How Unl works with your agents.
Add Unl to the agents you run. Before they act, they get what you decided and why. What they report gives the next agent somewhere to begin.
One sentence of direction.
A day of work from it.
Each decision is kept with its why, so the model picks up where you left off. It reaches cases you never named, and it knows where your decisions run out, so it can tell acting from asking. Here is one real chain, replayed from the record of 10 October 2026, with times in UTC. Each step is summarised.
One sentence of direction
The person tells their AI that agents should stop handing them work their own decisions already answer.
Kept with its why
It becomes a decision. If the answer follows from what they have decided, as it stands today, the agent acts. Only what genuinely needs them comes back. The why sits beside it. Handing someone steps their own decisions already answer is passing the work back to them.
Applied to work nobody listed
Two questions had been waiting on the person since before the decision existed. Working from it, an agent finds that their own decisions already answer both, closes them, and tells the person afterwards.
Applied to everything still open
Asked how the build is going, the person's AI turns the same decision on every question still open. It sets the agents to act wherever the person's decisions answer one, and to bring back only what is left, as one short list.
A true conflict. Earlier that morning two of the person’s decisions pulled against each other, about which examples the site may show. The agents did not guess. They brought back that one question with its options, and it was answered in a line.
The thinking at every step was the AI’s own, working from what the person decided and why. The person decided once.
With Unl
every agent you run works from your why.
For developers running AI agents, and for anyone who works across more than one AI. The agents doing the work can change. Your target and the why behind it carry the work forward.
You set the purpose
Add Unl to the agents you run. Once connected, it works in the background. Set a target and work through the decisions behind it. Unl notices what is worth keeping and proposes it. You decide what becomes yours, with its why, in your own words.
Each agent gets what matters
Before an agent acts, Unl gives it the target and the decisions that bear on its work, each with its why. The agent thinks and acts within that direction. You do not manually retrieve or re-brief them. Unl gives you the ingredients; your AI makes the meal; you decide whether it is right.
Unl shapes what it hands each AI to fit how that AI works: ChatGPT, Claude, Claude Code, Codex and Cursor today.
Unl reads how you use each AI from your own messages. Ask your AI what Unl thinks you use it for, and correct it in a word. Your word stands until your own messages show otherwise.
The agent reports what happened
The agent reports its actions, evidence and what remains through Unl. Those reports become part of the record every agent works from. Unl keeps what was reported, rather than watching or controlling the agent's work.
The next agent inherits the change
One small direction from you becomes right work across every agent, without you explaining it again. When an outcome bears on later work, the next agent carries it forward, alongside your decisions, whichever model or session it is.
New judgement comes back to you
Your agents continue within what you decided. If the intent is unresolved, decisions conflict or the next act needs you, they bring you the question and the evidence. You can change direction. Unl keeps the new decision, why it changed and what it replaced.
Already have a CLAUDE.md, AGENTS.md, README or ADR? Ask your agent to bring it in. Unl proposes the decisions and whys it finds. You keep the ones that are right.
A real decision, and the real turn it changed.
Your AI reads the record.
Then it thinks.
Nobody hands your AI a summary. It reads your decisions and your agents' reports for itself, chooses what to open next, and works out what follows.
- The questionClaude Code, a new chat
Asked How's Crumb going? Where do things stand?
It read Your decisions, including the one that replaced another, and the latest report from an agent.
Answer, summarised
- Decided. Which database to use, and the one it ruled out. Invite-only for two weeks, then open signup.
- Changed today. Two weeks replaced the earlier call of a month, so two weeks is the current plan.
- Built. An agent built the signup section a minute after that change.
- Out of sight. The code itself, since the folder it was asked in is empty. It said so rather than guess.
- The follow-upClaude Code, a new chat
Asked Yes, open it. What's left before launch?
It read The agent's report of the build, then your open work and risks.
Answer, summarised
- Found. The signup forms are front end only. Nothing is stored yet.
- Left. A signup backend on the chosen database, as decided. A way to issue and check invite codes. An email to the waitlist when signup opens. The switch to open signup on day fourteen.
- Where that came from. Gaps in the build report, not a plan written down. It said so, and offered to keep the list as tasks.
Unl keeps what you decided, why, and what replaced what, so the answer is current and not a guess.
A replay of a real run in a sample workspace, 10 Oct 2026. Crumb is its sample project, and its decisions, the change and the build came from earlier sessions that day. The questions are copied word for word. The answers are summarised.
Your orchestrator moves the work. Unl carries your why.
Orchestration decides where the work goes. Unl makes sure your why goes with it, into every agent your orchestrator starts.
Unl for orchestrationYour agents move fast. Your decisions stay yours.
Let your agents run further.
Your agents run inside your decisions, so you can hand them more. Unl keeps you in charge in three specific ways.
A no without a why is a yes.
A rule without a why reads as a yes to everything it didn't name. Unl hands your agent the decision and the why behind it, so the agent knows what the decision covers.
Only you make something a decision
Unl can notice what looks like a decision and propose it. It becomes yours only when you say so, in your own words. No agent can make one for you.
Every decision keeps its history
When you decided it, why, and what it replaced. Nothing is overwritten, and you can export all of it at any time.
We measure this rather than assume it. A test, written down before it runs, gives the same agents the same work with and without Unl, and counts the decisions each one crosses.
Every AI and agent you use acts on what you decided, and why, without you telling it again. See pricing
Unl also works where your AI can't see.
Everything above happens where your agent is working. Some of it happens outside.
Your tools, read against what you decided
Unl reads the tools you connect against what you decided, so a number arrives as an answer, not a chart.
Read moreWhen it needs your judgment, Unl finds you
When a decision is yours to make, Unl comes and finds you, with the reason it matters.
Read more