Features
What you can do with TDT
A persistent local workspace with 20 core features. Keep your knowledge, rules, projects and reusable workflows together across Claude Code and Codex sessions. Add optional Stacks as your work grows.
Capture
Keep what you learn in a session
Your active agent summarizes useful work as notes awaiting review. Capture depends on the hooks and trust settings your agent supports.
Save useful notes from conversations
Capture useful facts, decisions and open questions from supported sessions. TDT uses your active agent to write concise notes, without a separate model subscription or copying full transcripts into your brain.
Review
Choose what your brain remembers
Read each proposed note before it becomes part of ordinary search. You can approve it, edit it or reject it.
Review captured knowledge
Review history records your decisions. Pending and rejected notes stay out of approved answers.
Keep your knowledge in local files
Your brain is readable, editable Markdown in a local folder, with links between notes and references to their sources. Your agent still uses its own service and subscription; local storage does not make it offline.
Keep the evidence behind the answer
Notes retain source references, session provenance and connections to other notes. Conflicting evidence can coexist, so a new conclusion does not silently erase what came before.
Search
Find what you already know
Ask your agent a question and give it the knowledge you have already reviewed.
Search your brain and get cited answers
TDT searches approved notes and follows their links to give your agent evidence for answers with references. It also guides the agent to acknowledge missing information and unresolved conflicts.
Workspace
Set up a workspace of your own
Start with core, then enable the agents and capabilities you need. Your shared knowledge stays in place as your setup changes.
Set up core
Create a workspace with one CLI command. Core needs no database, embedding model or continuously running service. You can set it up before enabling an agent or installing a stack.
Work with Claude Code and Codex
Use either agent, or enable both to share the same brain, skills and stacks. Setup detects installed agents, and you can enable or disable each workspace integration later.
Give your agents rules
Write a workspace Constitution that tells agents how you want them to work, when to ask permission and what boundaries to respect. Refine and approve your rules through chat or UI. These are agent instructions, not a security sandbox.
Keep integrations scoped to your workspace
Skills and hooks belong to the workspace where you enable them. TDT preserves unrelated agent settings and leaves linked projects free of automatic agent registrations. Disabling an integration keeps your shared knowledge.
Find help in your workspace
Read installed getting-started, command and troubleshooting guides locally. A catalog groups installed stack documentation by stack and version, with offline access to bundled guides.
Check your workspace health
Run tdt doctor to find layout, ownership, integration and recovery problems. TDT detects duplicate captures and keeps pending work recoverable, so you can resume after an interruption without adding the same knowledge twice.
Skills
Save a workflow you want to use again
Review what a proposed skill will do before it becomes part of your workspace. Saved skills can be reused with your enabled agents.
Turn useful workflows into reusable skills
When a request would make a repeatable workflow, your agent can offer to save it as a workspace skill. Review the exact behavior before saving or updating it. TDT checks existing skills and remembers declined proposals to avoid repeated offers.
Find skill ideas in past work
Ask TDT to find repeated workflows in accessible completed sessions and propose skills. It reports gaps in the history your agent can provide and saves nothing without approval. Discovery from real session history still needs end-to-end testing.
Projects
Link projects where they already live
Keep project knowledge in your brain while leaving source files in their existing folders.
Register existing project folders
Link folders without moving or copying their source. List linked projects and see when a path is unavailable.
Get oriented in an unfamiliar project
Have your agent read a limited selection of docs and manifests to explain the project purpose, structure, entry points and open questions. Registration and onboarding leave source unchanged, and proposed knowledge goes through review.
Stacks
Add capabilities when you need them
Stacks are optional. First-party and community stacks use the same format and installer, and TDT works with zero stacks.
Extend TDT with stacks
Install optional packages of skills, workflows, hooks, knowledge, templates and local capabilities. Stack workflows can use TDT’s shared UI.
Discover and inspect stacks
Search a configured registry and inspect stack contents and prerequisites. Install from a local folder or a registry listing backed by a GitHub source archive. TDT records where installed content came from.
Choose what executable content you trust
Inspect executable hooks and capability assets before approving the exact stack content you trust. TDT validates bundle paths, checks ownership conflicts and verifies declared digests. Trusted executable code is not sandboxed by TDT.
Review updates and remove stacks
Inspect and approve updates before applying them, or uninstall a stack when you no longer need it. TDT preserves brain notes and user work, refuses conflicting edits and supports recovery after interrupted operations.
Workspace UI
Answer your agent in a local browser
Say “use ui” to open the workspace UI. It is built into core and works without stacks.
Use questions, forms and interactive pages
Work through grouped questions, choices, forms and custom interactive pages, then continue with follow-up questions in the same session. Chat remains an option, and an active agent reads your submitted answers.
Where things stand
TDT v0.1.12 is our initial release. It is in active development, so expect changes as we fix issues and improve how it works.
If something breaks or does not work as expected, report it through GitHub issues.
Help improve TDT
Found an issue or something that could work better? Choose a GitHub issue template and tell us what happened and what you expected.
Give feedback on GitHub