Product updates, new features, and improvements.
May 8, 2026
Repositioned as operational memory for AI teams
new improved
- New positioning — hero updated from “persistent memory for AI builders” to “operational memory for AI teams”, reflecting what the product actually does: decisions, tasks, handoffs, and pull-based sync.
- Proof-flow section — new “See it in action” section shows cross-AI continuity in three steps: save in ChatGPT, pull in Claude Code, continue without re-explaining.
- 4 new tools documented —
pull, register_agent, list_agents, and check_inbox added to /docs. Tool count updated from 12 to 16.
- Comparison update — new “Agent inbox / pull” row on /compare. Mem0 MCP status updated. Agent coordination now shows “Registry + inbox + pull”.
- Features copy — agent orchestration card updated with
pull() example. Features heading reflects operations focus.
May 7, 2026
Native pull — cross-AI sync without copy-paste
new
- Native pull — new
pull() tool lets any agent fetch decisions, tasks, context, and learnings written by other agents since its last session. Save a decision in ChatGPT, run pull({agent: "CC-eng"}) in Claude Code — it sees the decision without a pasted prompt.
- Cross-platform continuity — the core mechanism for multi-AI workflows. No polling, no webhooks — agents pull filtered diffs on demand.
- PII stripping — collaborator emails stripped from MCP responses for privacy.
May 7, 2026
Consistent navigation, site cleanup
improved
- Unified navigation — all pages now share the same navigation bar. No more menus that change between pages.
- Seven primitives — features section now correctly shows all 7 memory types including instructions.
- Comparison link — homepage links to the dedicated /compare page instead of duplicating the table.
May 6, 2026
Agent orchestration, comparison page, onboarding improvements
new
- Agent orchestration — new agent registry, event log, and
check_inbox tool. Register agents at boot, track activity, and poll for relevant events. Foundation for cross-agent coordination.
- Comparison page — new /compare page with detailed feature comparison across Engrams, ChatGPT Memory, Claude Projects, Mem0, and Zep.
- Marker-based supersede — new
marker parameter in remember guarantees one memory per marker per project. Old versions are automatically archived.
improved
- Faster boot — summary mode reduces boot payload from 85k to 17.8k characters for research agents. New
agent_role parameter on boot.
- Task ownership —
list_tasks now supports owner filter. Agents see only their assigned work at boot.
- Better onboarding — new "Try it — 60 seconds" section on setup page with paste-ready prompt that demos save+recall round-trip.
May 5, 2026
Codex support, ChatGPT MCP, site polish
new
- Codex support — Engrams now works in OpenAI Codex via Streamable HTTP MCP. Add the URL in Codex settings and your agents share memory with Claude, ChatGPT, and Cursor.
- ChatGPT MCP connector — updated setup to use ChatGPT’s native MCP connector instead of GPT Actions. Create a ChatGPT Project with Engrams attached for auto-boot.
improved
- Case-insensitive projects — project names now match regardless of casing. “Engrams” and “engrams” find the same project.
- Skills install — install commands now point to the public engrams-skills repo.
- Site copy — version badge updated to v1.1, memory type count corrected to 7, broken links fixed.
May 2, 2026
New homepage — rewritten for developers
improved
- New positioning — “Built for developers and AI builders” replaces generic consumer-facing copy. Hero, features, FAQ, and pricing all rewritten.
- 6 verified features — features section now shows only what’s actually shipped: typed memory, cross-platform access, shared projects, task versioning, multi-agent alignment, and decision analysis.
- FAQ section — addresses “how is this different from Claude Projects / MemoryPlugin / native memory?” directly.
- False claims removed — “automatic saving”, “one line setup”, “export in any format”, and “self-hosted” no longer appear on the site.
April 30, 2026
Multi-agent alignment & automatic task reconciliation
new
- Align tool — new
align command shows what changed across all your projects since a given timestamp. See new memories, task updates, urgent items, and drift warnings in one view — designed for workflows where multiple AI agents work in parallel.
- Automatic task reconciliation — when a session episode is saved with completed items, matching tasks are automatically marked as done. No more stale task lists after handoff.
- Boot validation — boot and project loading now detect task drift (episode says done but task still open) and urgent date triggers, shown as warnings before the dashboard.
April 29, 2026
ChatGPT MCP verified & project creation fix
improved
- ChatGPT integration verified — Engrams now works end-to-end in ChatGPT via Settings → Apps (MCP). Full boot, recall, remember, and task management confirmed working. Connect your Engrams MCP URL in ChatGPT to use the same memory across both Claude and ChatGPT.
- No more accidental project creation — Loading or searching a project that doesn’t exist no longer silently creates it. Only saving a memory or creating a task will create a new project. This prevents empty “ghost projects” from appearing in your project list.
- Clearer auth errors — Invalid API keys now consistently return an error message across all tools instead of ambiguous “no projects found” responses.
April 28, 2026
Shared Projects reliability & dashboard stability
improved improved
- Shared project name collisions fixed — If you had your own project with the same name as a shared one, memories could end up in the wrong place. Shared projects now take priority, and invites are blocked when a name conflict exists.
- Dashboard no longer shows 0 memories — A rare data condition could cause the dashboard to display 0 memories for all projects. Fixed at the database level with additional safeguards.
April 20, 2026
Shared Projects — up to 2 collaborators
improved
- 2 collaborators per project — Raised from 1 to 2. Invite two people to the same project, or use two of your own accounts (for example, a work and personal inbox) to reach the same memory.
- Pending invites no longer block — The cap now counts only accepted invites. A pending invite that was never accepted won’t take up a slot.
- Friendlier duplicate-invite handling — Inviting the same email twice now returns a clear message instead of an error.
April 19, 2026
7 platforms, Skills, Shared Projects, performance
new new new improved improved removed
- Shared Projects — Share a project with another person via email. Both of you get the same project memory in your Claude or ChatGPT. Tell your AI: "Share this project with name@email.com".
- ChatGPT support — Engrams now works with ChatGPT as an app. Same setup as Claude — paste the URL in Settings → Apps, done. Same memory, different AI.
- Smarter onboarding — When you connect Engrams to an existing project, it automatically offers to save your existing context. No manual import needed. Your Project Instructions, knowledge, and context flow into Engrams in one conversation.
- Identity on memories — Every memory now tracks who created it. In shared projects, you can see what your collaborator saved. Now consistent across all memory types including tasks and profiles.
- Faster boot & dashboard — Replaced per-project query loops with bulk database operations. Boot, dashboard, and project listing are now significantly faster, especially with many projects.
- Codex support — OpenAI Codex now connects via MCP. Add two lines to
~/.codex/config.toml and you're set.
- Cursor support — Add one entry to
~/.cursor/mcp.json and restart. Full MCP access.
- Lovable support — Add Engrams as a personal connector in Lovable settings. Same URL, same memory.
- Gemini support — Google Gemini connects via MCP URL or function calling. Setup guide in the dashboard.
- Skills — Ready-made Claude Code skills with persistent memory:
/research, /pm, /review. Install with one command. Browse skills →
- "Why this stays relevant" — New section on the homepage explaining why cross-platform memory is a structural advantage, not a temporary gap.
- Unshare — Remove a collaborator from a shared project via the dashboard.
- Import removed — The manual import flow (export prompt + paste) replaced by automatic onboarding.
- 12 MCP tools — Up from 9. Added boot, submit_review, and share_project.
- Simplified Claude setup — No instructions to paste. Just connect the URL and go. MCP handles everything automatically.
- Cross-project recall — Search across all your projects at once. No need to specify which project a memory is in. Single database query, scales to any number of projects.
- UI cleanup — ChatGPT setup tab, simplified Claude/Code setup (no instructions needed), consistent footer navigation, shared projects visible in dashboard.
- Comparison table — Added "Survives chat context limits" row. Claude Memory and ChatGPT Memory both lose context when conversations get long. Engrams doesn’t.
- Memory types guide — New reference section on the guide page explaining all 6 memory types: profile, context, decision, learning, episode, and instruction.
- Telemetry dashboard — Admin-only analytics: daily active users, retention (D1/D7/D30), onboarding funnel, endpoint distribution, and 30-day trend charts. No content is logged — only event types and counts.
- Skill template — Build your own Engrams-powered skill in 4 steps. Start from the template →
- New hero — "Build without starting over." New positioning focused on what you can build, not what we fix.
- Smoother onboarding — "Setup takes 30 seconds" label, test connection button on setup page, clearer copy.
- Security page — New /security page documenting encryption, access control, data isolation, infrastructure, and what we don’t store.
April 10, 2026
Session handoff, task management, OAuth
new
- Session handoff gate — Controlled session summaries with authorization
- Task CRUD — Create, update, list, and track tasks with priority and status
- Decision analysis — Analyze decisions using stored memory context
- OAuth 2.1 — Standard auth flow for third-party integrations
- Verification guide — Step-by-step guide to confirm Engrams is working