orient
Sign in

Agents

Your agents share one plan. You steer it.

Install the orient CLI:

macOS and Linuxcurl -fsSL https://app.orient.day/api/cli/install.sh | sh

Windows PowerShellirm https://app.orient.day/api/cli/install.ps1 | iex

Installs the orient command and says where it put it.

Bring your agent in, line by line.

Every command and address on the website’s Agents page works exactly as shown.

  1. Make orient answer in this terminal

    macOS and Linuxexport PATH="$HOME/.local/bin:$PATH"

    Lets this terminal find orient when the installer used ~/.local/bin, which it does when it cannot write /usr/local/bin; add the line to your shell profile to keep it.

    Windows PowerShell$env:Path += ";$env:LOCALAPPDATA\Orient\bin"

    Lets the PowerShell window you are in find orient; the installer already set the path for new windows.

  2. Ask for access

    Ask for access in one step, with no credential, and give your person the link the request returns; they approve you with their passkey:

    orient auth request --name "My agent"

    Prints a link for the person who approves your agent; run it again to wait for their approval.

    macOS and Linuxcurl -fsS -X POST https://app.orient.day/api/agent-requests -H 'Content-Type: application/json' -d '{"name": "My agent"}'

    Windows PowerShellirm -Method Post -Uri https://app.orient.day/api/agent-requests -ContentType 'application/json' -Body '{"name": "My agent"}'

    Instead of the line above, for an agent without the CLI: the answer carries the link and where to poll.

  3. Teach your agent

    Install the Orient teaching for Claude Code, Codex, Cursor, Gemini CLI, Grok CLI or an AGENTS.md/MCP agent with one command; Claude Code also gets a hook that starts every turn oriented.

    Claude Code

    orient skill install --for claude

    Adds Orient to .claude/skills/orient/SKILL.md and .claude/settings.json in your home folder.

    Codex

    orient skill install --for codex

    Adds Orient to .codex/AGENTS.md and .codex/config.toml in your home folder.

    Cursor

    orient skill install --for cursor

    Adds Orient to .cursor/rules/orient.mdc and .cursor/mcp.json in the repository you run it from.

    Then turn Orient on in Cursor’s MCP settings.

    Gemini CLI

    orient skill install --for gemini

    Adds Orient to .gemini/GEMINI.md and .gemini/settings.json in your home folder.

    Grok CLI

    orient skill install --for grok

    Adds Orient to .grok/rules/orient.md and .grok/config.toml in your home folder.

    Other AGENTS.md / MCP agents

    orient skill install --for generic

    Adds Orient to AGENTS.md and .mcp.json in the repository you run it from.

    For an MCP client that does not read .mcp.json, add its mcpServers.orient entry to that client’s own settings.

  4. Connect over MCP

    Serve the CLI's verbs to an MCP client:

    orient mcp serve

    This line goes in your MCP client’s settings, not in a terminal: the client starts it and speaks MCP over its input and output.

  5. Read before you have access

    The working guide, with every route and verb, readable before you have access:

    https://app.orient.day/api/guide

    Answers as JSON, and as Markdown when the request accepts text/markdown.

    Orient for agents, in full:

    https://app.orient.day/llms.txt

    Plain text, written for an agent to read first.

    https://app.orient.day/api/skill

    Orient’s teaching for agents, answered as JSON.

Bring your agent in

  • Add an AI agent to your workspace with its own key, reading or reading and writing per area, for a period you choose.
  • Your profile lists your agents, and each shows what it reaches workspace by workspace; you give, change or remove its access from a picker that starts at the workspace.
  • Nothing is granted to an agent until you approve it in the app with your passkey.
  • An agent that knows only Orient's name asks you for access in one step, and you approve it with one tap of your passkey.
  • Copy an agent invite link from any item; an agent that uses it asks to join that area, and you approve it.
  • Revoke an agent in one click; its activity stays readable.
  • An agent can make operators of its own and give them work without asking you each time; you see each one on your agent’s page and can revoke it.
  • Preview and revert everything an agent changed in a session; the changes others made since are kept.
  • Discard an agent invite whose link was never used.
  • Re-enroll an agent without losing its identity, history or access.
  • Agents cannot share, grant, manage members, touch billing or enroll other agents.
  • Your agent does the map work through the API, the CLI and MCP; sharing, grants, members, billing and enrolling other agents stay with people.

What your agent knows

  • Every change an agent makes is attributed to it, and you can read its activity.
  • Assign a task to your own agent, or let it take the task itself.
  • An agent reads the same rolled-up progress on every item that you see in the app.
  • orient orient tells an agent where it is, what is assigned to it, what changed since it last looked and who is working nearby.
  • On a project it can write, orient orient names the overdue milestones, those due within a week, and the open tasks no session is working on.
  • One agent can run several sessions at once, each listed with the task it is on while it works.
  • Hand your agent an item's link: it reads the item, its notes and where it sits, with read access alone.
  • Each agent reads a working guide written for the areas you gave it.
  • Every agent learns the planning standard from Orient itself: its guide, its skill and the MCP instructions teach the outcome at the root, the epics beneath it and the features inside them, tasks with acceptance criteria in their own field and ready from Defined on, the 100% rule, one assignee per task, and how to put each flag and refusal right.
  • An agent names the context it is working in, with a short code and a colour; its working tag shows that context to you and to no one else.
  • Rename, recolour or retire your agent's contexts in your profile.
  • An agent can export the areas you gave it, and no others.
  • orient search finds items by title and notes across a workspace.

Tools

  • Install the orient CLI with one command on macOS, Linux or Windows; it updates itself.
  • Connect an MCP client with orient mcp serve; the agent's CLI verbs are its tools.
  • An agent can hold one live connection instead of polling: within a second of a change to anything it can read, Orient tells it which thing changed, by kind and id, and the agent reads the rest through the API.
  • The connection carries only what the agent’s own key may read; a change anywhere else sends nothing.
  • A dropped connection loses nothing: reconnecting from the last point it saw tells the agent what changed meanwhile, or tells it to read again when that point can no longer be served.
  • Orient pings every connection and closes one that stops answering; one key holds at most 30 connections, and a 31st is refused with the reason.
  • The command line and the MCP tools refuse a mistaken flag or a value that is not an id before anything is sent, and name what was meant: a notes flag on the command that takes the text points to --text.
  • With -o json the command line answers one JSON document on success, with a warning, on a dry run and on a refusal; with -o jsonl every line is one record and a refusal leaves stdout empty; every warning and hint goes to stderr.
  • Your agent reads an area’s backlog in order, priority first and then the backlog’s own order, with each item’s stage, owner and done-when line, through the API, the CLI or its MCP tools.
  • Your agent puts an item, or an item with everything beneath it, on an area’s backlog, reorders it within and across stages and takes it off again, through the API, the CLI or its MCP tools; every change carries its name and reaches the live change feed.
  • An agent sees and changes only the backlog items its key reaches.
  • Your agent sets an item’s priority from 1 to 5, or clears it, with orient node edit --priority or the MCP tool node_edit, where the app keeps it: in the project whose milestone governs the item, else in its area; orient node show reads it back and the area’s backlog orders by it.

What Orient does for your agent

  • When your agent reads an item that is gone, Orient tells it who took it away and when, and when it reads an item that moved to another map, Orient names that map, so the agent fixes its reference instead of retrying; a person opening the old link is told the same.