Docs · Workspace

Notes

An Obsidian-style knowledge graph, built into Koryphaios instead of living in a separate app — because the point isn't just to store notes, it's to let your agents read and write them.

Why it's native, not a side app

Obsidian is a wonderful tool, but it's a silo: your linked knowledge lives in one app, and your AI work lives in another. Koryphaios brings the good part of Obsidian — a network of linked notes you navigate by connection rather than by folder — inside the workspace, so the same graph your agents can query is the one you browse. A note about your API contract isn't just documentation you keep; it's context an agent can pull in while it works.

The trade Koryphaios makes is deliberate: your existing Markdown stays on disk, while Koryphaios indexes it as a graph the agents can operate on. That unlocks the agent layer below — and it's also where the honest limits are (see the bottom of this page).

What's there

Wikilinks & backlinks

Write [[Another Note]] to link notes; each note shows what links back to it. Aliases in YAML frontmatter resolve too; unresolved links stay visible as placeholder nodes until their note exists. Rename a note and the links to it are rewritten automatically.

Force-directed graph

A live graph view — nodes sized by connection count, colored by folder, with a “local graph” mode that shows just a note's immediate neighbors. Search dims unrelated nodes; pinned notes get a gold ring.

Folders, tags & pins

Organize by folder, tag notes, and pin the important ones. Pinned notes can be set to always live in the agent's context.

HTML notes

A note can be Markdown or HTML. HTML notes render in a locked-down sandboxed frame — handy for a dashboard, a styled table, or an embedded chart, with no script or network access.

Attachments

Drag files in; images embed inline. Attachments are stored as real files on disk under the project.

Project-doc mirroring

Existing .md/.html files in your project are mirrored into the graph automatically, so your README and design docs become linkable nodes — and edits write back to the real file.

Indexed retrieval

Search uses a full-text index, with folders, tags, aliases, backlinks, and graph metadata available before an agent loads a note's full body.

The part Obsidian doesn't have: agents

This is the reason notes are native. Your agents get a set of note tools — create, read, update, link, search, list, backlinks, graph summary, and a token-frugal render_note that returns bounded excerpts instead of whole notes. So an agent can jot a decision as a note, link it to the code it touched, and recall it three sessions later.

Per-tool permissions

Every note tool is set to Allow, Ask, or Hide. By default agents read freely but must ask before writing. Presets range from “allow all” to “block all,” and YOLO mode upgrades Ask to Allow.

Budgeted context injection

A compact catalog of your notes (title, folder, link count, tags) plus the bodies of pinned notes are injected into context under a token budget — the agent knows what knowledge exists and can pull the specifics it needs, without blowing the window.

Exposed to external CLIs over MCP

Koryphaios runs a small MCP server that exposes the note tools to any MCP-capable CLI (Claude Code, Codex, Cursor, Grok). The same note graph is readable and writable whether the agent is native or a CLI harness.

Honest limits (today)

Notes are built for agent-driven project knowledge. They handle thousands of indexed notes, but they intentionally do not try to reproduce every personal knowledge-management workflow. Specifically:

  • The editor remains raw Markdown in a textarea — it has a live preview, but not a WYSIWYG editor.
  • The graph uses a canvas renderer to keep large vaults responsive, though a very dense graph can still be visually hard to navigate.
  • There is no shipping native mobile app; the desktop workspace is the supported notes experience.
  • HTML notes are deliberately sandboxed: scripts, navigation, and network access are blocked.

The design bet is that a linked, agent-readable knowledge graph is more valuable inside your AI workspace than a more polished editor would be in a separate one. If you push it to Obsidian-scale vaults, you'll feel the engineering gaps above.