Skip to main content
The persistence of thought

Thoughts,
gently melted
into structure.

You talk, we listen. OpenMind turns the drift of a conversation into a typed knowledge graph. Ask that graph and every answer names the nodes it cites — or tells you it couldn't find it in your graph. No black box. Dalí made sense of dreams; we make sense of your project.

Request an inviteSee how it works
Graph answers cite their sources
Skip to app →
01 — The idea

However tangled your thoughts arrive — they leave as a map.

→ you, talking
Linnea finds a mortar
in grandmother's garden
made of bronze
herbs… which ones?
an evening ritual
grandma's recipe
she grinds something
a childhood memory?
Sonnet 5 · extract · reconcile · persist
→ a typed graph
Linnea
person
Mortar
object
Grandmother's Garden
place
Herbs
concept
Recipe
concept
Bronze
object
Grandmother
person
Evening ritual
decision
02 — Three pillars

Talk it. Map it. Own it.

01

Talk it.

A normal chat. Drop in PDFs, MDs, DOCX — on self-host, audio and video too — OpenMind hears you, matches re-mentions against what you've said before, keeps uncertainty tagged.

02

Map it.

Every message you send is read for typed nodes and edges. Ask it anything: an answer from your graph lists the source nodes it cites, so you can check the work. When the graph can't answer, you're told it isn't there, and offered a clearly labelled answer from the model's general knowledge instead. Six export formats, Mermaid is canonical.

03

Own it.

Self-host with Docker — your Postgres, your vectors, your keys. Bring Anthropic, OpenAI, Ollama, or LM Studio. Stored data stays on your network; with Ollama or LM Studio for chat and embeddings, no model call leaves it either.

03 — The app

A conversation on the left. A living map on the right.

persistence.live/demo/linnea
open interactive demo →
Conversation · Linnea and the Mortar
you
Linnea finds an old bronze mortar in the garden of her grandmother.
+3 nodes, +2 edges
openmind
A bronze mortar in her grandmother's garden — do you know what she used it for?
you
The grandmother left the recipe for herb paste in an old cookbook.
+2 nodes, +2 edges
openmind
So the mortar and the recipe belong together — a family tradition, handed down.
Linnea
person
Mortar
object
Grandmother's Garden
place
Herbs
concept
Recipe
concept
Bronze
object
Grandmother
person
Evening ritual
decision
loading live graph…
Or explore it at scale: a living graph of more than 6,500 nodes from z/OS documentation, with curated questions whose answers show their sources →
04 — The proof

Every graph answer shows its work.

Quoted verbatim from the live demo — nothing staged

How does RACF control access to Zowe resources on z/OS?

RACF controls access to Zowe resources on z/OS by defining resource classes and profiles that specify permissions for users. For example, the ZOWE resource class is used to define access permissions to Zowe-related resources, including API service information, and the APIML.SERVICES resource profile controls access to API service information in the ZOWE resource class, requiring READ access for authorized users.

The nodes do not specify the exact mechanisms or processes RACF uses to enforce these access controls beyond defining resource classes and profiles.

Sources (3 nodes)
APIML.SERVICES(racf_resource)ZOWE resource class(racf_resource)RACF Resource(concept)
One click on a source reveals
APIML.SERVICES
Zowe DocumentationPrimary
versioned_docs/version-v1.28.x/extend/extend-apiml/service-information.mdCC-BY-4.0 · 2d0c447
ZOWE resource class
Zowe DocumentationPrimary
versioned_docs/_version-v2.14.x/user-guide/obtaining-information-about-api-services.mdCC-BY-4.0 · 2d0c447

Every link opens the original file at the exact commit it was extracted from. Not a citation style — a click path.

Check it yourself in the live demo →
05 — What's in the box

Built like a real tool.

Traceable answers
Ask the graph; an answer from it always lists the source nodes it cites. When the graph can't answer, you get a plain "couldn't find it" — and the offer of a clearly labelled answer from the model's general knowledge.
Multi-LLM
Anthropic Sonnet 5 by default, or OpenAI — switch in Settings. Ollama and LM Studio on a self-hosted instance.
Reconcile
Canonical key, then a case- and punctuation-insensitive label match. A re-mention by the same name usually lands on the existing node; duplicates that slip through can be merged.
Attachments
TXT, MD, PDF, DOCX in-process. MP4, MOV, MKV, WebM, audio via self-host worker.
Exports (×6)
Markdown, Mermaid, OPML, FreeMind, PNG, SVG. Mermaid is canonical.
pgvector
Dimension-agnostic embeddings. Swap providers without a migration.
Edge suggestions
Cosine-similarity SQL RPC proposes likely links. You accept / reject, we audit.
Self-host
Docker, your Postgres, your keys; source access by invitation for now. Stored data stays on infrastructure you control, and with Ollama or LM Studio for chat and embeddings so does every model call.
06 — Early access

Want early access?

We're onboarding people one batch at a time. Drop your email and we'll send an invite when a spot opens. No newsletter, no tracking — just the invite.

We'll only use your email to send you an invite. No newsletter, no tracking.