October 2026
Improved
A quieter app
Activity, connections, memory, sharing and account screens drop status pills, frames, dividers and decorative art. Status reads as plain words, and only failures are colored.
Improved
Plainer marketing pages
The public pages drop animations, mockups, tags and numbered labels for plain headings, short text and the full list of agent logos.
New
One-step setup for OpenClaw, Hermes Agent, Amp and Devin CLI
The setup command now finds and connects OpenClaw, Hermes Agent, Amp and Devin CLI alongside Claude Code, Codex and Cursor, each with its own key and one browser approval. Inside Devin, Claude Code's hooks stay quiet so Devin's work is credited to Devin.
Under the hood
Instinct and Aider logos restored
Reverts the earlier removal, so the agent list on the homepage is unchanged.
Under the hood
Instinct and Aider logos removed
Briefly removed from the homepage agent list, then restored in #66 before any release.
Improved
Old keys turn off when you reconnect
Running setup again now turns off the access keys this computer no longer uses, including those of folder setups you switch. Keys still in use are left alone, and turned-off keys stay listed as revoked under Connections.
Improved
Smoother first-time setup
The setup command no longer stops for npx's install prompt, Codex keeps your per-tool approval settings when it reconnects, Engram's hooks carry readable labels, and setups made earlier for single folders can be switched to the new connection in one answer.
New
Connect a computer in one step
Run
npx @ani-hq/engram-mcp@latest, approve in your browser, and Claude Code, Codex and Cursor, including their desktop apps, are connected with their own keys. Engram then works in every repo, with no config or Markdown files to edit.Under the hood
Release record and Cursor results
Records the 28 October release and the live Cursor transfer runs: 6 of 9 pass, and Cursor-first interruption is not yet verified.
New
A public changelog
This page: every change to Engram, newest first, with the time it landed and a link to its source.
Improved
Bridge 0.6.0 with task checkpoints
The
@ani-hq/engram-mcpbridge moves to 0.6.0 with the checkpoint command and continuation hooks. Live transfer results across Claude Code, Codex and Cursor are recorded in the continuation spec.New
Cookieless site analytics
Public pages count views, a few named calls to action and sign-up sources with a first-party script. No cookies, IP addresses, user agents or emails are stored, and Do Not Track and Global Privacy Control are honoured.
New
Task handoffs between agents
Agents can save a compact task record: objective, plan, decisions, blockers and next step. Another agent, or the same one after a crash or usage limit, can find it and continue from the last confirmed save. Activity gains a Handoffs view with a copyable continue prompt.
Improved
Sharing in team workspaces
Team owners and members can browse the approved excerpts they received, open team memory, and switch to a personal workspace to prepare a new share.
Improved
Compact action buttons
Buttons share one compact 32px size on desktop across the interface. Touch screens and narrow widths keep 44px targets.
New
A complete agent connection prompt
Connections offers a full setup prompt to paste into an agent, with the endpoint and intended workspace. It uses native MCP setup where available and exact guidance otherwise.
Improved
Simpler agent connections
Connect an agent shows the shared MCP endpoint right away. Access keys and agent-specific help are optional steps.
Improved
Focused Connections, checks in Activity
Connections lists your agents and opens a dedicated setup dialog for each one, and reconnecting keeps existing settings. Workspace checks moved into Activity.
Fixed
Reliable memory save confirmations
Explicit memory saves get a longer, bounded deadline, so a slow write that reached the server is no longer reported as unavailable. Animation playback controls were removed.
Fixed
Useful memory at small budgets
Installed agent guidance is shorter, so a 1 KiB retrieval budget still leaves room for useful context. Grok setup steps match its current connector flow.
New
Guided memory setup and repair
One guided command from Connections installs or repairs the memory workflow, without editing Markdown or configuration files. File changes are checked first and rolled back on failure.
Improved
Reliable, bounded agent memory
Retrieval is task-aware and capped so it does not flood an agent's context. Adds native runtime adapters, portable guidance, and JavaScript and Python SDKs.
New
Connection lifecycle and profile photos
Owners can archive, restore or delete connections without losing saved work, and upload, replace or remove a profile photo. Public profiles get quieter activity cards.
Improved
Calmer handoff diagram on mobile
The landing-page handoff diagram is quieter and fits short screens, with compact agent cards on mobile.
Improved
Shared controls and animated handoffs
The landing page explains scoped sharing with a looping save-and-recall illustration and shows 21 agent products with locally bundled logos. Motion pauses offscreen.
New
New interface system and profile layouts
One graphite and mint design system now covers the app, public pages, email and social cards. Public profiles offer two layouts, Living studio and Field notes, with up to three written highlights.
Under the hood
Stricter release smoke checks
Before issuing a temporary read credential, the release check verifies the exact candidate revision, image digest and provider-free worker configuration.
New
Reviewed selective sharing
Describe which context belongs with a team or another agent, preview bounded excerpts, and approve the exact text. Shares are fixed snapshots, so new source text needs a fresh review.
Improved
Clearer beta pricing
Hosted beta pricing shows 60 days free with an invite instead of $0, and says access starts after email verification. The site states the plan to sponsor open-source projects from revenue.
Under the hood
Fix worker preparation retries
Re-running mail or billing worker preparation updates existing schedules correctly, without enabling delivery or billing.
Under the hood
Separate connection for operator setup
Operator bootstrap and maintenance use a scoped provisioner connection instead of the restricted request pool.
Under the hood
Fix scheduled worker configuration
Mail and Jev billing jobs get the correct Cloud Run execution settings and database attachment.
Under the hood
Explicit operator workspace for Jev
The operator's default Jev workspace is named explicitly, alongside the exact owner email.
New
60-day friend invites and onboarding emails
Invited friends get 60 days of unlimited core knowledge transfers from verified redemption, with a workspace countdown and an extension link. Adds an onboarding email series and optional Jev processing.
Security
Email-only console sign-in
Hosted sign-in and invite screens use email verification only. Agent access tokens can no longer open a console session, though agents still use them.
Improved
Invite screen copy
The invite screen now reads: “No card required. Verify your email to get started.”
Security
Invite-only hosted beta
Hosted sign-up requires a single-use invitation bound to the recipient's email, plus inbox verification.
New
Hosted launch
The hosted service launches with workspace-scoped credentials, isolated workers, billing, privacy and support pages, and an activity view that separates observed memory calls from tasks agents report. Guided onboarding advances as setup succeeds.
Improved
A helpful page at /mcp
Opening the MCP URL in a browser explains how to paste it into a ChatGPT, Claude or Grok connector instead of showing a blank error.
New
Sign in from chat connectors
Engram runs its own OAuth authorization server. Claude, ChatGPT and Grok connectors sign in with the same email link as the console.
New
Share slices from a personal brain
Notes land in a personal brain first. Share rules copy matching slices into a team brain, and a share can be undone.
September 2026
Under the hood
README as the project front page
The README explains what Engram is, how to self-host it and how to ask for hosted access.
Under the hood
Keep Jev on across deploys
Deploys keep the Jev provider key mounted, so a routine release cannot switch Jev off.
Under the hood
Restore proven deploy settings
Deployment returns to the last working Cloud Run configuration so the Jev release could go live.
Under the hood
Deploy before provider access is ready
Deploys skip provider keys until access is granted. Jev stays off and memory keeps working.
Under the hood
Optional provider keys
Releases mount the Jev and embedding keys only when they exist; otherwise the gateway runs without them.
Under the hood
Run the Jev tests in the build
Tests preload the database setting they need, so the Jev suite runs in Cloud Build.
Under the hood
Fix the provider key build step
The build escapes its shell variables so it can create provider key placeholders.
New
Jev memory cycle and nightly consolidation
Optional Jev judgments route topics, classify notes and rerank recall, and a nightly job consolidates memory. Saves still work when Jev is down.
New
Graph view, live usage and provenance
The console shows which agent first wrote a page and who has added to it since, a graph of pages and their links, and live usage.
Improved
Console redesign
The console drops its paper-and-ink theme and Japanese vocabulary for a plainer design and plain terms.
Under the hood
Record which page each call touched
Each call records the page it touched: the groundwork for showing where a memory came from and who contributed to it.
Fixed
Long topic pages stay current
Topic pages are bounded and roll over, instead of recall cutting the answer off mid-sentence and returning an old state.
Under the hood
Fix the deploy trigger
The automatic deploy names a log destination, so the trigger can run.
Under the hood
Notes on a hidden test bug
Adds notes on how a global test mock removed exports and why a local container run hid the problem.
Fixed
Test mocks keep real exports
Module mocks keep the exports they do not replace, fixing the first automated build failure.
Under the hood
Deploy on merge, gated on tests
Merging to main deploys automatically, and only after the test suite passes.
Fixed
No failed connection after idle
An idle Engram no longer fails an agent's first connection while it wakes up.
Fixed
No lost saves from concurrent writers
Concurrent saves to the same page verify what they wrote, so one agent's entry is not dropped by another's.
Under the hood
Headless agent permissions guide
Documents the tool permissions headless agents need to reach Engram, and scopes the OAuth work.
Fixed
Saves to one page run in order
Saves to the same page run one at a time, so two agents appending together no longer drop an entry. Tool counts in the docs were corrected.
New
remember and recall
Adds the remember and recall tools, short tool descriptions, response size caps and MCP server instructions.
Under the hood
Bridge setup from npm
The wiring guide drops its clone-and-run workaround now that the bridge is published on npm.
Under the hood
Bridge command path
Fixes the bridge's command path so the published package matches the repository.
Under the hood
Bridge package links
The npm package links to the correct repository, homepage and issue tracker.
Under the hood
Ready for other people to run
Removes an unused cloud dependency, corrects the package description and adds CI that runs the tests on every change.
Improved
One brain
Engram collapses to a single brain with one tool surface, and forgetting a page can be undone.
New
A browser console
Adds a browser console to search, browse, read and quickly capture memory. The gateway serves it with no build step.
August 2026
New
MIT license and stateless MCP
Engram is MIT licensed and supports the 2026-07-28 stateless MCP specification.
New
Engram v0
A scoped MCP gateway over gbrain, with per-agent tokens and a separate database for each scope.
New
First commit
The first gateway: MCP over HTTP with per-agent bearer tokens, an audit log, an admin CLI and a docker-compose self-host path.