Every conversation with Claude Code, Codex and friends — synced from every machine, redacted locally, distilled into context any AI pulls back with one tool call.
Why
The difference
What resuming work looks like, with and without a central memory.
Paste files, repeat decisions, reconstruct what was already settled.
The context lives in a .jsonl on ANOTHER computer — if it still exists.
Why did we choose X? Nobody remembers, so you re-litigate it.
handoff-it ships every conversation, redacted and compressed, as it happens.
State, decisions and pending work — saved at every milestone via MCP.
One call returns the distilled context. The AI continues as if it never left.
Works across tools: start in Claude Code, continue in Codex.
How it works
Three moving parts, one shared memory.
→ recall_project("public-it")
← 3 checkpoints · 5 notes · 12 sessions
→ search_conversations("stripe webhook")
← the exact moment, sessionId + offset
Capabilities
The archive is the contract — everything else can be rebuilt from it.
Append-only copies of every transcript, chunked and compressed. Re-parse anytime.
Tokens, keys and passwords replaced on your machine — deterministically, before upload.
Find the exact moment a decision was made, across every tool and machine.
Seven tools: recall, search, drill-down, checkpoint. Works with any MCP client.
On the Team plan the workspace shares one archive — everyone reads and feeds it.
Same product inside your network. Conversations never leave the building.
Sources
Claude Code and Codex today; the raw archive is designed so new tools like Cursor plug in as adapters.
Integrations
A CLI for the raw stream, MCP for the AIs, REST for everything else.
One init, then it keeps itself current.
npx handoff-it init --key hof_...
npx handoff-it watchAny MCP client, one command.
claude mcp add --transport http handoff-it \
https://handoff-it.com/mcp \
--header "Authorization: Bearer hof_..."The first thing to ask on a new machine.
> pull the memory of project X
→ recall_project("X")
← state · decisions · pendingPricing
Tiers differ in storage and seats. The memory is yours on every one of them.
Try your central memory
For daily AI work
For power users and consultants
Your team's shared memory
Ready?
Connect a machine and pull your first recall in minutes.