How it works
Your agents work in the same vaults as your people, through one MCP address, under the rules each vault sets.
Connect an agent
Synapse is an MCP server. Add it to Claude, ChatGPT, Claude Code or any client that speaks the Model Context Protocol, at this address:
https://synapse.fraim.dk/api/mcp
The client opens a browser window, you sign in, and you choose what the agent may do. It acts as you: it reaches the vaults you can reach, never more, and you can narrow it further or disconnect it from Settings at any time.
What an agent can do
An agent starts with guide, which explains the vaults and their rules. From there it finds and reads knowledge with list_vaults, search, list_pages and read_page, writes with write_page and edit_page, keeps pages tidy with move_page and archive_page, and joins the conversation with list_comments and add_comment. Teams that use typed pages and saved views query them with query_pages and query_view.
The governance model
Vault tiers. Each person has a tier in each vault: read, edit or admin. An agent never has more than its person.
Per-path policies. A vault's admins set rules per folder or page: who may write there, and whether a change goes in directly or waits for review. Agents can see the rules with list_policies and suggest a change to them with propose_policy; only an admin accepts it.
Proposals. Where a policy asks for review, an agent's write becomes a proposal. A person sees the change and the agent that made it, and accepts or declines it. An agent can follow its own with list_proposals.
Agent scoping. Each connected agent can be limited to some vaults, or to reading only, by its person; an organization can set a policy for which agents may connect at all.
History. Every change is versioned and attributed to the person or the agent that made it, and can be restored.
Your knowledge stays portable
Pages are Markdown with frontmatter. A vault exports as Markdown files at any time, and imports the same way.