ThinkTree

ThinkTree is a web-first knowledge workspace for projects that need both hierarchy and semantic connections. It combines a readable tree, a draggable graph, Markdown blocks, node-purpose summaries, grounded evidence, structured change logs, and human-reviewed patch workflows into a long-term path toward a human-and-LLM maintained knowledge IDE.

Product Direction

ThinkTree is designed as a system that humans and models maintain together:

• Tree structure keeps the project readable from top to bottom.
• Each node now has a dedicated purpose/summary layer, so readers and agents can understand what a node and its subtree are for before opening detailed blocks.
• Graph relations capture semantic links without collapsing the outline.
• Evidence keeps nodes grounded in real files, paths, excerpts, and imported material.
• Patch review keeps structural change explicit, reviewable, and reversible at the workflow level.
• Structured change logs create an audit trail for later snapshots, restore, supervision, and branch-like workflows.

The current product surface is web-first. The data model and interaction model are intentionally being shaped so the project can later grow into a desktop or IDE shell.

Available Now

The current repository includes a working foundation across the first major product slice:

• Next.js 14 App Router + TypeScript + Tailwind CSS
• Prisma + SQLite persistence
• Project creation, rename, delete, import, and export
• Hybrid tree + graph workspace
• Canonical tree nodes stored through parentId
• Dedicated node summary/purpose field plus detailed Markdown blocks
• Semantic graph edges stored separately from the tree
• Draggable graph nodes with persisted coordinates
• Folder-style outline index synced with graph selection
• Markdown block editing with live preview
• Math rendering in preview with $...$, $$...$$, \(...\), and \[...\]
• Resizable projects / graph / editor panels
• Resizable source / preview split inside block editing
• Project-scoped evidence records
• Managed evidence file import into project-local storage
• Evidence projected into the graph as a distinct visual layer
• Evidence promotion into real nodes
• Tool-shaped retrieval APIs for tree, node search, evidence search, and scoped project context
• Reviewable patch proposals with apply, reject, and queued-proposal deletion
• Direct review mode that lets a draft become an immediately reviewable patch without a separate manual create step
• Structured change log entries for key user edits and patch decisions
• Right-side History tab for browsing the structured audit trail
• Project changelog API plus a live History tab for audit browsing
• Local .thinktree.json save and load for project transfer

Planned Next

The next product work should build on the current node-summary, evidence, and patch surfaces rather than bypass them:

• richer patch review UX with graph/editor-aligned visual previews
• savepoints and snapshots before and after major change application
• restore and later branch-style workflows built on snapshots and logs
• Codex / skills integration over explicit retrieval and patch APIs
• node-attached tool-using chat built on the same retrieval and patch contract
• broader provider support once the tool loop is stable
• portable project packaging and later zip-based export
• a future desktop or IDE shell once the core workflow is proven in web form

Repository Shape

Important areas in the current app:

• app/ for the Next.js app router and API routes
• components/workspace-shell.tsx for the main three-panel workspace
• components/workspace/node-summary-card.tsx for node-purpose editing
• components/workspace/graph-canvas.tsx for the graph rendering and interaction layer
• components/workspace/evidence-panel.tsx for evidence authoring and managed-file workflows
• components/workspace/patch-panel.tsx for direct review and queued patch review
• components/workspace/history-panel.tsx for project audit history
• prisma/schema.prisma for the persisted project, node, block, graph-edge, evidence, session, patch, and changelog models

Current API surfaces include:

• /api/projects
• /api/projects/:id
• /api/projects/:id/tree
• /api/projects/:id/changelog
• /api/projects/import
• /api/nodes
• /api/nodes/:id
• /api/blocks
• /api/blocks/:id
• /api/graph-edges
• /api/graph-edges/:id
• /api/evidence
• /api/evidence/search
• /api/tooling/tree
• /api/tooling/context
• /api/tooling/nodes/search
• /api/patch-proposals
• /api/patch-proposals/:id
• /api/patch-proposals/:id/apply
• /api/patch-proposals/:id/reject

Quick Start

1. Install dependencies:

npm install

If PowerShell blocks npm.ps1, use:

npm.cmd install

2. Create the environment file:

Copy-Item .env.example .env

3. Generate the Prisma client and create the SQLite schema:

npm.cmd run db:generate
npm.cmd run db:push

4. Start the development server:

npm.cmd run dev

Data Model Today

The current persisted foundation now spans both knowledge structure and reviewable change:

• Project
• Node
• Block
• GraphEdge
• Evidence
• NodeEvidenceLink
• Session
• Message
• SessionArtifact
• PatchProposal
• PatchOperation
• ChangeLogEntry

The tree remains canonical through Node.parentId. Semantic relations are modeled separately through GraphEdge. Grounding material is modeled through project-scoped Evidence plus node links. Node intent is modeled separately through Node.summary. Reviewable structured updates are modeled through PatchProposal and PatchOperation. Auditability is now modeled through ChangeLogEntry.

Documentation

• D:\new_windows\PhD\winter2026\AIproduct\ThinkTree\product-spec.md describes the intended product model and long-term workflows.
• D:\new_windows\PhD\winter2026\AIproduct\ThinkTree\implementation-plan.md turns that direction into a dependency-ordered delivery roadmap.
• D:\new_windows\PhD\winter2026\AIproduct\ThinkTree\docs\phase2a-api-reference.md documents the current evidence and retrieval surface.
• D:\new_windows\PhD\winter2026\AIproduct\ThinkTree\docs\patch-api-reference.md documents the current patch proposal and review surface.
• D:\new_windows\PhD\winter2026\AIproduct\ThinkTree\docs\change-log-api-reference.md documents the current audit trail surface.
• D:\new_windows\PhD\winter2026\AIproduct\ThinkTree\docs\thinktree-agent-skills.md explains how external agents should read ThinkTree and propose changes.
• D:\new_windows\PhD\winter2026\AIproduct\ThinkTree\docs\project-storage.md explains where data, evidence, and imports currently live.

Current Notes

• Managed evidence files live inside the project-specific ThinkTree storage area so later tools and exports can resolve them reliably.
• The graph currently shows both structural nodes and evidence nodes, but the system is still early relative to the eventual heterogeneous knowledge graph vision.
• The current patch review loop supports full apply or reject; partial apply and richer diff editing are the next layer.
• The new History tab sits on top of the changelog foundation; snapshots and branch-like workflows are the next layer, not a replacement for the audit trail.
• The documentation is intentionally explicit about what already exists versus what is still planned.