Context Tools

Context Handoff Builder

Continue work in a new chat without losing what mattered. Paste the session and get a handoff package — not a summary for humans, a state package for the next AI session: decisions to honor, constraints to respect, tasks to resume, open questions to keep tracking — extracted verbatim, never paraphrased. Runs entirely in your browser.

Paste the chat or working notes to hand off — the detector extracts decisions, tasks, constraints, questions, and risks verbatim.

Fidelity Mode

Forensic carries everything — every decision and assumption travels, and overriding one requires raising a question.

Handoff Preview (live — detected state, not the package)

            

AI Resource Library

Resources for this tool

View All Resources →

Workflow Playbooks

Playbooks that use this tool

All Playbooks →
Context Workflows · 3 steps

AI Project Handoff Workflow

Carry a project into a new chat, model, or teammate without the context evaporating — capture the state, distill what's worth keeping, and rebuild it as durable context on the other side.

View Playbook →

How it works

Paste the conversation or session notes you need to continue elsewhere, optionally state the current objective and the recommended next step, and pick a fidelity mode. The State Extraction Engine detects the session's working state deterministically — decisions, constraints, pending tasks, open questions, risks, assumptions, and completed work — and carries every item VERBATIM, deduplicated but never paraphrased; the live preview shows the detected counts as you type. Click Build Handoff Package for the nine-section package: current objective and state, key decisions, constraints, open questions, pending tasks, risks (plus assumptions in Detailed and Forensic), a recommended next step, and handoff instructions that tell the new session to accept the context, preserve the decisions, leave assumptions alone, track the questions, and continue rather than restart. Paste the package as the FIRST message of the new chat. Anything the detector missed gets an honest bracketed placeholder, never an invented item. Nothing leaves your browser.

Use cases

  • Continuing a project when the chat window fills up
  • Carrying decisions and constraints into a fresh session
  • Handing a long working session to tomorrow's you
  • Checkpointing multi-day AI collaborations

Pro tips

  • Hand off the recent working session, not the whole archive — the detector extracts state, and state lives in the last stretch of real work. An old exploratory ramble adds noise, not context.
  • Review the package before sending — it tells you to. Items are extracted verbatim from your session, so a sarcastic 'sure, let's ship it broken' can land under decisions; thirty seconds of review beats a session built on it.
  • Use Forensic for anything with real decisions at stake: it carries everything, adds the assumptions section, and instructs the new session that overriding a decision requires raising an explicit question first.
  • Fill the Current Objective field even though it's optional. The detector finds what was decided and what remains, but only you know what the work is FOR — and the new session starts from that line.

FAQ

Is this just a chat summarizer?

No — and the difference is the whole product. A summary tells a human what happened; a handoff package tells the NEXT AI SESSION how to continue: which decisions are settled, which constraints bind, what remains to be done, what is still open. It's organized state with standing instructions, not narrative. The Structured Summary Prompt owns summaries for humans; this owns state for machines.

How does the extraction work — does an AI read my conversation?

No AI reads anything: detection is deterministic pattern matching in your browser. Lines that signal decisions ("we agreed on…"), constraints ("must not change…"), tasks ("still need to…"), questions, risks, and assumptions are recognized, cleaned of speaker labels and timestamps, deduplicated, and carried VERBATIM into the package. What the patterns miss, you add by hand — the package marks every empty section honestly instead of inventing content.

What do the fidelity modes change?

How much state travels and how strict the instructions are. Compact carries the top three items per section — decisions, constraints, the next step. Standard takes five. Detailed takes eight and adds the assumptions section. Forensic drops nothing: every detected item travels, assumptions are first-class, and the instructions add two rules — decisions can only be challenged by raising an explicit question, and prior state gets quoted from the package rather than reconstructed from memory.

Why do the handoff instructions matter?

Because a new session's default behavior is to start fresh: re-derive things, re-open settled questions, politely ignore constraints it wasn't part of deciding. The instructions flip those defaults — accept the context as established, preserve decisions, leave assumptions alone, keep tracking open questions, and continue from the next step. Without them, the package is information; with them, it's a working contract.

When should I split the conversation instead of handing it off?

Split when the AI needs the original MATERIAL — every word of the document, every line of the code; that's the Long Prompt Splitter. Hand off when the AI needs what the conversation ESTABLISHED — the decisions, constraints, and remaining work. A 200K-token session usually compresses into a 1K-token handoff, because most of a conversation is the road, not the destination.

How is this different from the Project Context Builder?

Time scope. The Project Context Builder (coming soon) defines what the AI should permanently know about your project — conventions, glossary, stack — the stable knowledge that outlives any session. The handoff package is a snapshot of NOW: this objective, these open tasks, this next step. PCB is the employee handbook; the handoff is the shift-change briefing.