Agentmemory changelog: what's new each month
Every stable Agentmemory release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 3 months covered; the current month updates daily.
October 2026 · so far
1 release: 0.9.30
October brought significant security and reliability improvements to Agentmemory, with authentication now enabled by default and new durable capture capabilities to protect against data loss.
- AGENTMEMORYSECRET environment variable now optional; a secret is auto-generated if unset.
- Write requests must include Content-Type: application/json header.
- Origin header validation is stricter; only loopback, the viewer, or allowed origins are accepted.
Highlights
- Authentication is now on by default with automatic secret generation for new installations.
- Durable capture system ensures events are safely stored and retried even if the server goes down.
- New status endpoint and Health tab provide visibility into system state with live monitoring.
- Import and export paths are now confined to designated directories for improved security.
- Optional Redis backend support for distributed state management.
New
- Durable capture with persistent inbox, event IDs, retries, dead letters, and offline hook spool.
- Capture drain command to recover pending events when the server is back online.
- GET /agentmemory/status endpoint for system status queries.
- Health tab in the viewer showing system status.
- Vector pending log that survives force-kill and resumes backfill across restarts.
- Optional Redis backend via AGENTMEMORYSTATEBACKEND environment variable.
- Viewer rebuilt with snapshot plus live stream event architecture supporting 12 tabs.
Improved
- Writes now require Content-Type: application/json header and stricter origin validation.
- Import and export operations are confined to the data directory or specified import root.
- Upgraded to iii engine 0.22.1 with preserved builtin worker names in custom configurations.
- Viewer performance improved with snapshot-based plus live stream event model.
Fixed
- Vectors now persist through force-kill and backfill resumes correctly across server restarts.
August 2026
1 release: 0.9.29
August brought major expansions to Agentmemory's integrations and capabilities, with new connectors for Cursor, Devin, and DeepSeek, plus improvements to core memory and graph features.
Highlights
- Cursor plugin now available with 7 auto-capture hooks, 17 skills, and full MCP support.
- Devin integration lets you connect your Devin agent and use Agentmemory skills via /agentmemory commands.
- Knowledge graph extraction now works without requiring an LLM API key.
- Memory provenance tracking records where every piece of information came from.
- New skills for memory discipline and turning corrections into confidence-weighted rules.
New
- Cursor plugin with IDE integration and cursor-agent CLI support.
- Devin connector with native hooks and plugin manifest for /agentmemory:<skill> commands.
- DeepSeek Harness connector with auto-capture via --with-hooks.
- Memory discipline skill to help decide when to recall and save information.
- /lesson skill to convert corrections into confidence-weighted rules.
- Keyless graph extraction for populating the knowledge graph without an LLM key.
- AGENTMEMORYLLMNOTHINK environment variable to skip hidden thinking during graph extraction.
Improved
- Claude Code plugin feature parity reached with connect pi.
- Provenance tracking now stamps the origin channel at capture, save, and import time.
- connect pi command now actually installs the extension as expected.
July 2026
1 release: 0.9.28
July brought critical stability fixes and new tooling for Agentmemory, addressing memory leaks and reconnection issues while adding convenient demo capabilities and expanded API functionality.
Highlights
- Fixed a cross-agent memory leak that could occur under isolated scope settings.
- Resolved hook crashes when receiving null payloads from code execution tools.
- Eliminated engine conflict hangs by improving WebSocket reconnection recovery.
- Added one-command demo mode for seeding and recalling memory without extra setup.
- Session summaries are now available via the GET /agentmemory/sessions endpoint.
New
- One-command demo mode (agentmemory demo --serve) for testing seed and recall workflows.
- OpenCode connect adapter for expanded integration options.
- Session summaries returned by GET /agentmemory/sessions API endpoint.
- 15 new self-updating skills documenting the entire system.
Improved
- Onboarding agent picker now single-sourced from adapters for consistency.
- Engine version gating in onboarding is less restrictive.
- Engine conflict recovery provides guidance for resolution.
- Onboarding statistics no longer drift from actual state.
Fixed
- Cross-agent memory leak via /agentmemory/context under isolated scope.
- Hook crashes caused by null payloads from Codex CLI and Claude Code.
- Engine conflict WebSocket reconnection hangs.
- Onboarding stat synchronization issues.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Agentmemory: pricing, features and alternatives · All changelogs
