Blog
Published: October 23, 2026
We make Harbor, one of the options below. Everything here about Supermemory is from its own post, changelog and repository, read on 23 October 2026. The other tools are from their own docs and pricing pages, read in October 2026.
Supermemory stopped its hosted Company Brain on 9 September 2026 and released it as open source on 25 September. To keep the same Slack bot, deploy that build on your own Cloudflare account. If your coding agents also read the brain, the open-source build cannot serve them today. Unblocked, Cognee and Harbor can.
What happened to Supermemory's Company Brain? #
In a post dated 10 September 2026, Supermemory's founder wrote that the company had stopped the Company Brain the day before, along with Nova, its personal knowledge app. Everyone who was charged was refunded. The reasons given are focus, and not competing with the developers who build on its memory API, which is where the company is now putting its time.
Two things did not stop. Supermemory's MCP server and its plugins for Claude Code and other agents moved to the developer platform and still work for personal use. And on 25 September, the changelog announced the Company Brain itself as open source: supermemoryai/company-brain, under Apache-2.0, rebuilt to run on Cloudflare.
What the open-source build keeps, and what it drops #
The README describes it as the whole product, open-sourced. It is a teammate in Slack that remembers the channels it is in, answers from them, and acts in GitHub, Linear, Notion, Google Workspace and other tools over MCP. Reading follows Slack's own access: a private channel stays private. When a request needs a tool only a teammate has connected, that teammate gets an approve or deny card first. On the brain's home page, memories export as Markdown.
What it drops is the way in for coding agents. The hosted Company Brain could be read from Claude Code, ChatGPT or Cursor over MCP. The self-hosted build's guide says it exposes no MCP server and no API for asking it questions, so there is “no endpoint to point a coding agent at.” If your engineers asked the brain from their editor, that habit has nowhere to go.
To run it, you need a Cloudflare account and two keys: a Supermemory API key, where the memory lives, and a model key from Anthropic, OpenAI, Google, xAI or OpenRouter, billed by that provider directly. It runs on Cloudflare's free plan. The README recommends Workers Paid at $5 a month, because the free plan allows 50 outbound calls per request and can cut long answers short.
1. Self-host the open-source Company Brain #
- Pick it because: it is the product you had, with the same Slack habits, and it can still act in your tools, which none of the options below do.
- Watch out for: coding agents cannot read it. Memory still lives in a Supermemory account. What it learns from a channel is saved without a review step; the approve or deny cards cover its actions, not its memory.
- Price: free software. Cloudflare at $0 or $5 a month, plus your model usage and Supermemory API usage.
2. Unblocked: answers in Slack, and in every major agent #
A hosted context engine for engineering teams. It reads about 45 sources, from code and pull requests to Slack, Microsoft Teams, Confluence, Notion, issues and CI, and answers in Slack, on the web and through a CLI. Coding agents reach it over an MCP server and skills, in Claude Code, Codex, Cursor, Copilot, Windsurf and more.
- Pick it because: the widest set of sources and agents here, and it works in Microsoft Teams as well as Slack.
- Watch out for: what it learns, including conventions it mines from pull request reviews, is decided automatically. No review step before a memory is used is documented.
- Price: no free plan. $29 per user a month billed yearly, or $35 monthly, after a 21-day trial.
3. Cognee: open source, with recall in Claude Code and Codex #
Open-source graph memory under Apache-2.0. Cognee Cloud reads three live sources, Slack, GitHub and Google Drive, and answers in Slack with /cognee-recall. Its plugins add memories on every prompt in Claude Code and Codex, and other agents connect over MCP.
- Pick it because: you can run the engine yourself, and recall happens on every prompt rather than when the agent remembers to ask.
- Watch out for: only the workspace owner can connect sources, and capture is automatic, with no review step documented.
- Price: free to run yourself. Cognee Cloud is free up to 1M tokens, then $1 per million tokens plus $5 a workspace.
4. Harbor: the team's Slack decisions in every coding session #
Harbor learns the team's rules from the Slack channels an admin switches on, pull request reviews and docs, plus what agent sessions record. In Slack, /harbor answers a question with a line per cited fact, and only you see it. @Harbor remember, /harbor remember and the Remember in Harbor message shortcut save a decision for the team. In the editor, hooks give each Claude Code, Cursor, Codex or Copilot session the rules that apply to its task when it starts, with an MCP server for the rest.
You set the gate. Manual holds everything for a person. Hybrid, the default, adds routine rules and holds the rest, and a save that contradicts a live rule always waits in Review. Every rule shows how often it was served to a session and how often an answer cited it (which rules got used).
- Pick it because: what was decided in Slack reaches the coding agents, not only the people who ask, and nothing reaches every agent without passing the gate you set.
- Watch out for: it answers and saves; it does not open issues or run code in your tools, as the Company Brain did. It is hosted only, and there is no instant sign-up: every team starts with a 20-minute setup call with us.
- Price: Free costs nothing for your whole team, on one source: one channel, repo or folder. Pro is $24 per person a month billed yearly.
If it was only you, in Claude Code #
If you used Supermemory as personal memory from your editor rather than as a team brain in Slack, you may not need to move. Supermemory's post says its MCP server and plugins carry on for personal use. For other options built for one developer, see Mem0 alternatives for coding agents, which covers Supermemory's plugins next to claude-mem, Basic Memory and Zep.
Side by side #
| In Slack | Coding agents | Learns from | Review before it is shared | Host it yourself | Price | |
|---|---|---|---|---|---|---|
| Company Brain, open source | A teammate that answers and acts | None today | The channels it is in | Actions only, not memory | Yes, Apache-2.0, on Cloudflare | Free software; $0 or $5 a month on Cloudflare, plus model and Supermemory usage |
| Unblocked | Answers, also in Teams | MCP and skills in Claude Code, Codex, Cursor, Copilot and more | About 45 sources | Not documented | No | No free plan; $29 a user a month, yearly |
| Cognee | /cognee-recall |
Plugin on every prompt in Claude Code and Codex; MCP elsewhere | Slack, GitHub, Google Drive, agent sessions | Not documented | Yes, Apache-2.0 | 1M tokens free; then $1 per million plus $5 a workspace |
| Harbor | /harbor answers only you see; @Harbor remember saves |
Hooks at session start and MCP in Claude Code, Cursor, Codex, Copilot | Slack, pull request reviews, docs, agent sessions | A gate you set | No | Free on one source; Pro $24 a person a month, yearly |
How to choose #
- You want the same Slack bot back, acting in your tools, and can run it on Cloudflare: the open-source Company Brain.
- Your team is on Microsoft Teams, or needs the most sources and agents: Unblocked.
- You want open source with recall on every prompt in Claude Code and Codex: Cognee.
- You want what the team decides in Slack to reach Claude Code, Cursor, Codex and Copilot through a gate you set, with served and cited counted per rule: Harbor , set up with you on a 20-minute call.
- It was only ever you, in your editor: Supermemory's own plugins still work.
For the wider field, see the best company brain tools for coding agents and how Slack decisions reach coding agents. How /harbor works in Slack is in the docs. If a line here is out of date, tell us and we will fix it.
Questions #
Is Supermemory's Company Brain shut down?
The hosted Company Brain stopped on 9 September 2026, and Supermemory says everyone who was charged was refunded. On 25 September it was released as open source under Apache-2.0, to run on your own Cloudflare account. Supermemory's memory API, MCP server and coding plugins continue.
Can I self-host Supermemory's Company Brain?
Yes. The supermemoryai/company-brain repository deploys to Cloudflare Workers with two keys: a Supermemory API key, where the memory lives, and a model key from Anthropic, OpenAI, Google, xAI or OpenRouter. It runs on the free plan; the README recommends Workers Paid at $5 a month.
Can Claude Code or Cursor read the open-source Company Brain?
Not today. Its guide says the self-hosted build exposes no MCP server or API for asking it questions, so there is no endpoint for a coding agent. The hosted product had one. Unblocked, Cognee and Harbor each serve coding agents.
What is the best alternative to Supermemory Company Brain?
For the same Slack bot that acts in your tools, self-host the open-source build. For Microsoft Teams or the most sources, Unblocked. For open source with recall in Claude Code and Codex, Cognee. For your team's Slack decisions in every coding session through a gate you set, Harbor.
Does Supermemory's Claude Code plugin still work?
Yes. Supermemory says its MCP server and its plugins for Claude Code and other agents moved to its developer platform and still work for personal use. Only the Company Brain and Nova, its personal knowledge app, were discontinued.