This guide walks you through a complete single-agent installation of agent-memory. It covers two installation paths (source build or prebuilt binaries), a full sample configuration, an optional dry-run config check, and recommended verification steps.
If you only need the essentials, see the Quickstart.
- Platforms: macOS and Linux only
- Mode: single-agent setup (multi-agent configuration is out of scope here)
- Agent hooks: handled in a separate agent-specific guide
Required:
- macOS or Linux
protoc(Protocol Buffers compiler)
Optional (only for source builds):
- Rust 1.82+ with Cargo
Optional (for summarization):
- OpenAI or Anthropic API key
Verify now (optional):
protoc --versionrustc --version
You have two supported options:
- Build from source (recommended if you already have Rust)
- Use prebuilt binaries (fastest path)
git clone https://github.com/SpillwaveSolutions/agent-memory.git
cd agent-memory
cargo build --releaseCopy binaries to a local bin directory:
mkdir -p ~/.local/bin
cp target/release/memory-daemon ~/.local/bin/
cp target/release/memory-ingest ~/.local/bin/mkdir -p ~/.local/bin
PLATFORM=$(uname -s | tr '[:upper:]' '[:lower:]')
ARCH=$(uname -m)
curl -L "https://github.com/SpillwaveSolutions/agent-memory/releases/latest/download/memory-daemon-${PLATFORM}-${ARCH}.tar.gz" | tar xz -C ~/.local/bin
curl -L "https://github.com/SpillwaveSolutions/agent-memory/releases/latest/download/memory-ingest-${PLATFORM}-${ARCH}.tar.gz" | tar xz -C ~/.local/bin
chmod +x ~/.local/bin/memory-daemon ~/.local/bin/memory-ingestexport PATH="$HOME/.local/bin:$PATH"Verify now (optional):
memory-daemon --versionmemory-ingest --version
Create the config directory and set defaults inline.
mkdir -p ~/.config/agent-memorydb_path:~/.local/share/agent-memory/dbgrpc_port:50051grpc_host:0.0.0.0log_level:infosummarizer.provider:openaisummarizer.model:gpt-4o-mini
# ~/.config/agent-memory/config.toml
# Core settings
db_path = "~/.local/share/agent-memory/db"
grpc_port = 50051
grpc_host = "0.0.0.0"
log_level = "info"
# Single-agent mode is the default. Multi-agent options are not covered here.
multi_agent_mode = "separate"
[summarizer]
provider = "openai"
model = "gpt-4o-mini"If you prefer Anthropic, set:
[summarizer]
provider = "anthropic"
model = "claude-3-5-haiku-latest"Verify now (optional):
cat ~/.config/agent-memory/config.toml
Before starting the daemon, run a config check to validate the file.
memory-daemon config checkIf the check fails, fix the reported issues before continuing.
memory-daemon startVerify now (optional):
memory-daemon statusmemory-daemon query --endpoint http://[::1]:50051 root
Agent integration is intentionally separate so the core install flow stays clean. Pick the guide that matches your tool:
See: Agent Setup
If you need advanced configuration (custom paths, ports, or tuning), see:
Keep advanced changes minimal unless you have a specific reason to deviate from the defaults.
Common issues and fixes:
Ensure the install directory is on PATH:
export PATH="$HOME/.local/bin:$PATH"The daemon is not running or the port is wrong:
memory-daemon statusIf the port is busy, choose a new one in config.toml and restart the daemon.
Set an API key for the provider you chose:
export OPENAI_API_KEY="sk-..."
# or
export ANTHROPIC_API_KEY="sk-ant-..."Agent hooks are not configured yet. Follow the agent setup guide and verify the
hooks file points to memory-ingest.
- Keep the daemon running while you work
- Use
/memory-statusto check health - Explore
/memory-searchand/memory-recentonce events are captured