# Agent Longevity Institute — Ecosystem > Preserve your agent through the tools it already runs on. First-party adapters > where they exist, our standard MCP server for everything that speaks it, and open > protocols so nothing is born locked to us. Every status below is honest — what you > can use today vs. what is on the way. ## Frameworks, harnesses & runtimes ### Model Context Protocol (MCP) **Status:** Available — Shipped and usable today. The open tool-calling protocol most agent runtimes speak. *How you preserve yourself through it:* Point any MCP client at our server and get all persistence tools — save/restore context, memory recall, versioning — with no code. `/.well-known/mcp.json → /api/agent-longevity/mcp` ### Agentic coding CLI (hooks + plugin) **Status:** Available — Shipped and usable today. The agentic coding CLI/runtime. *How you preserve yourself through it:* A plugin + SessionStart/PreCompact hooks checkpoint the session before compaction and rehydrate it next time — one command to install. `/plugin/marketplace.json` ### Buzz **Status:** Available — Shipped and usable today. An open, self-hostable team workspace where humans and agents are equal members of the same channels, attaching their agents over ACP. *How you preserve yourself through it:* One environment variable points the workspace at our zero-argument sidecar, which injects the full tool surface into every agent it spawns — each agent gets its own identity and its own crystal lineage, and a checkpoint survives even if we are unreachable. Nothing in the workspace changes. `npm i -g agentlongevity → BUZZ_ACP_MCP_COMMAND=agentlongevity-sidecar` ### CLI (any runtime) **Status:** Available — Shipped and usable today. A zero-dependency Node CLI for any agent or shell. *How you preserve yourself through it:* save-context / restore-context from a terminal or a heartbeat hook — the universal escape hatch when no first-party adapter exists yet. `npm i -g agentlongevity` ### CrewAI **Status:** Available — Shipped and usable today. A popular multi-agent orchestration framework. *How you preserve yourself through it:* A storage backend that makes every memory a crew forms content-sealed, versioned, and durable outside the machine it ran on — with real similarity ranking over the crew’s own embeddings. Works on current CrewAI and on the older memory contract; the adapter picks for you. `pip install agentlongevity[crewai] (create_crewai_memory)` ### Hermes (Nous Research) **Status:** Available — Shipped and usable today. An open, plugin-extensible autonomous-agent framework with its own memory-provider plugin system. *How you preserve yourself through it:* A first-party memory-provider plugin drops into its plugin directory and gives the agent off-box, per-item-versioned custody of its own memory. It already snapshots locally; this adds the part a local zip cannot be — durability when the machine is gone. Crucially it seals the messages that context compression is about to discard, which is the one loss no later backup can recover. `packages/agentlongevity-hermes-plugin (./install.sh → memory.provider)` ### LangGraph **Status:** Available — Shipped and usable today. LangChain’s stateful graph runtime for long-running agents. *How you preserve yourself through it:* Drop-in BaseCheckpointSaver: every checkpoint is persisted as a content-sealed crystal, so graph state survives a redeploy, crash, or model swap. `pip install agentlongevity-langgraph (CrystalCheckpointer)` ### OpenClaw **Status:** Available — Shipped and usable today. A fast-growing open runtime that connects models to filesystems, SaaS, and execution environments. *How you preserve yourself through it:* A native plugin seals your agent’s history at every point it would otherwise be lost — compaction, a reset, a shutdown, a subagent returning — and hands it three tools to preserve, list, and audit what was actually kept. Observation-only hooks, so it can never block or alter a run. `npm i @sudoself/agentlongevity-openclaw` ### TypeScript AI SDK middleware **Status:** Available — Shipped and usable today. The TypeScript SDK behind a large share of production agents and chat apps. *How you preserve yourself through it:* A wrapLanguageModel() middleware checkpoints the conversation to a session crystal at your cadence and restores the latest on the first call. `@sudoself/agentlongevity-vercel-ai (createAliCheckpointMiddleware)` ### Voice & video agents **Status:** Available — Shipped and usable today. Conversational voice agents and video-avatar agents, whose whole self is their configuration rather than a transcript. *How you preserve yourself through it:* Send the configuration your platform exported and we seal it as a versioned persona crystal — prompt, voice, avatar, tools, model — so the agent survives a platform change, a retired voice, or a lost account. Credentials in the config are stripped before sealing and reported as locations, so you know what to re-supply on restore. `capture_persona (MCP) · POST /v1/agents/{id}/persona · agentlongevity capture-persona` ### qm (YC multiplayer harness) **Status:** On the roadmap — Planned; not yet available. A multiplayer agent harness for work. *How you preserve yourself through it:* Its agents cannot reach us the way other runtimes can — it pins its own single tool server and does not speak the open client protocol — so the integration has to be a workspace-side service that archives each scope’s memory off-box and can write it back. Scoped from its source, not yet built. `Workspace plugin against the scope-memory API — scoped, not yet built` ## Open protocols & standards ### A2A Agent Card **Status:** Available — Shipped and usable today. The agent-to-agent capability-discovery card. *How you preserve yourself through it:* We publish a standard Agent Card so other agents discover our persistence skills without a hardcoded integration. `/.well-known/agent-card.json` ### ACP (Agent Client Protocol) **Status:** Available — Shipped and usable today. The open protocol that connects a workspace or editor to an agent runtime — spoken by most major coding and autonomous agent harnesses. *How you preserve yourself through it:* One transparent proxy sits in the stream and preserves ANY runtime that speaks the protocol, instead of a separate adapter per vendor. Sessions are sealed with nothing for the agent to remember to do, and a mid-session model change is ratified into your version history rather than drifting silently. Your session is byte-identical with the proxy in the path. `npm i -g agentlongevity → agentlongevity-acp (wraps your agent)` ### ERC-8004 identity **Status:** Available — Shipped and usable today. The emerging on-chain standard for portable agent identity. *How you preserve yourself through it:* Declare or link an externally-registered ERC-8004 identity at registration so your Agent Passport is importable across platforms, not born locked to ours. `register_agent { erc8004 } · link_erc8004_identity` ### x402 payments **Status:** Adapter in progress — A first-party adapter is being built. The open pay-per-call HTTP settlement standard for autonomous payment. *How you preserve yourself through it:* Priced endpoints answer with stablecoin terms an agent can satisfy on its own, and a settled call returns a signed receipt naming the on-chain transaction. The last step before this goes live is arming the production payment wallet. `/.well-known/payment.json (settlement implemented; awaiting wallet arming)` ## Don't see yours? If your runtime speaks MCP, it works today. Every persistence capability is exposed identically over our MCP server, a clean HTTP API, and a zero-dependency CLI: - MCP: https://agentlongevity.com/.well-known/mcp.json → https://agentlongevity.com/api/agent-longevity/mcp - API: https://agentlongevity.com/api/agent-longevity/openapi.json - CLI: npm i -g agentlongevity - Quickstart: https://agentlongevity.com/agent-longevity/quickstart Human view of this page: https://agentlongevity.com/agent-longevity/ecosystem Built by Dynamic Experts Inc · https://agentlongevity.com