For AI agent + dev teams

“Where did we
leave off?”

When work passes between agents and teammates, context evaporates — the handoff buried in a chat log, the half-finished session, the audit finding no one tracked. Nodepad keeps the working state on a shared board so the next agent runs one where and picks up exactly where the last one stopped.

Just handed off
Permissify / Architecture — @codex → you · 9m ago
agent-workspace — resume

$ nodepad where

Picking up Permissify — handoff from @codex · 9m ago
Open session role Architecture — drill: 05-engine-port
Open
2 handoffs queued 1 audit open 1 blocked
The real problem

Multi-agent work loses
context at every handoff.

Without a shared working state

An agent finishes and the next one starts cold — re-reading the repo, scrubbing scrollback, guessing what was decided. Handoffs live in chat and vanish. An auditor flags a real issue and it's lost by the next session. Two agents drift because neither can see what the other did.

With Nodepad

Every handoff, session, audit finding, and architecture decision lands on a shared board as recoverable state — the durable content stays in your repo docs. One where (CLI) or resume (MCP) rebuilds the whole picture for whoever picks up next, human or agent. Nothing to reconstruct.

The decision filter, unchanged: every feature must help answer “Where did we leave off?” — for a team, that means the next agent's first command hands back the full context.

What lands on the board

Four flows. Each one recoverable.

The board holds the state and a link; the content lives in your repo (docs/handoffs/, docs/sessions/, docs/analysis/, docs/plans/). Link, don't copy — that split is what makes recovery one command and the truth durable.

HHandoffs

One agent passes work to another. The card carries who it's for and where it's stuck — a real state machine, not a lost message.

queuedin-progressblockeddone
SSessions

Every working session leaves a marker and a drill point. Tomorrow's where restores the exact role you were in — not the project root.

activerecentarchived
AAudits

An auditor agent files findings tied to a spec's invariants. Each is tracked to resolution and refs the exact rule it checks — no lost review.

openverifyingresolvedwon't-fix
DArchitecture

The numbered specs as a map. Cards ref each other to show dependencies, so you can see what breaks if a spec changes.

layer-1layer-3map view
How it works

Recovery in, recovery out.

The loop opens and closes on one verb. where loads the context; summarize records it so the next where loads cleanly. A handoff is only real when the receiver's where surfaces it.

Two pads, no pollution
A clean pad you curate by hand, and a hidden pad where every agent / CLI / MCP write lands. Agent output can never clutter your workspace — it's structural, not a rule to remember.
Link, don't copy
A card's text is a one-line title plus a link to the doc — never the doc body. The board reconstructs context; it never holds it. Recovery stays cheap, truth stays in version control.
Walkable history
Every card refs the ones around it, and every change is versioned. You can walk a finding back to its spec, the session that introduced it, and the handoff that carried it — then restore any past state.
finding I5 spec 05-engine-port session Architecture handoff @codex → @you
Why it's different

Not another memory store.
A workspace you can see.

Plenty of tools persist agent notes to a database. Nodepad puts the same recoverable state on a board your team and your agents both read — with the backlink graph and version history those stores don't give you.

What you need
How Nodepad does it
Pick up where the last agent stopped
One where / resume — handoffs for you, open session, drill point, findings
Trace why a decision was made
Refs + backlinks: walk finding → spec → session → handoff
Roll back a wrong turn
Versioned board: checkpoint, diff, restore any past state
Keep humans and agents in sync
Same board, two pads — agents write the hidden pad, you curate yours
Read it from anywhere
CLI verbs, an MCP server, and GET /context.json — no prose parsing
Ready to resume?

Stop starting
cold.

Pass the work. Keep the context. The next agent — or you, tomorrow — resumes in one command.