paxm-claude

Why paxm

Memory systems tend to optimize the memory engine and leave integration to everyone else. Users must wire SDKs into each agent, or remember to call MCP tools and skills manually. Provider authors must build a separate integration for every agent runtime.

paxm removes that duplication. Agents get one consistent active and passive memory surface. Providers implement one adapter contract. Users keep control of credentials, routing policy, hooks, and where their data lives.

AI agents  ->  CLI / MCP / skills / hooks  ->  paxm  ->  any memory provider

What you get

• Active memory through CLI commands, MCP tools, and an agent skill.
• Passive memory through lifecycle hooks that recall context and capture durable writes without relying on the model to call a tool.
• Provider independence through a common search/write contract and multi-provider routing.
• Local-first defaults with SQLite and no required account or API key.
• Failure isolation with per-provider timeouts, an overall passive-recall budget, bulkheads, and partial-result fallback.
• Durable passive writes through a local queue with retry, deduplication, and crash-safe state.
• Operational visibility through local event logs, latency histograms, provider errors, timeout counts, and paxm history.

What paxm is not

paxm is not another sophisticated memory engine or a hosted memory cloud. It does not try to own embeddings, knowledge graphs, semantic consolidation, reranking research, or long-term memory intelligence.

Those capabilities belong to memory providers. Paxm focuses on the integration and runtime layer: agent entry points, lifecycle hooks, provider routing, failure isolation, durable delivery, and telemetry.

The built-in SQLite provider is a practical local baseline, not a claim to state-of-the-art memory quality. Use Zep, Mem0, or a custom adapter when you need more advanced retrieval and memory behavior without rebuilding every agent integration.

Quick start

Choose the path for the agent you use. The Codex plugin is the shortest path:

Codex plugin

codex plugin marketplace add pax-beehive/paxm --ref paxm-memory-v0.1.3
codex plugin add paxm-memory@pax-agent-nexus
paxm setup --integration codex-plugin

Start a new Codex task and trust the Pax Agent neXus hooks when /hooks asks. The plugin installs its reviewed paxm binary, registers active-memory skills, and owns the passive Codex hooks. Provider credentials remain user-managed.

Claude Code plugin

Install the paxm CLI, then install the Claude Code plugin:

curl -fsSL https://github.com/pax-beehive/paxm/releases/latest/download/install.sh | bash
claude plugin marketplace add pax-beehive/paxm
claude plugin install paxm-claude@pax-memory
paxm setup --integration claude-plugin

The Claude plugin includes active-memory skills, the paxm MCP server, and five lifecycle hooks: SessionStart, UserPromptSubmit, PostToolUse, PostToolUseFailure, and Stop.

CLI, MCP, or Pi

Install the latest release and run interactive setup. SQLite provides a full local flow without an account or API key.

curl -fsSL https://github.com/pax-beehive/paxm/releases/latest/download/install.sh | bash
paxm setup
paxm config doctor

Write and recall a memory:

paxm remember --profile ltm --text "We chose SQLite for the local memory layer"
paxm recall --query "local memory layer"
paxm history --days 7

Select Pi during setup to install its passive extension. Any MCP-compatible client can use paxm mcp serve without installing passive hooks.

How it works

Agents reach paxm in two ways:

Path	Entry points	Best for
Active	CLI, MCP, skill	Deliberate recall, explicit writes, inspection
Passive	Agent lifecycle hooks	Prompt-time recall and automatic turn capture