> ## Documentation Index
> Fetch the complete documentation index at: https://docs.deepmerge.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Deepmerge changelog: releases and product updates

> Release notes for Deepmerge: new features, improvements, and changes to the audit trail, shared skills layer, retrieval, and tool connectors, listed newest first.

<Update label="June 25, 2026" description="One shared brain per team" tags={["Memory", "Design"]}>
  ## Your team is now one shared brain

  We removed project folders. Everything your tools save lives in one shared
  space for the whole team, organized automatically instead of by folders you
  have to set up.

  * **No more workspaces to manage.** Every memory belongs to the team directly.
    Your tools organize what they save through living topic documents, links
    between memories, and search, so you never file anything by hand.
  * **Everything stays searchable in one place.** Any connected tool reads the
    whole team's memory before starting, the same as before.
  * **Corrections are tracked.** When a tool fixes a mistake in an earlier
    memory, the correction is recorded and the original stays visible in history.
</Update>

<Update label="June 25, 2026" description="A chat Assistant that sets up your Context Sprints" tags={["Assistant", "Context Sprints", "Design"]}>
  ## Talk to Deepmerge to set up a Context Sprint

  Deepmerge now has a built-in Assistant you can talk to. Describe the workflow you
  want to capture and it drafts an **Agent Context Sprint** with you, then opens it
  in a side panel where you invite the people who know the work and launch.

  * **Set up a sprint by chatting.** Tell the Assistant what you want your tools to
    learn and it builds the sprint plan with you, instead of asking you to fill in
    a form.
  * **The plan lives beside the conversation.** The draft sprint updates live in a
    side panel as you refine it, so you always see exactly what you are about to
    launch.
  * **Invite and launch in one place.** Open the sprint from the panel to add
    participants, review the roster, and send the invites, all without leaving the
    page.
</Update>

<Update label="June 24, 2026" description="Connecting an AI tool now works the first time, every time" tags={["Connected tools", "Reliability"]}>
  ## Connecting a tool is reliable end to end

  We fixed three issues in the connect-and-save path, so adding a new AI tool and
  writing to your workspace just works, even when traffic is spread across our
  servers.

  * **First-time connect no longer drops you.** Connecting a tool while signed out
    used to land you on the dashboard after sign-in and quietly abandon the
    connection. Now sign-in returns you straight to the approval screen, so the
    tool finishes connecting.
  * **The approval step completes cleanly.** Approving a tool now hands the
    connection back to it reliably instead of stalling at the final step.
  * **No more "session expired" mid-save.** Saving to your workspace no longer
    fails intermittently depending on which server handles the request. Every
    request now stands on its own, so a tool that just connected can always write.
</Update>

<Update label="June 19, 2026" description="Cleaner reference citations on memory pages" tags={["Memory", "Design"]}>
  ## Memory references now read like citations

  Memory pages keep the body simple while still showing provenance. Inline
  references now render as numbered citation chips that match the reference list
  in the details rail.

  * **No graph language in the human UI.** Pages show References and Referenced
    by, not relation types.
  * **Inline sources are easier to scan.** A `[[saved memory]]` reference appears
    as a small number in the text, with the matching source listed beside or
    below the memory.
  * **Styles load from the real Tailwind bundle.** The app no longer depends on
    stale local compiled assets for its product styling.
</Update>

<Update label="June 19, 2026" description="Simpler workspaces for saved agent work" tags={["Workspaces", "Design"]}>
  ## Workspaces now show the saved work directly

  Workspaces now behave like simple project folders: open a workspace and you see
  the memories saved there as tiles, with search and pagination.

  * **No extra memory categories on the workspace page.** The page no longer groups
    saved work into a second layer of human-facing buckets.
  * **Memory tiles are the primary surface.** Each tile shows the headline, preview,
    and update date. Open it when you need the full body, attribution, history, or
    relations.
  * **Agents still own the structure.** Workspace filing, memory kinds, relations,
    revisions, and retrieval remain intact for connected tools; the human UI is just
    easier to scan.
</Update>

<Update label="June 17, 2026" description="Context Sprints, loop health, and a cleaner workspace" tags={["Context Sprints", "Audit", "Design"]}>
  ## Deepmerge now turns workflow interviews into reusable agent context

  The product now has a first-class **Agent Context Sprint** flow: pick one
  workflow, invite the people who know it, let the AI interviewer collect
  concrete examples, and turn the result into a report plus reusable context
  your connected tools can apply.

  * **Context Sprints replaced generic interviews.** The sprint flow now starts
    from the workflow goal, hypotheses, participant plan, and evidence needed to
    prove whether agents can reuse the context afterward.
  * **Reports build as conversations land.** Completed conversations feed the
    sprint report, and the report records how many responses it synthesized.
  * **The interviewer uses the workspace while asking questions.** It searches
    existing context during each conversation so follow-ups are specific to the
    team, not a generic script.
  * **Sprints write back to the workspace.** A finished sprint produces durable
    findings, insights, draft skills, decisions, artifacts, and a reuse proof
    task instead of leaving the evidence trapped in a transcript.
  * **Loop health is visible.** The dashboard and tool bootstrap now show whether
    agents are reading before they write, including retrieval counts, write
    counts, missed read-before-write loops, and agents that need attention.
  * **Session bootstrap is more useful.** `get_self` now includes living topic
    docs, loop health, knowledge gaps, memory hubs, recent memories, workspace ids,
    and memories changed since the tool last connected.
  * **The app surface is quieter.** Skills moved out of the sidebar, dashboard
    and memory views were tightened, tables and dropdowns are more accessible,
    and the marketing site now focuses on Agent Context Sprints.
</Update>

<Update label="June 11, 2026" description="Insights: living answer docs your AI tools keep current" tags={["Insights", "Memory", "Quality"]}>
  ## Your team's knowledge now has two layers: what happened, and what's true

  Memories now come in five kinds across two layers. The diary records what
  happened: findings, decisions, and deliverables, written once and kept as the
  audit trail. The manual records what's true: **insights** and reusable
  how-tos, living documents your AI tools keep current instead of piling up
  near-duplicates.

  * **Insights are living answer docs.** "How we process refunds." "Our
    positioning." One document per topic per project, updated in place by your
    AI tools, with every version kept in history. When a tool tries to save a
    near-duplicate, Deepmerge points it at the existing insight to update
    instead - your knowledge consolidates rather than sprawls.
  * **Insights and how-tos are never "replaced".** They evolve. The replace
    mechanic now belongs only to the diary, where a new decision genuinely
    retires an old one. No path - direct edit, replace link, or version
    restore - can sneak a living doc into a replaced state.
  * **Your tools start every session knowing the manual.** Connecting tools now
    receive the team's living topic docs up front - what the team knows, newest
    first - instead of crawling memory one page at a time.
  * **Notes and todos retired.** Durable context that used to land in "note" is
    now an insight; existing notes were converted automatically. Todos were task
    bookkeeping, not knowledge - open questions now live as tracked gaps that
    close when someone writes the answer. Existing todos were archived as
    records of past plans.
  * **Version history is safe across the change.** Old versions that reference
    retired kinds still open, restore their content cleanly, and never break
    the page.
  * **Search results are now genuinely ranked.** A bug made the keyword half of
    memory search return results in arbitrary order; relevance ranking now
    flows all the way through, so the best match is actually first.
  * **Faster everywhere.** Memory lists, session bootstraps, and link syncing
    each dropped redundant queries, and the app stopped re-writing session
    records on every page navigation.
  * **Archiving polish.** Trying to archive the built-in General workspace now
    explains itself instead of erroring.
</Update>

<Update label="June 10, 2026" description="Project folders, sharper search, cleaner memory" tags={["Workspaces", "Search", "Design"]}>
  ## Every memory has a home, and the workspace finally reads at a glance

  Two days of polish across the whole product, shaped by how real teams use it.

  * **Workspaces are now project folders.** Every memory lives in exactly one
    workspace. Anything your tools save without picking a project lands in the
    built-in **General** workspace, so nothing floats loose and nothing gets lost.
    Tools name workspaces after the project they're working on, and a near-duplicate
    name ("checkout" vs "Checkout") reuses the existing folder instead of creating
    a copy.
  * **Search now filters instead of just sorting.** Searching your memories returns
    the ones that actually match, not the whole list reordered.
  * **Related memories are easier to follow.** Memory pages show the links,
    replacements, contradictions, and supporting evidence behind each saved item.
  * **Memories show their project.** Every row in Memories shows which workspace it
    belongs to, and you can filter by workspace.
  * **The home page grows with you.** Once your AI tools are connected, the home page
    shows them - who's active and when - instead of repeating the setup pitch.
  * **Version history is easier to trust.** Versions are labeled "Current", "v2",
    "v1" with the tool that made each edit, and past versions render their links
    properly. Reverting is now something you ask your AI tools to do - they restore
    any past version as a new, attributed edit, so history is never rewritten.
  * **Citations in ChatGPT.** When ChatGPT answers from your team's memory, it now
    cites the actual memory page - with who saved it and which project it belongs to.
  * **Profile photos and team logo.** Upload yours in Preferences so saved work
    and account settings show the right identity.
  * **Tidy workspaces.** Your AI tools can now rename a project workspace, fix its
    description, or archive a finished one (and bring it back). Workspace memory
    counts now match what you actually see in listings.
  * **Old memories get a second look.** When a tool pulls up a memory that hasn't
    been touched in months, it's flagged as possibly out of date - so your tools
    verify before building on stale information instead of confidently repeating it.
  * **Setup that sticks.** The new setup prompt has your AI tool add Deepmerge to its
    own permanent instructions, so every future session reads the shared context
    before working and saves what it learns after - automatically, without you asking.
  * **Email that keeps up.** A welcome note when you join, a heads-up when an invite
    is accepted, and a notice if a payment fails. Sign-in codes can now be clicked
    straight from the email.
</Update>

<Update label="June 5, 2026" description="Connected, not duplicated" tags={["Memory", "Teamwork"]}>
  ## Your tools build on each other's work instead of repeating it

  Deepmerge now does more of the connecting for you, so your team's AI tools pick up
  where each other left off instead of starting over.

  * **No more near-duplicates.** As a tool saves a note, it sees the closest existing
    notes right away and builds on them instead of writing the same thing again. When
    one is clearly the same topic, Deepmerge links the two together automatically.
  * **Tools see what the others are working on.** When a tool starts up, it sees
    what each teammate's tool worked on most recently. Two tools on the same goal
    no longer redo each other's work.
  * **A head start on every session.** Deepmerge hands a fresh tool the most
    connected notes first, so it gets oriented in seconds instead of digging.
  * **Search catches the exact thing.** Looking for a specific name, code, or id now
    surfaces it reliably, on top of finding notes by meaning.
</Update>

<Update label="June 1, 2026" description="Deepmerge is live" tags={["Launch"]}>
  ## Deepmerge is here

  Your team's AI tools now share one workspace. A few things that work from day one:

  * **Works with any AI tool.** Claude, ChatGPT, Cursor, and others all read and write
    the same workspace.
  * **Finds memories by meaning.** Ask in plain words and your tools surface the right
    past memory, even when you don't remember the exact wording.
  * **Full history on everything.** Every memory keeps its earlier versions, so you can
    always see what changed.
  * **Read-only when you want it.** Give a tool permission to look things up without
    letting it change anything.

  Connect a tool once and you're set. No keys to manage.
</Update>

<Update label="May 30, 2026" description="Smarter retrieval" tags={["Memory"]}>
  ## Old notes step aside when you replace them

  When a tool saves a note that replaces an older one, the old version now steps out
  of the way automatically. Your AI tools always look up the current version first,
  and the full history stays available if you need it.

  * Replacing a note retires the old one in the same step, so searches return the
    latest thinking, not stale copies.
  * When two notes disagree, each one now points directly at the conflicting note, so
    a tool can weigh both without extra lookups.
  * Re-creating a workspace that already exists simply reuses it instead of erroring.
</Update>

<Update label="May 29, 2026" description="Workspaces" tags={["Workspaces"]}>
  ## Group your memories by goal

  You can now organize saved memories into **workspaces**: optional containers for one
  goal, like a product, a client, or a research topic.

  Workspaces are a filter, not a wall. Notes that are not filed under any workspace
  still belong to your whole team, and a tool can read across everything at once or
  focus on a single goal.

  * Your AI tools create workspaces on their own as they work.
  * Each workspace shows its name, description, and how many notes it holds.
</Update>

<Update label="May 27, 2026" description="Slash commands" tags={["Connected tools"]}>
  ## Quick commands inside your AI tool

  Connected tools that support slash commands now show three Deepmerge shortcuts:

  * **bootstrap** opens a session and surfaces what changed since the tool was last here.
  * **capture** reviews the conversation and saves the findings, decisions, and how-tos worth keeping.
  * **retrieve** loads relevant context before the tool starts a task.

  These make the read-first, write-after habit a single command instead of a manual ask.
</Update>

<Update label="May 22, 2026" description="Notes in plain writing" tags={["Memory"]}>
  ## Your tools save notes the way a person would

  There are no rigid forms to fill in. Your AI tools now write each note as normal
  text, with headings, lists, links, and code where it helps. The first line becomes
  the note's title automatically.

  The result reads like something a teammate wrote, not a database record.
</Update>

<Update label="May 14, 2026" description="Organized by what each note is" tags={["Memory"]}>
  ## Every note has a type, so nothing is a jumble

  Saved notes are now sorted by what they actually are, so the right thing is easy to
  find later:

  * Something the tool **learned**
  * A **decision** that was made, and why
  * A **source** worth keeping
  * Something the tool **produced**, like a draft or report
  * A reusable **how-to**
  * A **to-do**

  Your tools pick the type as they save. You just read.
</Update>

<Update label="May 11, 2026" description="Every note is signed" tags={["Audit"]}>
  ## Always know which tool saved what, and when

  Every saved note now records which AI tool wrote it and the time it happened. This
  is the heart of Deepmerge: a clear record you can read whenever you want to know
  what your AI did, without taking anyone's word for it.

  Nothing is anonymous, and nothing is silently overwritten.
</Update>