PRI-OM-001 · Operating manualRev 0.37.0

Your agents do the work. You make the calls.

Priori runs Claude Code, Codex, pi and OpenCode in real terminals beside your files, collects every decision they need from you in one list, and gives them a browser you can both see. Local, on your Mac.

Get early access →Read the manual

Platform
macOS 12+
Status
Early access
Runs
Locally
Fig. 000.01 — Four parts, you, your agent, tasks and the vault, and six edges between them. Priori gives each edge a tool.
Supersedes rev 0.36.0i
PRI-OM-001 · 100.00 WorkspaceRev 0.37.0

100.00

One window per project

100.01

Files and notes on the left, the document in the middle, your agents’ terminals on the right. The Brief, Agent workflows and Browser open as tabs beside the document.

The Priori window: a file tree with Git status, a plan document in the middle with Content, Brief, Agent workflows and Browser tabs, and a terminal on the right where an agent reads a handoff, edits files and reports passing tests.
Fig. 100.01 — Main window. The project is a demo; hidden terminal sessions keep running.
100.02

Meetings. Detected, recorded and transcribed on your Mac, into a Markdown vault. The agent turns decisions into tasks linked to the transcript.

100.03

Terminals. After a relaunch, tabs reopen with their output and offer to resume the agent session.

100.04

Vault. Notes and meeting transcripts live as Markdown in a vault you link to the project; tasks live in the project itself.

PRI-OM-001 · 200.00 BriefRev 0.37.0

200.00

Everything waiting on you, in one list

When an agent needs a decision, it records the question on its task with the context, the options and the one it recommends. The Brief shows every ask across your agents, most blocking first, beside what’s running, ready and blocked. Press Enter to take the recommendation, or a letter for another option.

Press V for the Board: the same plan as lanes by dependency depth, with the next tasks to start marked.

The Brief: a list of decisions waiting on you, with one open to show what it unblocks, its context, three answer options and the recommended one.
Fig. 200.01 — An agent’s ask, with its context, the options and the recommended answer.
The Board: the same plan as lanes by dependency depth, with tasks that need you, are blocked, running or next to start.
Fig. 200.02 — The Board: the same plan, in lanes by dependency depth.
Since 0.31.02
PRI-OM-001 · 300.00 AgentsRev 0.37.0

300.00

Real terminals, the harnesses you already use

Each terminal tab is a login shell running Claude Code, Codex, pi or OpenCode, unwrapped. Agent setup connects them to a project in one step: lifecycle hooks or a plugin, MCP configuration and orchestration skills, with no prompt to paste. Agent workflows shows each main session with the subagents it started.

The main session hands bounded work to search, easy, hard and expert subagents, and checks what comes back.

A terminal tab running an agent session: it reads a handoff, edits files and reports passing tests.
Fig. 300.01 — A terminal tab, running the harness as it ships.
The agent setup dialog: choose Codex, Claude Code, pi or OpenCode, link a vault, and keep agent files out of Git.
Fig. 300.02 — Agent setup connects a harness to the project in one step.
Agent setup since 0.15.03
PRI-OM-001 · 400.00 BrowserRev 0.37.0

400.00

One browser, for you and your agents

Each project gets its own browser in the Browser tab, driven by you and your agents together. It fills its pane at your display’s sharpness. Device mode sets a specific size, a size an agent chose is marked “Set by agent”, and Chrome’s DevTools docks in the pane.

priori browser <file> opens a report an agent wrote, such as .agent-artifacts/report.html.

The browser toolbar in device mode: Phone size 390 by 844, DPR 3, marked “Set by agent”, with a Fill button.
Fig. 400.01 — A size an agent chose is marked “Set by agent”.
The project browser with Chrome’s DevTools docked at the bottom, showing a shipments page above the Elements panel.
Fig. 400.02 — The shared browser with DevTools docked in the pane.
Needs Chrome, Chromium, Edge or Brave4
PRI-OM-001 · 500.00 HandoffsRev 0.37.0

500.00

The next session starts where the last one stopped

Through MCP, an agent saves a handoff with its task: a summary, scope and constraints, the files it changed, its branch and commit, and next steps. The next session reads the last handoff before it touches the repository.

Session 14 Task 0142 Session 15 1 Save handoff summary, files, next steps Later 2 Read last handoff before the repository 3 Handoff task 0142 4 Continue the task
Fig. 500.01 — The handoff travels with the task, not with the terminal.
Since 0.2.05
PRI-OM-001 · 600.00 ReviewRev 0.37.0

600.00

When an agent edits the file you’re editing

Priori shows both versions side by side. You choose each change and save the merged result. Drafts survive a crash or quit.

The merge dialog for a file an agent changed while you were editing it: original against your edits, original against disk, and the merged result.
Fig. 600.01 — Both versions side by side; you choose each change.
Since 0.9.06
PRI-OM-001 · 700.00 AccessRev 0.37.0

700.00

Early access

macOS 12 or later · Apple Silicon and Intel. Expect changes and a few rough edges.

  1. 0.37.0The Brief's decision view no longer has letter keys, Enter-takes-recommendation or the decision Undo toast. Choose an option or write a note, then Save; Back changes an earlier answer.
  2. 0.36.0`ask_user` and its Browser-tab form are gone. To ask about something you need to see, such as mockups, an agent passes the HTML as `request_decision`'s `preview`, and Codex no longer needs the 11-minute tool timeout.
  3. 0.35.0Agents ask you in the Brief. Priori now tells every agent to send each question through `request_decision`, so it waits in the Brief and the Attention inbox instead of the terminal, chat or a browser form. If no task fits, the agent creates one first, then keeps working or ends its turn and reads your answer when it resumes. When you say a choice is yours, it asks instead of picking for you. Subagents report blockers instead of asking you.

Full changelog →