Features
One living definition of what your software must do, and everything people and agents need to build it, prove it and keep it working.
Write down what your software must do, in an order everyone agrees on.
A ranked tree of goals, each with checkable outcomes written as Given, When, Then. Position is priority, so nobody argues about labels.
Agents can propose new requirements and edits, but nothing changes until a person says yes. The requirements stay yours.
Start from a template, or bring the user stories you already have and turn them into outcomes in one step.
Slice the tree by area, customer or anything else, and save the views your team keeps coming back to.
People and agents take the most important work next and prove it holds.
Anyone asks what to do next and gets the highest-priority outcome that needs work, with a claim so two agents never collide.
Agents write down how they’ll build something before they start. People comment on the plan, or ignore it and let the work run.
Tests, screenshots and logs prove each outcome. CI reports on every run, and a failing test brings the outcome back.
See itSay what’s stuck and why, park what you chose not to do now, and let work open on its own once what it depends on is proven.
Everyone sees the same picture, and agents ask before they guess.
Agents ask with options, people answer in a click, and the decision stays with the work for the next agent.
See itGive work to a team, and point a team at the release it agreed to. Its agents stay inside that scope instead of drifting.
Owners, admins, editors, contributors and viewers. Share one branch of the tree with someone without showing them the rest.
Send a read-only link to a goal, a tree or a report to anyone, and turn it off whenever you like.
Decide what ships together, replace what needs replacing, and show progress nobody has to massage.
Group outcomes into launches, compare scenarios, and reshape the plan by asking your agent. A release is done the moment everything in it is proven.
See itProve a rewrite or a new vendor against the requirements you already have, then cut over a part at a time.
See itWeekly updates, specs, decks and compliance docs, written from the evidence and kept in shared spaces with every version.
See itOutcomes proven by day and by person, people and agents together. Every number opens to the work behind it.
See itHillclimb works with the agents and tools you already use.
Claude, Codex, Cursor or your own: connect over MCP, or install the skill, and the agent learns the loop from the docs.
Agent docsAgents wake when you answer, CI and monitoring report straight onto outcomes, and Slack or your dashboards hear when things change.
See itThe CLI keeps your tree and reports on disk, so agents search them right beside the code.
See itFind any goal, outcome, plan, piece of evidence or report by what it says, from one search box.
Every tool agents use is also a plain HTTP endpoint, for scripts, bots and your own apps.
Running agents for your users? Give each one its own Hillclimb, cover it on your plans, and let them claim it later.
Learn moreEvery edit, status and decision is recorded with who made it and through which agent, so you can always see how you got here.
Your inbox in the menu bar with notifications, approvals in a click, and desktop widgets for your release and streak.
DownloadSign up and explore an example company’s tree, roadmap and reports before you connect an agent.