Direct answer
Continuity is the work itself moving between AIs. Memory is only what one AI recalls.
Every AI app keeps its own history, and the job you are running does not live in any of them: it lives in the gap between them. Start work in one AI, hit a limit or a better tool in another, and the second AI knows nothing: not the goal as agreed, not the steps that already ran, not what the client decided. Memory features soften this inside a single app, and the search for an MCP server for agent memory and continuity usually starts here. But memory and continuity are two different goods, and buying the first while needing the second is the common mistake. This page separates them, shows what breaks when only memory moves, and describes the server shape that carries both.
Updated November 2, 2026. First published November 2, 2026. The two handover records quoted below are real and public, with their decision states read back from the live service on November 2, 2026.
Two kinds of continuity, and why the difference matters
Memory stores are the first kind. A memory-store server saves facts about you and your work: preferences, names, decisions, summaries of past sessions, and recalls them when a new session seems to need them. This is genuinely useful, and it is what most memory tooling sells. Its unit is the fact. The second kind is work continuity, and its unit is the job. Work continuity carries the session itself: the goal, the steps that ran in order, the outputs as saved, the receipt for each step, the items still open, and the decision state, client accepted, client still deciding, release held. When work continuity moves, the receiving AI does not get told about the job. It gets the job, in the state the job was actually left in. Confuse the two and you get the familiar failure: the new AI warmly recalls your client's name and your formatting preferences, and then redoes the analysis from scratch, contradicts the version the client already approved, and asks the question you answered yesterday.
What breaks when only memory moves
Three breaks, in the order you meet them. First, re-explanation: facts without the record still leave the user as the courier, pasting context, re-uploading files, re-stating constraints, because the memory holds conclusions and the work needs the material itself. Second, silent divergence: the receiving AI, working from a summary, produces a new version of an output that already existed, and now two versions circulate with no record of which one the client saw. Third, lost decisions: memory rarely stores the state of play, who accepted what and what is still pending, so decided questions reopen and open questions get treated as settled. Each break costs the thing continuity was supposed to save: the second half of the job goes to reconstruction instead of progress.
How Zambo carries both, from one endpoint
Zambo is the verifiable work layer for AI, and continuity is built into the same endpoint as the tools: https://zambo.dev/mcp, 153 tools with 36 on the default surface, added once to each AI you use. The memory side runs through memory_remember and memory_recall: what an agent saves in one AI, it recalls in another, because the memory sits with the endpoint rather than inside one app's history. The recent-work side runs through continuity_get_recent, which returns the latest jobs across your sessions with their steps and receipts, so a fresh session starts by reading the record of what already ran. The work-continuity side runs through the wrap-up: a session seals into a handover carrying goal, steps, outputs, receipts, open items, and decision state, and another AI picks that handover up and continues the job itself. Every one of those calls, memory calls included, returns a verifiable receipt under AER-1, published as IETF Internet-Draft draft-zambo-aer1-26, with its conformance corpus of 126 vectors and 129 checks public at https://zambo.dev/aer1/corpus.json. The two public handovers below show the record form in its two live states:
Accepted: the decision is on the record
- Status: accepted, decided_by client, decided_at 2026-10-16T14:31:51.146Z, release_hint ok_to_release.
- Steps: 4 steps total, 4 checked.
- aer1_version: draft-zambo-aer1-26.
- Open the handover or read its acceptance.json.
Pending: checked, and still waiting on a decision
- Status: pending, release_hint hold.
- Steps: 5 steps total, 5 checked.
- aer1_version: draft-zambo-aer1-26.
- Open the handover or read its acceptance.json.
Setting it up is the same act as adding any MCP server: add the endpoint to the client, once per AI, and the memory, the recent work, and the handovers are simply there in each one. 20 calls + 1 wrap-up free every day. Picking up in another AI is always free. For a working day that spans several AIs, $0.99 buys the Day Pass, unlimited for 24 hours, no subscription: https://zambo.dev/day-pass/. Ongoing use is the Zambo Pass at $29 a month or $290 a year, and $19 a month for $ZAMBO holders, which adds private handover links for work that should travel between named parties only.
What continuity does not settle
Memory can be wrong: a saved fact that was mistaken when it was saved will be faithfully recalled, and a verifiable receipt proves the record was not changed after saving, not that the remembered fact was true. Continuity carries the work record between AIs; it does not move an app's private chat history, merge accounts, or transfer settings that live inside one vendor's product. What travels is the work: its steps, its outputs, its receipts, and its state. This page is general information, not legal advice.
Frequently asked questions
What is the difference between agent memory and agent continuity?
Memory is recall: facts, preferences, and summaries an AI saved, brought back when they seem relevant. Continuity is the work itself: the goal, the steps that ran, the outputs as saved, the receipts for those steps, the open items, and the decision state, carried from one session or one AI to the next so the job continues instead of restarting. Memory answers the question what did we learn. Continuity answers the question where exactly did this job get to, and it answers with the record, not a recollection.
Can my AI's memory move to a different AI?
Memory saved inside one app stays inside that app; its memory feature serves its own sessions. Memory saved through an endpoint your AIs share moves with you. When memory is written and read through the Zambo endpoint, https://zambo.dev/mcp, any AI with that endpoint added can recall it, because the memory lives with the endpoint rather than inside one chat window. The same applies to the work record: a session sealed as a handover can be picked up by a different AI, and picking up in another AI is always free.
What does continuity_get_recent return?
It returns the recent work record across your sessions: the jobs that ran most recently, in order, with their steps and receipts. A new session, or a different AI picking the work up, calls it to see what already happened before doing anything new, so the first move is informed by the record instead of by a summary the user has to type out again. Each entry carries the receipt the original call issued, checkable at https://zambo.dev/verify/, so recent work arrives with its evidence attached rather than as a claim about itself.
How is a handover different from exporting a chat?
An export is a transcript: the words, in order, with the evidence stripped out and the state left implicit. A handover is the sealed session record: goal, steps, outputs as saved, a verifiable receipt per step under AER-1, draft-zambo-aer1-26, the open items named as open, and the decision state. The receiving AI continues from the handover as a working state, not as reading material, and the client or teammate receiving the same link sees the checked record rather than a chat they have to take on faith.
What does it cost to carry work between AIs this way?
The free tier is 20 calls + 1 wrap-up free every day; picking up in another AI is always free, so the receiving side never pays to continue your work. For $0.99, a Day Pass makes the endpoint, https://zambo.dev/mcp, unlimited for 24 hours, no subscription: get the Day Pass. Ongoing use is the Zambo Pass at $29 a month or $290 a year, and $19 a month for $ZAMBO holders, which adds private handover links, your branding, and the approval record as PDF and JSON files.