EN
Back to the archive

The archive · AI & Models · Product decision · 2025–2026

Red Planet HQ bets open-source user-owned memory becomes the shared brain across AI apps

Red Planet HQ launched CORE on HN as a shareable, user-owned memory graph for any LLM, then grew it into an open-source personal AI OS.

Red Planet HQ (RedPlanetHQ)

The betThat memory, not models, is the moat: users will adopt an open-source, self-hosted relational memory vault any LLM can share, and anchor an always-on personal AI OS.Live

What the business is

Open-source, self-hosted 'personal AI OS' built around a shareable temporal knowledge graph: CORE indexes a user's email, meetings, GitHub, Linear, Slack and AI chats, then an always-on assistant (voice, scratchpad, chat or WhatsApp/Slack/Telegram) plans and delegates coding, browser and terminal work to Claude Code or Codex sessions on the user's own subscription.

How it started

Red Planet HQ (GitHub org RedPlanetHQ), the team that had built the SOL personal assistant, kept hitting the wall the launch post opens with: 'each AI app remembers me in its own silo. ChatGPT knows my project details, Cursor forgets them, Claude starts from zero', so users re-explain themselves dozens of times a day. They concluded the missing layer was memory, and that it had to be open and user-owned: CORE ('Context Oriented Relational Engine') stored every fact as a versioned statement with who said it, when and why in a reified temporal graph, retiring contradictions instead of wiping them, so any LLM (ChatGPT, Cursor, Claude, SOL) could share one queryable context.

What happened

Launched as a Show HN on 2025-07-01 with a hosted free tier, demo video and docs, drawing 112 points and 42 comments; founders answered the skeptical thread in detail. Commenters asked why a graph beat 'a simple text file accessible by MCP' or a plan.md, and compared CORE with Zep's Graphiti; the founders replied that static files cannot answer relational or timeline questions ('what medications have I tried, why did I stop each one, and when?') and positioned CORE for individual users needing portability and full explainability across Claude, Cursor and Windsurf, versus Zep's B2B, platform-locked memory. By the 2026-09-04 fetch the repo had evolved from memory-graph showcase into 'Your Personal AI OS', adding tasks, scratchpad, voice, WhatsApp/Slack/Telegram messaging, coding-agent delegation and 50+ MCP connectors.

No ending yet — it is still running.

Background

CORE, by the GitHub organization RedPlanetHQ, launched in July 2025 as an open-source 'memory vault': a shareable, user-owned knowledge graph that any LLM — ChatGPT, Cursor, Claude, their own SOL assistant — could query for the same persistent context instead of each app remembering the user in its own silo. By September 2026 the same repo pitches a full 'Personal AI OS': a temporal knowledge graph indexing email, meetings, GitHub, Linear, Slack and conversations, with tasks, scratchpad, voice, WhatsApp/Slack/Telegram messaging and Claude Code/Codex delegation.

The founders, the team behind the SOL assistant, opened their Launch HN with the pain: 'each AI app remembers me in its own silo. ChatGPT knows my project details, Cursor forgets them, Claude starts from zero,' forcing users to re-explain themselves dozens of times a day. The bet was architectural — memory, not models, is the moat. CORE stored every fact as a versioned statement with who said it, when and why (a reified, temporal graph), retiring contradictions instead of deleting them, so an assistant could answer 'what did I say, and when did it change?' with full provenance.

The project launched on Hacker News on 2025-07-01 with a hosted free tier and drew 112 points and 42 comments. The thread was demanding: commenters asked why a graph beat 'a simple text file accessible by MCP' or a plan.md, and compared it with Zep's Graphiti. Founders answered that static files cannot answer relational or timeline queries — 'what medications have I tried, why did I stop each one, and when?' — and positioned CORE for individual users who need portability, contradiction handling and explainability rather than B2B platform lock-in.

By the 2026-09-04 repo fetch, CORE presented as 'Your Personal AI OS', still open source and self-hosted in Docker or Railway, with live docs at docs.getcore.me, a claimed 88.24% average on the LoCoMo memory benchmark, CASA Tier 2 certification, a Mac app waitlist, and 2.0k stars, 189 forks and 1,210 commits. It connects to 50+ apps through MCP and can run entirely on a user's existing Claude or Codex subscription. The material discloses no funding, revenue or user counts; the bet is still running.

What has to be true

  • Verifiable traction: Launch HN on 2025-07-01 drew 112 points and 42 comments, and the repo carried 2.0k stars, 189 forks and 1,210 commits when fetched 2026-09-04.
  • The bet is explicit and falsifiable: relational, versioned, provenance-tracked memory beats sticky-note memory, argued in-thread against plan.md/text-file critics.
  • A visible pivot arc: launched as a memory graph for LLMs in July 2025 and by September 2026 presents as an always-on personal AI OS built on that graph.
  • Adoption signals are in the material: 50+ app connectors, coding-agent delegation, a Mac app waitlist and live docs/changelog, with no funding or revenue claims to verify.

What can be applied

Betting graph memory against 'just use a text file' means launch threads demand proof of relational, temporal questions plain files cannot answer; ship that demo or simpler defaults win.

Aftermath

As of 2026-09-05 CORE is live and building in public: the repo (2.0k stars, 189 forks, 1,210 commits) now leads with 'Your Personal AI OS' — an always-on assistant keeping a temporal knowledge graph across email, meetings, GitHub, Linear, Slack and chats, with a gateway that runs Claude Code/Codex sessions on the user's own subscription or Docker/Railway. Docs and changelog are live and a Mac app is waitlisted; the team publishes LoCoMo (88.24%) and security (CASA Tier 2) claims. No funding or revenue is shown; the memory-graph bet has widened into an early-adoption personal-OS product.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases