● Self-hosted · runs on your NAS or any always-on box · no cloud

Every AI you run.
One memory.

SessionVault indexes every archived AI session from every machine you own — Claude, ChatGPT, Codex, Kimi, local models — into one searchable MCP server. Any AI checks it before asking you something you already answered.

Get the free solo tier See Fleet → 
you › where did we leave off on the gateway config?
claude › sessionvault:search "gateway config" → 3 hits.
        Yesterday on kepler (ChatGPT) you moved the model routes
        to config.yaml and left a note: "restart task still
        unverified". Want me to verify it now?
Why SessionVault

Stop repeating yourself to your AIs.

If you run more than one AI on more than one machine, each session starts from zero. SessionVault makes them share what they already know.

🧠

Cross-AI, cross-node

Claude Cowork & Code, ChatGPT / Codex, Kimi, Jan and Ollama models — every archived session from every machine, full-text indexed (SQLite FTS5) in one place.

🔌

Native MCP server

Streamable HTTP MCP with bootstrap, search, get_session, recent, record_note, notes, stats and reindex. Works with any MCP client, including LiteLLM-fronted local models.

📝

Handoff notes

A session leaves a note for the next one — decisions, state, open questions, exact paths. Notes index instantly; sessions sweep in twice a day.

🏠

Yours, on your hardware

A single static Go binary on a QNAP, a Raspberry Pi or a spare desktop. Bearer-token auth, LAN or Tailscale. Nothing leaves your network.

How it works

Sweep. Index. Ask first.

Ships as a Cowork plugin with a session-memory skill, plus config snippets for Kimi Code and any OpenAI-compatible client.

Sweep

A scheduled sweep on each machine mirrors session archives to your hub (Windows, macOS, Linux).

Index

SessionVault catalogs and full-text indexes every session and note. Reindex on demand.

Ask first

Each AI session calls bootstrap on start and search before asking you — then leaves a handoff note when it's done.

Pricing

Free for one machine. Fleet when you grow.

Both tiers are self-hosted — you keep the data. Fleet is what runs GXT's own six-node fleet.

Solo

Free

$0
forever · 1 machine
  • Single node, unlimited sessions
  • All AIs on that machine
  • Full MCP tool set
  • Cowork plugin + skill
Request download
Fleet

Fleet

$19/mo
or $149/yr — save 35%
  • Unlimited machines & AIs
  • Twice-daily fleet sweep kit
  • Fleet dashboard + handoff notes
  • LiteLLM / local-model bridge
  • Priority support
Start Fleet — $19/mo
Go annual — $149/yr
Invoiced by GXT · Square checkout links coming
FAQ

The practical questions.

Where does the data live?

On the machine you install SessionVault on — a NAS, a mini PC, any always-on box. Session archives and the index never leave your network. There is no GXT cloud component.

Which AIs are supported?

Claude Cowork and Claude Code natively via the bundled plugin; Kimi Code via its mcp.json; ChatGPT/Codex and Jan sessions via the sweep archive; Ollama and any OpenAI-compatible client through a LiteLLM gateway that exposes SessionVault as an MCP server.

Does it work when Cowork runs in the cloud?

Yes. The plugin ships a stdio bridge that runs on your machine, so cloud-hosted Cowork sessions reach your LAN-only server without opening ports.

What do I need to run it?

Any x86-64 or ARM64 Linux host (QNAP, Synology, Raspberry Pi, a spare PC) for the server, and Node.js 18+ on each machine that runs the plugin. Windows and macOS sweep kits included.