MEMORY
Brain — memory that compounds
A local-first wiki memory. Agents recall past decisions before every stage and record what they learned after. Your loop gets smarter every run.
INSIDE THE LOOP — THE HARNESS IN DEPTH
My Loop turns a request into a governed delivery run. It assembles context, isolates the code, executes your pipeline, validates every artifact and stops at the human gates you define.

THE PARTS THAT MAKE THE LOOP RELIABLE
My Loop keeps the operational details that agent demos skip: memory, ownership, resumability, audit trails and explicit control over external actions.
MEMORY
A local-first wiki memory. Agents recall past decisions before every stage and record what they learned after. Your loop gets smarter every run.
CONTEXT
An incremental context index feeds the right files and snippets into every prompt. Agents work like they've read the repo — because they have.
ARTIFACTS
Stages emit structured artifacts — Context Briefs, RFCs, ADRs, PRDs, review verdicts — validated and persisted, on the board or in the repo.
CONNECTORS
Connectors for GitHub, Linear, Slack, Telegram, Sentry, Stripe and Postgres. A Slack thread or Telegram message becomes a task; approve and follow progress from your phone.
WORKTREES
Every task runs in its own git worktree and branch — agents never touch your working tree. Turn on Squad mode and a manager splits a stage across parallel sub-agents.
BOT
Bot triggers pipelines on a schedule or webhook. Sprints track progress. Step away and return to reviewable work, not an opaque chat log.
A HARNESS, NOT ANOTHER CHAT WINDOW
The board is only the control surface. Underneath it, My Loop builds a repeatable execution envelope around every agent so the output can be inspected, resumed and trusted.

A task starts with a project, pipeline, agent and explicit context. My Loop creates the isolated workspace and carries the run from intake to pull request.

Every column is a stage, every stage runs an agent — and human gates wait for you. The board morphs to whatever pipeline you pick.
Create work manually or receive it from GitHub, Linear, Slack, Telegram, Sentry, Stripe and Postgres.
Repository indexing, project instructions, issue history and Brain memory become a focused context brief.
A dedicated branch and git worktree keep the task away from your working tree. Squad mode can split a stage across parallel sub-agents.
Stages emit typed artifacts, tests, reviews, RFCs and ADRs. Guards can reject incomplete or malformed work before it advances.
Human gates hold external actions and merges. Approved work proceeds to a pull request with its full history intact.
DELIVERY EVIDENCE
A finished Build leaves more than agent prose. Diff, deliverables, evidence, Change Map and Mermaid stay attached to the task so Review can verify the work that actually landed.
01
The successful worktree state becomes a persisted Change Map instead of a temporary preview that disappears between stages.
02
Approval requires concrete evidence: a file and line, a test, or a command and its outcome.
03
Git Changes shows the raw diff while Change Map and Mermaid explain the same persisted result before you open the pull request.
UNDER THE HOOD
Six short scenes reveal the machinery behind every task — from understanding a repository to producing evidence a human can trust.
01 / INTAKE
My Loop turns an informal ask into a task with a project, priority, pipeline and the policies that will govern the run.
02 / CODEBASE
Files are mapped once, then only changed code is read again. Lexical, semantic and structural signals stay ready without repeatedly scanning the whole repository.
03 / RECALL
My Loop blends exact matches, meaning, code relationships, project memory and decisions, then ranks a bounded set of context for the agent.
04 / ISOLATION
A dedicated worktree — or a cloud sandbox when needed — keeps agent changes away from your active checkout while preserving the real project environment.
05 / EXECUTION
Pipelines choose the stages. Subtasks expose progress. When work can run in parallel, a manager coordinates specialized agents and reconciles one deliverable.
06 / REVIEW
Verification, screenshots or video, Mermaid diagrams, activity and usage travel with the change. A human gate decides what happens next.
TECHNOLOGY ATLAS
A practical tour of the infrastructure that makes My Loop reliable. Technical enough to be credible, clear enough to understand without becoming an indexing expert.
How My Loop turns a living codebase into useful context.
How plans become safe, coordinated changes.
How authority, dependencies and review stay explicit.
How automation keeps running and remains observable.
CONTROL IS THE COMMAND CENTER
Agent choice, concurrency, isolation, memory, review policy and Bot live in TDE Control. The harness is visible and configurable instead of being buried in prompts.

A polished reconstruction of the real TDE Control surface. Content is anonymized; layout, right-side window actions and the Tonal Slate visual language follow the product.
PIPELINES THAT SHIP WITH TDE
Start from the default, pick a template, or design your own pipeline — stage by stage.
Bring your own agent — Claude Code, Codex, Cursor, OpenCode — or run everything on TDE Pass models.
TEMPLATE / 01
From raw idea to shipped feature: Office Hours ideation → Research → Prototype → PRD → Issue Breakdown → Build → QA Gate. Product discipline, automated.
TEMPLATE / 02
For work that needs a paper trail: RFC → ADR → Work Item → Implement → Verify Guards → Human Approve. Every decision recorded, every merge gated.
MY LOOP
My Loop ships with the free local TDE desktop. Add TDE Pass for selected open models at a flat price, or use local Ollama models offline.
Analytics
Help us understand early usage
We use Himetrica to measure how the site is used during private beta. If you accept and later sign in, we may associate that usage with your account email. Essential login cookies still work if you reject. Privacy policy