Can I use Obsidian as long-term memory for Claude or ChatGPT?
Yes. A vault plus a community plugin or a small MCP server gives one person durable, local, Markdown memory that an AI can read and write. It holds up well until a second person needs part of what is in it, because a vault is shared as a whole and a company is not.
What a vault is genuinely good at
Obsidian keeps plain Markdown files on your own disk. You can link notes to each other, search them, keep version history, and move the whole folder somewhere else next year. Nothing sits in a vendor's database that you cannot read without the vendor.
The AI setups people build on top of it are real, not toys. Some servers read the Markdown files directly and need no plugin at all, so the vault answers even when Obsidian is closed. Others run through the Local REST API community plugin, which needs the app open. Either way your AI can search the vault, read a note, and write a new one, and the files stay yours.
If you are one person keeping your own working notes, that is a good stack. HexaHQ is not trying to take it away from you. Keep drafting where you think.
A personal vault, used by an organization
None of these are bugs. They are the honest edges of a tool built for one person's thinking, asked to carry a company's operating rules.
Sharing is all or nothing
Obsidian's own collaboration documentation says fine-grained permissions are not supported yet, and that every collaborator gets the vault owner's permissions except the ability to invite other collaborators. So the day one note should reach Sales and another should not, the answer is a second vault.
A second vault is a second copy
Split the vault and the same rule now lives in two files. They agree the day you split them. Six weeks later, one has the current discount ceiling and the other does not, and nothing in either file tells a reader which one is current.
Two notes that mean the same thing look fine
Version history tells you a file changed. It cannot tell you that "we do not discount past 15%" and "anything over 15% goes to Finance" are the same rule written twice, with one of them out of date. A person has to notice, and usually nobody owns noticing.
A note does not say when it applies
"Approve it" is a fine sentence in a note and a poor instruction for an AI. Which customers, which region, which deal size, which of the three people who might ask? Prose written for a reader keeps that in the reader's head.
Reading a vault is not doing the work
Vault memory can tell your AI the rule. It cannot update the CRM record through your own account, pause for approval when the change is consequential, and leave a record of who asked for it.
The stack is a person
Someone chose the plugins, wired the server, and keeps both working through upgrades. While that person is busy, everyone else's AI stops getting better.
Source: Obsidian Help, "Collaborate in a shared vault", checked 28 August 2026.
HexaHQ keeps the settled part, and decides who receives it
Obsidian is a strong personal memory system. The harder problem begins when the memory has to become organizational. A company needs to know which version is current, who should receive it, where it applies, and which parts should stay personal. HexaHQ is built for that layer.
In practice that means four things a folder of files does not do on its own. A concept has one current version, so a correction lands in one place instead of in whichever copy you happened to open. Sharing is a separate decision from writing, so the same canonical document can reach one team and not another without being duplicated. Applicability is recorded with the rule, so the AI can tell whether it binds this deal. And the same context that guided the decision is what the AI uses when it goes and does the work in a real system.
Corrections get captured while real work is happening, and a person reviews what was saved before it spreads any further. You are not stopping to maintain a knowledge base by hand, and nothing becomes an organization wide rule because one person said it once in a chat.
Sharing a vault is not the same as choosing an audience
Sharing a vault is not the same as deciding which context should reach which person. HexaHQ can keep one organizational concept while exposing only the relevant and permitted parts to each person and their AI.
Each piece of context has an audience. Some stays personal, like how you want your own drafts written. Some reaches one team or project group. Some is approved for the whole organization. Some is packaged for a client, a contractor, or an advisor outside the company. A person's AI inherits that same boundary: it can use what its person is meant to have, plus that person's own working context, and it does not quietly pick up the rest.
That is what makes one canonical layer safe. Without it, "share the knowledge" and "share the vault" are the same action, and companies correctly decide not to.
The security questionnaire rule
A solutions engineer keeps a vault. It has her call notes, a page on how she writes follow-ups, and a page listing which security questionnaires the company will fill in and which ones Legal has said to decline. Her AI reads all of it, and her answers have been good for a year.
A teammate gets the same questionnaire question. Her options are to hand over the whole vault, paste the paragraph into a chat, or answer it again herself. The first exposes her call notes. The second creates a copy that will go stale. The third is the reason she is the bottleneck.
With HexaHQ, the questionnaire rule is one document shared with the Solutions team. Her writing style stays personal. Her teammate's AI answers the next questionnaire the way Legal wants and never sees a single call note. When Legal changes the rule, it changes once, and both AIs start from the new version on their next task.
Obsidian and AI memory, answered
Can I use Obsidian as long-term memory for Claude or ChatGPT?
Yes. A vault of Markdown files plus a community plugin or a small MCP server lets an AI read, search, and write your notes, and the files stay on your own disk. For one person it is a good setup. It stops being enough when a second person needs part of what is in the vault, because a vault is shared as a whole.
Do I need an Obsidian plugin to connect my AI to a vault?
It depends which server you pick. Some read the Markdown files straight off disk and need no plugin, so the vault answers even when Obsidian is closed. Others go through the Local REST API community plugin, which means Obsidian has to be running for your AI to reach anything.
Can I share part of an Obsidian vault with one teammate?
Not with per-note permissions. Obsidian's collaboration documentation says fine-grained permissions are not supported yet, and that collaborators get the vault owner's permissions except the ability to invite other collaborators. The usual workaround is a second vault, which means a second copy of anything both groups need.
Does HexaHQ replace Obsidian?
No. Obsidian is where you write and think. HexaHQ keeps the distilled version your organization wants AI to apply: the current rule, who it reaches, when it applies, and which actions it governs. Plenty of people keep drafting in a vault and share only the settled part through HexaHQ. More on how company knowledge works.
Is HexaHQ a note-taking app?
No. HexaHQ is the layer between people, their AI, and the systems where the work happens. It holds canonical knowledge and reusable workflows, works out which slice a task needs, controls which people and which AI receive it, and runs real actions under your organization's approval policy.
What happens when the person who built the vault stack leaves?
With a self-assembled stack, the plugins, the server, and the conventions leave with them. HexaHQ artifacts belong to the organization, so a teammate connects their own AI client and starts from the shared context that is already there.
Related comparisons
Keep the vault. Share the rule.
Give one teammate the part of your context they need, and nothing else. Free to start.