 --model cohere.embed-v4:0 install-daemon

This will:

1. Copy the binary to /usr/local/bin/embedrock
2. Write a systemd unit file to /etc/systemd/system/embedrock.service
3. Run systemctl daemon-reload, enable, and start

The CLI flags you pass (--port, --region, --model) are baked into the service's ExecStart command.

To check service status:

systemctl status embedrock

API

Health check:

curl http://127.0.0.1:8089/
# {"status":"ok","model":"amazon.titan-embed-text-v2:0"}

Single embedding:

curl -X POST http://127.0.0.1:8089/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{"input": "Hello world", "model": "amazon.titan-embed-text-v2:0"}'

Batch embeddings:

curl -X POST http://127.0.0.1:8089/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{"input": ["First text", "Second text"], "model": "amazon.titan-embed-text-v2:0"}'

Response format:

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": [0.123, -0.456, ...]
    }
  ],
  "model": "amazon.titan-embed-text-v2:0",
  "usage": {
    "prompt_tokens": 3,
    "total_tokens": 3
  }
}

Error responses:

{
  "error": {
    "message": "model 'cohere.embed-v4:0' is not available; this server is configured with 'amazon.titan-embed-text-v2:0'",
    "type": "invalid_request"
  }
}

Internal errors return a generic "embedding failed" message — no AWS internals are leaked to clients.

OpenClaw Configuration

"memorySearch": {
  "enabled": true,
  "provider": "openai",
  "remote": {
    "baseUrl": "http://127.0.0.1:8089/v1/",
    "apiKey": "not-needed"
  },
  "model": "cohere.embed-v4:0",
  "query": {
    "hybrid": { "enabled": true, "vectorWeight": 0.7, "textWeight": 0.3 }
  }
}

Prerequisites

• AWS credentials with bedrock:InvokeModel permission
• Bedrock model access enabled for your chosen embedding model

Development

# Install dependencies
go mod download

# Run tests
go test ./... -v -count=1

# Build
go build ./cmd/embedrock/

# Build with version info
go build -ldflags "-X main.version=v0.4.0 -X main.commit=$(git rev-parse --short HEAD) -X main.date=$(date -u +%Y-%m-%dT%H:%M:%SZ)" ./cmd/embedrock/

# Lint (requires golangci-lint)
golangci-lint run ./...

Architecture

See ARCHITECTURE.md for design details and AGENTS.md for AI coding agent guidelines.

License

MIT