Getting Started
What's new
The biggest changes to HyperMemory, newest first
October 2026
Organization memory, the new CLI, notes and the new platform.
Organization memory in every recall
Members of an organization now get its memory with their own on every hm_recall, searched in parallel and returned under organization_memory. Organization graphs are read-only for members; admins write to them with organization write mode. Projects can be bound to one organization, shared by the plugins and the CLI.
Agent skills with supporting documents
Upload a skill as a .zip with SKILL.md and its reference documents. Organization skills reach every member’s agents, and HyperMemory’s domain skills — software development, marketing, legal, memory quality audit — are now published system skills. See Skills Distribution.
The new hm CLI
The CLI now:
- sets up your agents with
hm setup claude | codex | hermes | kimi | openclaw | all, plushm doctorandhm uninstall; - signs in through your browser or with an API key kept in your keyring;
- has a command for every tool, with
--jsonoutput and stable exit codes for scripts and agents; - binds projects to organizations with
hm organd installs published skills withhm skill install.
HyperMemory Notes for iPhone, iPad and Android
The new notes app captures text, dictation, photos, video, voice, meetings and web pages, reads them with AI, and adds them to your memory — offline-first, with chat, sharing and push notifications. Also on the web, with a Markdown export of every note. See HyperMemory Notes.
A faster, sturdier platform
HyperMemory now runs on a new Rust platform:
- Every write is idempotent — retry with the same
X-Request-IDand nothing is duplicated. - Every response carries a request id.
- Errors use one consistent envelope with
retryableandRetry-After. See Errors & limits. - Large deletions run in the background, and orphan analysis no longer slows down big graphs.
New pricing
Plans are now usage-based, with one pool of queries and nodes, plus documents, meetings and notes. Paid plans continue past their allowance as overage instead of stopping. From 23 October 2026: Free gives a one-time 6,000 queries + nodes; Basic is $15 a month with 15,000, Pro $30 with 50,000, and Business $79 with 110,000, 3 seats included and extra seats at $49. Overage is $0.61 per 1,000 queries + nodes and $0.32 per document. See Plans.
Governance
The new Governance section covers HyperMemory’s security, privacy and compliance programme, including the control catalogue and subprocessor register.
August – September 2026
Plugins, organizations, token usage, meetings and new tools.
Plugins for Claude Code and Codex
Install HyperMemory from the plugin marketplaces. The plugins add the MCP connection, the HyperMemory skill, hooks that make the agent recall and save every turn, and a background memory-writer agent.
Organizations
Create an organization, invite admins and users, and buy seats for members. Organizations get a shared memory graph, organization skills and organization-wide token usage.
Token usage
See every agent’s AI token usage — by model, provider, tool and kind of work — in one dashboard. Organizations see it across members and can import exact usage from OpenAI, Anthropic and OpenRouter accounts.
Meetings
Connect Google Meet or Microsoft Teams and HyperMemory brings your meetings into memory. See Meetings.
Files
Uploads support audio and video, are processed by durable background jobs, and can be reparsed with an extraction lens for engineering, software, legal, financial and other documents.
New tools
hm_get_nodes— read up to 50 nodes in full.hm_get_chat_context— reload what one conversation touched.hm_tabular— query uploaded spreadsheets.hm_skill— fetch agent skills.hm_tokens— report token usage.
hm_add_relationships is available over MCP, and hm_update and hm_forget can target a single edge with edge_id, with optimistic concurrency. hm_store writes a node and up to 99 relationships in one transaction, and if_exists controls what happens when the key exists.
Changes that affect API clients
| Change | What to do |
|---|---|
hm_list_orphans pages with cursor / next_cursor. offset is rejected. | Pass next_cursor back as cursor. |
GET /api/v1/memory/export answers 410. | Use POST /exports, then GET /exports/{id}/download, or hm export. |
MCP request bodies are limited to 64 MiB (about 47 MiB of file data in hm_upload_file). | Upload larger files from the dashboard. |
A relationship in hm_store that points to a missing node fails the whole call. | Store targets first, or add the relationship later with hm_add_relationships. |
hm_update replaces a node’s data object. | Send the complete data object. |
| File uploads are available on every plan, counted against the monthly document allowance. | — |