Throne — an agentic engineering app built around intent.
Throne helps people manage AI agents around a concrete intent: plan the work, gather context, launch the agent, and understand the result faster. Local, a single binary.
curl -fsSL https://throne.whitesnow.tech/install | bashThe productive AI work loop
The interview unpacks the intent. Work delivers the result. The knowledge base captures context and feeds it back into the next pass — into the new interview and the new work.
Throne closes this loop in one place: framing, agent launch, review of the result, and memory that feeds back into the next pass.
Throne in ten minutes
A short walkthrough: creating an Intent, launching the agent from the embedded terminal, and improving the result through patches.
Context under your control
Throne gives the human one interface over intents, agent launch, and memory. Context no longer scatters across chats — it lives around the Intent.
Intent workspace
Every task lives as an Intent: a wording, history, links, questions, and work results. Context no longer scatters across chats.
Links between tasks
Connect Intents to each other: continuation, dependency, refinement, bug, fix. The agent sees a map of work, not an isolated request.
Launch the agent from Throne
Hit "Run agent" — Throne spins up a terminal right in the browser, hands over context and operational skills, and drives the session through hooks.
Memory and preferences
Your rules, style, and decisions are stored locally and applied automatically when the agent starts framing or a work session.
Dream retrospective
Ask your agent to review past conversations and propose a patch to your instructions. You accept or reject the change — memory improves transparently.
UX for the human
Throne shows what context a task has, which instructions the agent will receive, and what changed after the work. Memory management stays visible.
Three steps to your first Intent
Throne runs locally: a single binary serves the UI, the API, and the database. No Docker, no cloud, no network auth.
Install with one command
macOS arm64/x64 (Apple Silicon/Intel) and Linux x64. The script downloads the latest release, installs the binary and the throne command, appends PATH, and starts the daemon. Re-running does a clean reinstall.
curl -fsSL https://throne.whitesnow.tech/install | bashOpen Throne
The daemon starts and opens the browser at http://localhost:5008. Manage it with the same command; update to the latest in one line.
throne # start daemon and browser throne status # port, pid, version throne update # update to latestRun the agent
Create an Intent and hit "Run agent" — Throne spins up an agent terminal right in the UI, hands over context and operational skills, and drives the session through hooks. Host features — repositories, PR comments, open in VS Code — light up on their own as soon as the matching CLI is on PATH.
Releases and development discussion
The public Telegram channel is the main place for release updates, development discussion, and feedback. Manual builds for macOS arm64/x64, Linux x64, and Windows x64 stay in GitHub Releases.
Where Throne is heading
The core loop: brief the agent, give it context, verify the result, and keep useful takeaways for the next run. Each direction below strengthens one of these steps.
Easier verification
Diffs, change history, static analysis, diagrams. The goal — see faster what the agent got right and where another pass is needed.
Connecting work systems
Bug reports, task managers, messengers, and internal knowledge bases. Throne pulls in incoming context and turns it into an Intent.
Intermediate task memory
Steps, partial solutions, open questions, dependencies, and state across multiple agent passes.
Long-lived project context
Architectural decisions, repo rules, team conventions, and recurring patterns. The agent gets personal plus project-specific context.
Workspaces for teams
Roles, shared instructions, project memory, intent history, and access control. Throne becomes a team-level loop for working with agents.
Throne public channel
Release updates, development discussion, and feedback live in the public Telegram channel. For direct feedback, you can also message the author personally.