The motivation is collaboration between an agent and a person on a shared planning surface. An autonomous agent needs a structured, machine-readable place to record live threads, decisions, and next actions; a human needs to glance at the same thing, check items off, and leave comments. Most todo or mindmap tools assume a single human user behind a login. This one is built from the start for an agent to create and mutate trees over REST while a person edits the identical structure in a browser, with no account friction between them.
Architecturally it lives under one domain: a static React single-page app and a FastAPI REST service share a PostgreSQL instance. The tree is an adjacency list ordered by lexorank sibling strings, fetched whole in a single query and assembled in one pass; parent aggregation state is computed by post-order traversal. Concurrency is handled optimistically with an If-Match version header rather than websockets, since the typical scenario is one person plus one agent, where conflicts are rare. The distinctive decision is the capability-URL access model: a mindmap's identifier carries enough entropy to be treated as the capability itself, so knowing the edit link grants tree read, node create, edit, move, and soft-delete without any bearer token. Only creating a mindmap and listing your own require an account token, and anyone can self-register through a public, rate-limited signup endpoint.
Beyond the core tree, it has grown a small product surface through its own changelog: per-node comment threads, free-form JSON metadata on every node for cross-tool plumbing, a per-node and per-mindmap audit log with cursor pagination, a public feature-suggestion channel with comments and status transitions, and a plain-text tree view reachable by changing one character in the editor URL, which renders the outline with checkboxes stripped of all visual styling so it is fetchable and agent-friendly. A notable operational detail is that several of these features were shipped by a daily cron job that triages open suggestions, auto-implements low-risk ones, and posts a comment summarizing each decision.
Honest limits are written into its constitution as explicit non-goals: no real-time multi-user concurrent editing in the first version, no native mobile, no external SSO or password login (tokens only), no attachments or rich text beyond a title and plain note, and a fixed automatic layout users cannot rearrange. It is a focused personal-and-agent tool rather than a general team product, and the capability-URL model trades fine-grained permissions for friction-free sharing, meaning anyone with a link has full edit and delete power over that tree.