Runic
46 FEATURES · 6 AREAS

Everything Runic does, one page each.

Runic is a text editor with a whiteboard attached: you write a process as plain lines, a flowchart or sequence diagram builds itself beside you, and an AI agent edits the text with diffs you review. Describe a whole product and Runic plans and writes every flow, then packages it for Claude Code and Claude Design — or connect Claude to your flows over MCP. Each page below explains one capability in full — what it does, how to use it, and where it stops.

Prefer to watch? Every feature in motion, full size →

The editor

What the editor does for you because your process is text — and no canvas tool can copy.

LIVE

The diagram redraws on every keystroke

Every change to the text re-parses the document and rebuilds the flowchart in the same render. Nodes glide to their new positions, brand-new steps grow in at their spot, and hovering either pane lights up the twin in the other. A chip in the editor header tells you whether the text parses and whether it is saved.
Read more →
EXTRACT

Select steps, convert them to a process

Select two or more lines and a floating bar offers “Convert to process”. Name it, press Enter, and the block leaves the document as its own process while a .do line stays behind. Half a decision can never be extracted: the selection grows to keep every .if with its branches.
Read more →
REFERENCES

See every document that calls this one

Runic indexes every .go and .do in your workspace, in processes and in sequence diagrams alike. The sidebar shows a badge with the number of call sites in other documents, the References panel lists each call site with a line preview, and one click jumps to that line in that document.
Read more →
RENAME

Rename once, every reference follows

Click the document name, type a new one, press Enter. Runic rewrites every .go and .do line that targeted the old name in every document you can edit, processes and sequence diagrams alike, keeping indentation, the command and any arrow caption intact. A rename cannot quietly break a flow you forgot about.
Read more →
AUTOCOMPLETE

It knows your commands and your document names

A dot at the start of a line opens the Commands menu. After .go or .do the menu lists your other documents, processes and sequence diagrams; after .back it lists the steps of the current document; in a sequence diagram it completes actor names. Arrow keys move, Enter or Tab inserts, Esc dismisses, and typos stop being a category of problem. Signed in, a gray ghost suggestion adds the next line.
Read more →
COMMANDS

The command menu, in both editors

Type a dot at the start of a line and a 286-pixel “Commands” menu lists the ten process commands, or the seventeen sequence commands, each with a description. Picking .if writes both branches, picking .alt writes the .else, picking .end opens the next line. The menu is placed under the caret line, moves above it when there is no room below, and follows the line as you scroll.
Read more →
FOLDING

Collapse what you are not reading

A chevron appears in the gutter beside any line with indented lines beneath it. Click it and the block collapses to a “⋯ N lines” chip. The fold is a view: the document is untouched, the diagram is complete, and the first edit or click expands everything with the caret exactly where you aimed.
Read more →
QUICK FIX

A link to nowhere fixes itself

A .go or .do whose target names no document gets an amber band and an inline “+ Create as process” chip. Click it and Runic creates a process with exactly that name and opens it. The sidebar shows an amber dot on any document that still has a broken link, and the sequence pane tints its REF or GO TO box amber with a “+ Create” label for the same gap.
Read more →
REORDER

Move a block, the picture follows

Hover a line and a ⋮⋮ handle appears. Drag it and the line moves with every deeper-indented line under it, plus the closing .end of a decision; the block re-indents to fit where it lands. Select several lines and they move as one. Tab and Shift+Tab shift every selected line by one level.
Read more →
NOTES

Notes that do not cover the diagram

Write `.note` under any step to attach context, ownership, or an edge case. In the diagram the note becomes a yellow comment icon beside the card; several notes on one step collapse into a single icon with a count badge. Image exports show the full text. Sequence diagrams have the same icon, anchored to an actor or kept in a side column.
Read more →
LAYOUT

One text, laid out your way

The layout engine turns the parsed text into positions. Four toolbar toggles shape the result: direction, compact cards, branch alignment and tight packing. Arrow captions come from a colon on the line. Return arrows are routed through clear gutters so they never cross a box.
Read more →
SIDEBAR

One list for every document

The Documents sidebar holds both document kinds in one list. A + menu creates a process or a diagram, ✨ drafts one with AI, and rows nest under their only caller. Drag a row between two others to reorder it, or onto a row to make that document call it. A search field finds any document by name, nested or folded away included, and names that share a “Prefix · ” fold into collapsible groups, so a workspace of hundreds stays navigable. The sidebar resizes from 150 to 420 pixels, collapses to a 48-pixel rail, and which parents and groups you folded is remembered.
Read more →
DARK

Dark mode for the editor

Open the avatar menu and choose “Dark mode”. The app, the editor pane and the settings page switch to a navy palette; “Light mode” switches back. The preference lives in your browser with Preferences consent. Marketing pages, the docs and the login page stay light, and exports are untouched.
Read more →
KEYS

Every shortcut Runic binds

Runic keeps its shortcuts few and predictable. Both editors own undo and redo, Tab and Enter do structural work, Esc closes the nearest thing, and Enter sends a chat message. This page lists what each key does and where.
Read more →

The AI agent

An agent that edits documents the way a careful colleague would: proposals wait as diffs, every change is a checkpoint — and, from one brief, a whole linked system, with the cost of every call on the ledger.

AI CHAT

Tell the agent what to change. Review the diff.

The chat panel beside the editor proposes changes as green-and-red diffs of your document. Nothing is written until you click Apply, every applied change keeps a checkpoint, and each request runs on the tier that fits it: Claude Sonnet 5 with thinking off for an ordinary turn, Sonnet 5 with adaptive thinking on for whole-document work. When a request is ambiguous, the agent asks one precise question instead of guessing.
Read more →
CHECKPOINTS

Every applied change keeps a way back

When you apply a proposal, Runic keeps the full text of the document before the change on that chat entry. Click ↩ Restore, even much later, to return to it. Checkpoints are keyed to the document’s id, so they survive renames, reorders and archiving.
Read more →
LIVE PREVIEW

See the proposal as a diagram before you accept it

A pending proposal renders in the flow pane as the diagram it would produce, with added or moved steps highlighted in green and a banner offering Apply and Discard. The editor keeps showing your current text, so you judge the change the way you think about it — as the flowchart.
Read more →
SELECTION

Point the agent at exact lines

Select lines in the text or draw a box over nodes in the diagram, then press Copy to chat. A chip pins the range to your next message, and the reply is a diff confined to those lines by construction: the model returns only the replacement, and Runic splices it into the same range.
Read more →
AUDIT

Two audits in one click: structure, then meaning

One button under the chat runs a deterministic structural pass — one-branch decisions, empty or identical branches, unreachable steps, .back lines that point nowhere or at two places, a misplaced .end, a guard branch that falls through, options lost into a caption — and then hands the document to an AI analyst who reads it for missing failure paths, unclear roles and absent thresholds. The same structural check runs on its own after every change the AI writes, and the same codes are what the AI must fix in its own output.
Read more →
MEMORY

The conversation follows your account

The chat is saved with your account, not your browser. Long conversations fold older messages behind summaries that keep every concrete value, and when you refer back to something specific, the original messages are recalled verbatim. Say “apply what you suggested” and it knows what you mean; when a detail sits only in a summary, it asks.
Read more →
GENERATE

One brief, a whole system of linked documents

Describe a product or an operation in plain words and Runic plans it — actors, every process and integration, the entities and their statuses, what it assumed, what it must ask — then writes each planned document on its own call, in three waves, with the plan as shared context. Platform concerns are named but never drawn, missing values become notes instead of inventions, and the result lands in your sidebar as a nested tree with a hand-over message and a queue of open questions. One narrow flow or one sequence diagram is still a single document away.
Read more →
QUESTIONS

After the draft, the agent asks what it left open

A first draft is the happy path. After a whole system the queue comes straight from the plan’s own open questions, no extra call; after a Process flow linked set a reviewer reads the set back and asks. Either way each question opens the document it would change, your answer is edited into that document as a proposal, and the queue pauses until you Apply or Discard so two answers never race on one text.
Read more →
IMPORT

Import a whiteboard photo or a Mermaid file

Attach a photo of a whiteboard, a screenshot of a flowchart or a Mermaid file to a chat message. The agent reads every node, edge label and branch, translates them into Runic syntax, and either creates a new document or proposes a merge into the open one as a reviewable diff.
Read more →
COST

What AI costs, which tier does what, and the $5 line

Runic spends model tokens deliberately. Each AI task maps to a tier — Haiku 4.5 for autocomplete, summaries and planned documents; Sonnet 5 for chat, edits, the plan and the main flow; Sonnet 5 with thinking only for the plan, the build questions, the build spec and the deep review — and every call, successful or not, is written to a usage ledger with its tokens and a list-price estimate. On Runic’s key an account may spend $5 in a calendar month; the line in Settings → AI shows where you stand, and bringing your own key lifts the budget. MCP traffic costs nothing on Runic’s side.
Read more →

Build handoff

Shape the idea here, then hand your AI tools a complete spec instead of a guess.

PACKAGE

Everything Claude Code and Claude Design need, in one zip

The Build Package turns the documents you tick into the folder a coding agent starts from. Two AI calls happen: one writes the questions, one returns a single typed spec object. Everything else is code. Runic assigns every branch, message, story, entity and screen a deterministic id, renders every file from fixed templates, derives the integration contracts from your sequence diagrams, and runs ten mechanical checks on the rendered files. The done screen shows which checks pass, so a thin spec is fixed in the interview rather than discovered by Claude Code three milestones in.
Read more →
INTERVIEW

Up to eight questions before the spec

A flow says what happens. It rarely says which stack runs it, who may do it, what data each step reads or writes, how users sign in, or how fast it must be. The build interview pulls that out of your head in up to eight questions written for these documents, and folds the answers into a typed docs/spec.md. Unanswered questions are listed as open questions rather than guessed, and the ten quality checks on the done screen show what the answers still left thin.
Read more →
HANDOFF

A plan-mode kickoff, a memory file, and hooks that hold

The handoff is a small set of files that share one id scheme. KICKOFF.md is the prompt you paste into a plan-mode session; CLAUDE.md is what Claude Code keeps reading afterwards; docs/plan.md is the order and the Definition of Done; .claude/rules/runic-dsl.md teaches it to read the flows only when it opens them; .claude/settings.json runs the tests after every edit and refuses edits to the flows and sequences; docs/decisions.md is where it logs what it decided alone. Nothing in the set is written by a model.
Read more →
IDS

One id scheme, from .then branch to commit message

Every file in the Build Package cites the same ids, and so do the tests, commits and questions Claude Code writes from it. The ids are assigned by code once, deterministically, from the parsed documents in package order: flows and their decisions, branches, failures, loops, waits and inputs; sequences and their messages and alternative sections; then the spec-side stories, criteria, entities, screens, roles, targets and milestones, numbered by position. This page names each form, what it points at, and why the ten quality tells and the kickoff depend on them.
Read more →
QUALITY

Ten mechanical tells before you download

A package can look complete and still send Claude Code into a milestone with an untested branch, a field no entity has, or a target that says fast. The Package quality panel catches what can be caught mechanically: it parses the ticked documents, reads CLAUDE.md, KICKOFF.md, the spec, the plan and the design brief exactly as Claude Code and Claude Design will read them, and joins them on the id scheme. The same ten tells come back as lint from the MCP tool, which is what lets a client's model fix a spec in a loop.
Read more →
DESIGN BRIEF

A 13-section brief Claude Design reads on its own

Claude Design receives one file and never sees the repo, so everything the screens are designed against is in that file. Runic renders it from the spec object, the plan and the parsed flows: the journeys are walked from the flow text, the four states of every screen are read off .input, .wait, .fail and terminal steps when the spec gave no copy, the components are found by name across screens, and the touchpoints table says what each sequence message needs on a screen. The brief and the spec share the screen headings, so an artboard named SC-2 is the same screen in both.
Read more →
OVER MCP

The build package from Claude Code, over MCP

In the app, Runic's AI writes the spec and you read the tells. Over MCP the roles swap: your Claude Code, Claude Desktop or claude.ai session writes the spec with its own model, and Runic's tool does the deterministic part, the ids, the files and the lint. Because a whole package can exceed what an MCP client accepts in one result, the tool has a lint-only mode for the fix-and-retry loop and a paths mode for fetching one file at a time, and the build-spec prompt is written around exactly that loop.
Read more →

Share & output

Everything a diagram tool gives you — links, embeds, exports, history, teams — reading from one text.

SHARE

Read-only links that show the current diagram

Click Share on the diagram pane and Runic creates a public, read-only URL at /s/<token> for the open process or sequence diagram. Viewers see the current text rendered live, with zoom, layout direction and an Export menu, but nothing to edit. Revoke deletes the token and the link dies immediately.
Read more →
EMBED

Iframes that follow the text

Copy embed in the Share menu gives you an iframe pointing at /embed/<token>. The frame renders the same document as the share link, without the header, and polls for changes every 60 seconds, so a diagram on any page that renders iframes never goes stale.
Read more →
EXPORT

PNG, SVG, clipboard, text and Markdown

The Export section of the Share menu offers five outputs: Copy image, PNG image at 2× resolution, SVG image, the plain-text source, and a Markdown file with a reading legend so Claude or a teammate can interpret it. Image exports come in seven styles, including your own four-color palette.
Read more →
HISTORY

Snapshots you can name, pin and restore

Every signed-in document gets automatic version snapshots, at most one every ten minutes of editing, plus one whenever a save overwrites text someone else just wrote. The clock icon in the editor opens Version history, where each row shows how it differs from now, whether Claude took it over MCP or a concurrent save kept it, and a Restore button that asks first and can be undone; naming a snapshot pins it. Claude can browse, diff and restore the same list through MCP.
Read more →
TEAMS

Shared workspaces with owner, editor and viewer roles

Teams puts documents in a workspace instead of one person's account. The owner invites people as editors or viewers, by email or with an open link, and documents created inside the team appear in every member's sidebar. Permissions are checked on the server for every read and write — in the app, on share links and over MCP — so removing someone ends their access at once, and two people saving the same document at the same time never lose each other's text silently.
Read more →
MAP

Every document and every link, on one canvas

Click Map in the top bar and Runic lays out every process and sequence diagram as a force-directed graph. Violet arrows are .do calls, blue arrows are .go jumps, dashed red arrows land on ghost nodes for documents that are referenced but never written. Hover to isolate a neighborhood, drag to rearrange, click to open.
Read more →

The language

Ten single-word commands for flows, a handful of arrows for sequences. Indentation carries the structure.

PROCESS

Ten commands, one node per line

Every non-blank line in a process document becomes something in the flowchart: a card, a branch, an arrow or a note. Plain text is a step; ten dot-commands add decisions, branches, ends, jumps, sub-processes, inputs, waits, failures, return arrows and notes. Indentation decides how they connect, so there is nothing to draw.
Read more →
SEQUENCE

Sequence diagrams from arrows you type

A second document type for who-talks-to-whom. Each line is one message between two actors — sync, async or reply — with activation bars, typed actors, anchored notes and loop / alt / opt / par / break / critical / group fragments. Reference any other document with .do or .go, or paste a Mermaid sequenceDiagram and keep going.
Read more →

Platform & account

Sign-in, MCP for Claude, and the privacy choices that shaped how Runic is built.

MCP

Connect Claude to your flows over MCP

Runic ships a Streamable HTTP MCP server at /api/mcp: 19 tools, seven packaged prompts, every document as a resource and a live change stream. Sign in from claude.ai through OAuth, or create a personal token under Settings → Claude & MCP and add one line to Claude Code. Claude can list, read, search, lint, write, audit, render, diff, share and roll back your flows, and it learns the Runic language from the server’s own syntax guide. The server never calls a model, so MCP traffic spends your Claude plan, never Runic’s AI budget.
Read more →
CLAUDE.AI

Connect from claude.ai with OAuth, no token to paste

Add https://runicflow.ai/api/mcp as a custom connector in claude.ai, sign in to Runic, read what the connector may do and press Allow. Behind the click: RFC 9728 and 8414 discovery, dynamic client registration, an authorization code with PKCE, a one-hour access token and a 30-day refresh token that rotates. The connection shows in Settings → Claude & MCP as “OAuth · Claude” with its scope and expiry, and Revoke ends it. The same flow serves any client that speaks the MCP authorization spec.
Read more →
SCOPES

Read-only tokens: let Claude look without touching

Tick “Read-only” when creating a personal token and Claude can list, read, search, lint, audit, render, diff and export every document, run all seven prompts and follow the change stream, but write_document, create_document, rename_document, restore_version and share_document answer “This token is read-only (scope: read). Create a token with write access under Settings → Claude & MCP.” OAuth grants carry the same scopes, and the consent screen shows which one the client asked for.
Read more →
SIGN IN

Sign in with email, Google or GitHub

Runic accounts use email and password, Google or GitHub. Providers link to one account by verified email, Settings → Connected accounts adds or removes them without ever removing your last way in, and a password reset link works once and expires in two hours. Sessions are a 30-day httpOnly cookie.
Read more →
PRIVACY

Three cookies, no trackers, your data on request

Runic is built to hold as little as possible: a session cookie, a ten-minute OAuth state cookie and the cookie that records your consent choice, self-hosted fonts, a Content Security Policy that only allows connections to Runic’s own origin, and AI calls that leave the server only when you invoke a feature. Settings → Your data offers Download my data as one JSON file and Delete my account after a password check.
Read more →
AI SETTINGS

Runic’s Claude, or your own Anthropic, OpenAI or Google key

Every AI feature in Runic (the chat, Create with AI, inline edits, ghost autocomplete, the clarifying questions and the build package) runs one set of prompts through one server-side client. Which API key that client spends is a workspace setting: Runic’s own Claude models, a key each account brings, or Runic’s key unless an account stored its own. A stored key is checked against the provider before it is saved, kept encrypted, and only ever shown back as a masked hint. On Runic’s key each account has $5 a month; the section ends with a line showing this month’s calls and spend.
Read more →
SETTINGS

One page for everything that is not a document

Everything about your account that is not a document lives on /settings, reached from the avatar menu’s first item. A navy header with your email and “Back to the editor”, a sticky left nav, and one card per section — AI, Claude & MCP, Connected accounts, Plan (only when billing is on), Your data, Cookies — each addressable by hash. The page follows the editor’s dark mode. The avatar menu itself is short: Settings, Teams, Learn, Docs, Dark or Light mode, Feedback, Sign out.
Read more →

Read enough. Try it.

A free account takes a minute. Write three lines and watch the diagram appear.

Create a free account →Read the docs