Try Intent-System→
Product

Intent-System: an operational implementation of IDD

Intent-System is our implementation of Tree-Structured Operational IDD. It puts intent — not spec — at the center, then turns each ready slice into reviewed work. Released intent-cli v0.22.0 emits the commands and records or verifies the state for a herdr-only team; your agent or operator runs the external tools after approval.

Try Intent-System→See how it runs todayEvidence showcaseHerdr-only orchestrationRoadmap
In one diagram

The operational loop

Intent treecurated by humansIssueissue-ready sliceImplementerhuman / agent / AIReviewagainst intentCloseoutsync state, plan next slice▸ slice▸ assign▸ open PR▸ acceptnext slice — back to treeHUMANSAGENTSCLOSE

The loop runs on GitHub issues. A slice of the intent tree becomes an issue that already carries the architecture, contracts, and UI patterns it needs — so writing a clear issue, the highest-leverage skill in modern AI-assisted coding, becomes feasible. Implementation runs on context that is already there. Review compares the PR to the intent slice, not to the reviewer's taste. Closeout writes any new decisions back into the tree as canonical intent.

The tree itself doesn't need to be perfect on day one. A continuous Clarification step lets AI surface "what about this?" questions in bulk with recommended answers; humans accept the recommendation where they have no strong opinion and override it where they do. Intent grows incrementally, not in an upfront marathon.

Our primary product story is the released v0.22.0 four-thread orchestration model. Herdr-only is the preferred non-preview path for collocated teams; agmsg plus herdr remains supported for distributed or existing-agmsg teams, and the three-thread timer loop remains a supported alternative. See it → Herdr-only orchestration →

Current practice

Models each thread runs today

These are the models this project's threads run today:

Design

Opus 5 / Sonnet 5 / Fable 5 / GPT-5.6 Sol

Implementation

GPT-5.6 Luna Max / Sonnet 5

Review

GPT-5.6 Terra Max / Opus 5

This is a current-practice snapshot, not a requirement. Intent-System is model-agnostic: the four-thread model defines responsibilities and decision boundaries, not a mandatory model list.

The chapters

Read Intent-System in detail

Detail pages, in reading order. Each opens its own page; the same step indicator at the top of every one lets you move between them.

01Current Flow
How the implementation cycle actually runs today, region by region.
Open →02Evidence Showcase
One traceable prompt-to-merge case, plus the honest v0.22.0 authoring-only minimum start.
Open →034-Thread Orchestration
The primary four-thread model with herdr-only preferred for collocated teams in released v0.22.0.
Open →04Timer-Loop Alternative
The simpler three-thread timer-loop alternative.
Open →05Intent Tree
Purpose / User Context / Means, as a navigable folder of markdown.
Open →06Unified Intents
Product-experience intent and technical intent in one tree.
Open →07Human + AI Assignment
Humans decide, agents implement, you review against intent.
Open →08Review & Closeout Loop
Review is intent diffing; closeout sets up the next slice.
Open →09Roadmap
Where Intent-System is today, and how we decide what's next.
Open →

Built on our proposals

Intent-System embodies both proposals — but they don't depend on it. Intent Compass — the Intent Tree supplies the direction for each slice. Git-AI-Flow — the proposal describes an ai-develop lane; intent-cli itself uses the configured target branch or named lane.

What Intent-System is not

  • Not a code generator. Code is downstream; we don't ship a model.
  • Not a replacement for your IDE.
  • Not a project management tool, though it can integrate with one.
Try Intent-System→