agent-memory-orchestrator-cli

Thin npx installer wrapper for Agent Memory Orchestrator.

It installs the Python AMO runtime with pipx, writes local config, configures Claude/Codex hooks and MCP entries, and can initialize local stores.

Install

npx -y agent-memory-orchestrator-cli -- install --target codex --preset cpu-balanced --qwen-model qwen3:1.7b

The -- after the package name is intentional. It prevents npm/npx from consuming AMO flags such as --target.

Install for Claude and Codex:

npx -y agent-memory-orchestrator-cli -- install --target all --preset cpu-balanced --qwen-model qwen3:1.7b

If npx resolves [email protected], that registry package is too old for this command shape. Publish/use 0.1.6 or newer for peer sidecar install support.

The wrapper automatically skips Python versions that are too new for AMO's Kuzu dependency and asks pipx to create the AMO app with Python 3.10, 3.11, 3.12, or 3.13. This avoids the common macOS ARM failure where python3 points to Python 3.14 and pip tries to build Kuzu from source. If automatic discovery cannot find the right interpreter, install Python 3.13 and rerun the same command.

For unusual machines, you can override the interpreter explicitly:

npx -y agent-memory-orchestrator-cli -- install --pipx-python /opt/homebrew/bin/python3.13 --target codex

On a fresh device, install initializes the empty production marker automatically. The reset-production command is only for an existing AMO home with graph/retrieval data that must be backed up and cleaned explicitly.

Install is per user/device, not per repository. AMO hooks capture Codex/Claude sessions from any working directory. Closed-session production jobs resolve the actual Git repository, store a durable repo_id, and keep central memory, active GraphView, retrieval docs, embeddings, and dashboard views scoped by that repo. If a user works in another repo later, AMO creates separate repo-scoped memory from the same install.

Common Options

| Option | Meaning | | --- | --- | | --target codex|claude|all | Agent configs to patch | | --pipx-python <path> | Optional override for the Python used by pipx | | --preset cpu-light|cpu-balanced|gpu-quality | Local model profile | | --qwen-model <model> | Ollama Qwen model written to config | | --with-models | Install embedding/vector extras | | --with-slack | Install Slack Socket Mode extras | | --with-peer | Install and verify the signed Windows/macOS peer networking sidecar | | --download-models | Intentionally cache selected models once | | --dry-run | Show planned changes only | | --yes | Apply without confirmation | | --skip-init-db | Skip SQLite initialization |

Diagnostics

npx -y agent-memory-orchestrator-cli -- doctor --target codex
amo-cli doctor --target codex

If Windows reports No module named 'agent_memory_orchestrator.app.cli', the terminal is still finding an old pip-installed amo-cli.exe shim. Remove the stale package and rerun the npx installer:

py -3.11 -m pip uninstall -y agent-memory-orchestrator agent-memory-orchestrator-cli
pipx uninstall agent-memory-orchestrator
npx -y agent-memory-orchestrator-cli@latest -- install --from "git+https://github.com/spurbey/agent-memory-orchestrator.git@main" --target codex --preset cpu-balanced --qwen-model qwen3:1.7b --yes --force

Uninstall Managed Entries

amo-cli uninstall --target all

Optional Slack Runtime

npx -y agent-memory-orchestrator-cli -- install --target codex --preset cpu-balanced --qwen-model qwen3:1.7b --with-slack
amo-cli slack setup-wizard
amo-cli slack run --reply-mode answer

Full documentation: https://github.com/spurbey/agent-memory-orchestrator#readme