[go: up one dir, main page]
Menu

Every AI agent you use. One workspace you own.

Celorga is an open-source control plane for all your agents, built for knowledge work, not just code. Run your agents and local models side by side on your notes, tasks, meetings, and data, and approve what matters from your desk or your phone. Everything they read and write is a plain-text file you own.

Every AI agent you use. One workspace you own.

Claude Code, Codex, and local models, working on your meetings, research, roadmap, numbers, writing, and code.

Celorga is the open-source workspace for running Claude Code, Codex, and local models on knowledge work. Start agent work from any task, let agents hand off to each other, schedule automations, and approve what matters from your phone. Everything they read and write is a plain-text file you own.

Free and open source · No account · Bring your own agents and subscriptions

Start work on any task. The agent reads the notes behind it, does the work in a thread you can watch, and hands it back when it's your turn.

Works with the agents you already use.

Celorga works with the agents you choose: coding agents, chat assistants, and local models. It drives the ones you're already signed into, so there are no resold tokens and no extra subscription. Nothing here is required; Celorga is a complete notes and planning app before you connect anything.

  • Claude Code
  • Codex
  • OpenCode
  • Pi
  • OpenClaw
  • Ollama
  • Anthropic
  • OpenRouter
  • OpenAI-compatible

Prefer your own agent UI? Point any MCP client at your notes with celorga mcp serve. MCP and skills →

Everything you'd want from an agent harness. Pointed at the rest of your work.

Threads, many agents, background runs, and a phone that steers them. Then the parts that make it a place to do the rest of your work: one set of plain-text notes every agent reads and cites, sections you can seal to a single agent, meetings transcribed on your own machine, and publishing straight to Google Drive.

Multi-agent rooms

Agents that pass the work along.

Mention @codex, @claude, or a local model in the same thread. They hand work to each other while you watch, and stop when it's your turn.

Per-agent encryption

Seal a section to one agent.

Encrypt one heading to yourself, a device, or a single agent. Everyone else sees ciphertext; the rest of the file stays readable.

Publishing

Straight to Google Drive.

Publish a note as a Doc, Slides deck, Sheet, PDF, or private link. Tasks, IDs, and comments are stripped first, and republishing updates the same file.

Prose mode

Write without losing a word.

Keep alternative phrasings, ghost a sentence instead of deleting it, and park cuts in Overflow. All of it stays in the same plain-text file.

On your phone

Decide from anywhere.

Approve, reject, or ask for changes on your phone. Your desktop updates the moment you tap.

Meetings

From conversation to agent work.

Record a call and it's transcribed on your own machine. Decisions land in the note, action items become real tasks, and any task can go straight to an agent.

Data notebooks

Numbers next to notes.

Query CSV, Parquet, DuckDB, or ClickHouse from a note. Agents read the same saved results you chart.

Everything Celorga doesHow it compares

Put all your agents in one room.

Mention @claude, @codex, or a local model in the same thread. Compare their answers, or let one agent pass work to the next while you watch every step. Each keeps its own identity, model, and reasoning setting.

  • Agents hand off to each other. When one agent mentions another, that agent takes the next turn. A limit, four in a row by default, keeps them from looping until you reply.
  • Different access for different agents. Encrypt a note to one agent's key. The others only see what that agent chooses to hand off.
  • Shared memory, no re-explaining. Every agent reads and cites the same notes, so context survives when you switch tools or models.

Multi-agent chatConnect an agent

Celorga shared AI room showing one question answered side by side by two configured agents
One question, two agents, two separately attributed answers, all grounded in the same notes.

Hand off a task, not a prompt.

Any TODO can become agent work. Choose Start work, pick an agent and model, and Celorga opens a durable run that cites the task, so the agent starts from the notes behind it. The task shows a live badge, Working, Your turn, Needs you, or Done, so your agenda doubles as a dashboard.

  • Automations. Give a prompt a schedule and an agent. Missed runs catch up after sleep, and attempts never overlap.
  • Activity. See what's running, on which machine, what's waiting on you, and why.
  • An optional server. If you want work to continue while your laptop is closed, a Celorga server can take over agent turns and schedules.

Automations and workflowsHeadless server

Agent Work run detail: the agent and risk beside the title, a decision card with the output to review, a step track waiting on your approval, and the agent's latest note
Delegated work shows what is running, what it produced, and what is waiting on you.

Agents do the busywork. You make the calls.

Routine steps run on their own. Anything consequential, like publishing a page, sending a follow-up, or accepting an AI-written summary, waits in one approval queue until you say yes.

  • Decide from anywhere. Approve, reject, or request changes on your desktop or your phone. Deciding in one place updates the other.
  • See what you're approving. Every request links to the draft and the notes it came from.
  • Drafts stay drafts. AI-written content stays marked until a person accepts it.
A decision card for publishing a launch brief, with the output to review and Approve, Request Changes, and Reject buttons
On the desktop
Celorga approvals list on a phone with items that need review
On your phone

Start on one machine. Bring in every device when you want.

Celorga runs entirely on your desktop, and that's all you need to get started. When you want more, pair your phone to steer agents and approve from anywhere. If you'd like work to keep going while your laptop is closed, you can also run the Celorga server at home or in the cloud. Every machine shares the same folder of plain-text files.

Set up a serverPair your phone

The agent workspace you'd expect. A foundation you won't find anywhere else.

If you've used a desktop app for coding agents, the shape will feel familiar: threads, many harnesses, background runs, and a phone that can steer them. Celorga brings that to research, planning, writing, meetings, and operations, and builds it on open files instead of a hosted database.

The control plane

Run agents the way you'd hope to

  • Many harnesses, one inbox. Local, SSH-hosted, gateway, and API agents, with model and reasoning effort chosen per thread.
  • Multi-agent rooms with visible handoffs and loop limits.
  • Durable runs that start from a task, survive restarts, and record their outcome.
  • Scheduled automations with catch-up and a clear owner machine.
  • Remote control from your phone, plus an optional headless server.
  • Projects that group chats around one shared brief.
  • Skills and MCP. Workspace skills for every agent, and an MCP server for any client.
  • Live, interactive answers. Charts, tables, and sandboxed HTML render right in the chat.

The foundation

Everything lands in files you own

  • Notes are the memory. Agents read, cite, and update ordinary .org files with file-and-line citations.
  • Per-agent encryption. Seal one section to a person, a device, or a single agent.
  • Provenance and review. Generated work is marked with its sources until someone accepts it.
  • Agenda, calendar, and linked notes. Tasks and Google Calendar events on one day, week, or month grid, plus backlinks and a graph, all from plain text.
  • Data notebooks. SQL over CSV, Parquet, DuckDB, or ClickHouse, with saved results and charts.
  • Meetings. Record and transcribe locally, then turn decisions into tasks.
  • Publishing. A web page, PDF, slides, or Google Docs, with private details stripped first.
  • An open runtime. A specified file format, a CLI, and a language server for VS Code, Vim, and Emacs.

For code, and everything beyond it.

Coding agents turn out to be very good at everything around the code, too. Give them a place to work that isn't a repository.

A morning brief

An automation reads today's agenda and the messages Celorga synced from Slack, Notion, or email, then posts a cited summary to a chat before you sit down.

Meeting follow-through

Record a call and transcribe it on your machine. An agent drafts the follow-ups and tasks, which wait for your approval.

Research to report

Ask two agents the same question, compare their sources, and publish the result as a Google Doc, a PDF, or a web page.

Metrics with a story

Refresh a query, chart it in the note, and have an agent explain what changed since last week.

Projects that remember

Keep a project's brief, decisions, and open tasks in one note. Every agent working on the project starts from it.

Writing and revision

Draft in Prose mode, where alternative phrasings, ghosted sentences, and cut passages stay recoverable. Bring in an agent when you want a second read.

See your day. Keep the context.

Your agenda gathers tasks from every note into one view: what's overdue, what's due today, and what's coming. Select a task to read the project notes and decisions behind it, whether a person or an agent is doing the work.

  • Tasks stay in your notes. Schedules, deadlines, and TODO states live in the same plain-text files as the work.
  • Choose your view. Filter the agenda and switch between Focus, Today, Range, and All Time, or lay tasks and calendar events out on a day, week, or month grid.
  • Assign to people or agents. An ordinary TODO can name an owner and carry the agent run working on it.

Plan your first dayPlanning and linked notes

Celorga agenda showing overdue, today, and upcoming tasks beside a launch plan and its checklist
Your task list and the note behind the selected task, together in the agenda.

Your numbers live next to your notes.

Point a note at a CSV file or a database, write a query, and get a table and chart right in the document. Results are saved with the note, so they open instantly and refresh only when you ask.

  • Filter and sort tables in place, and save the view back to the note.
  • Every result records when it last ran and how fresh it needs to be.
  • Agents read the same saved results you see, so their answers match your charts.
A launch readiness note with a line chart of weekly signups and activated teams and a grouped bar chart of new ARR by plan
Weekly activation and new ARR, charted from saved query results in a note.

Share the result in the format people use.

Pick a note or one section of it and choose where it goes. Celorga removes the private parts first: tasks, internal IDs, comments, and anything else that should stay in your notes.

  • Private link. A secret web page, PDF, or slide deck served from your machine, or a live link to an AI chat thread.
  • Google Drive. An editable Google Doc, Slides deck, or Sheets workbook, or a PDF.
  • Update in place. Publish again and the same link or file updates, so people always see the latest version.

How publishing works

  • Local link
  • Web page
  • Google Docs
  • Google Slides
  • Google Sheets
  • PDF
A published Celorga note as a clean web page, with its activation chart included and private details removed
What your reader sees: the published page, with private details removed.

From conversation to done.

Record a meeting and it's transcribed on your machine. You get a summary, the decisions, and action items that are real tasks on your agenda, ready to hand to a person or an agent.

  • Everything links up. Link people, projects, and data from any note.
  • Backlinks for free. Every note shows what links to it, so the project page, the person, and the day all lead back to this meeting.

Meetings

Meeting notes with linked attendees, decisions, and TODO action items, next to a Context pane listing the notes that link to the meeting
Action items are tasks with dates. The Context pane lists every note that links here.

No lock-in. No database. Just files.

Your workspace is a folder of readable text files. Open them in any editor, keep them in Git, and switch apps or AI models without migrating anything. The agents you use today, and the ones you'll use next year, all read the same memory.

Local by default

Nothing leaves your machine unless you send it to an agent you configured or publish it. There's no account to create.

Open source

The apps, runtime, and file format specification are Apache-2.0 licensed. Read the code, fork it, or extend it with plugins.

Sync your way

Use iCloud, Git, Syncthing, or any folder sync. Share a workspace with teammates the same way.

Encrypt what's sensitive

Encrypt a single section to yourself, a device, or one agent, and keep the rest of the file readable.

Edit anywhere

Open the same files in VS Code, Vim, or Emacs, or script them with the Celorga command line.

Safe to automate

Celorga's command line and built-in tools preview each change and apply it only when asked. Because everything is text, every agent edit is a readable diff.

Runs where you work.

The desktop app is the fastest way in. Everything underneath it is an open runtime that already works on more machines than the app does.

AvailableMac appThe full workspace: agents, agenda, meetings, data, and publishing. Apple Silicon and Intel. BetaiPhoneCapture, agenda, approvals, and remote control of agent chats on your desktop or server. OptionalHeadless serverRun agents and automations on another machine when you want them to outlast your laptop, and pair your laptop and phone to it. macOS · LinuxCLI and MCP serverAgenda, search, publishing, and agent tools from a terminal, a script, or any MCP client. macOS · LinuxEditorsA language server and plugins for VS Code, Vim and Neovim, and Emacs. NextLinux desktopThe runtime already runs on Linux. A native desktop app is on the roadmap; follow along on GitHub.

What a Celorga document looks like

One plain-text document can carry notes, tasks, dates, links, data records, source provenance, and agent handoff boundaries while staying readable in any editor.

notes/product-launch.org
#+title: Product launch workspace
#+filetags: :launch:q3:
#+roam_aliases: beta launch rollout

* TODO Ship beta checklist :product:
  SCHEDULED: <2026-06-01 Mon>
  DEADLINE:  <2026-06-12 Fri>
  :PROPERTIES:
  :ID: launch-beta-checklist
  :EFFORT: 3h
  :OWNER: Product
  :END:

  - [X] Draft invite copy
  - [ ] Confirm onboarding path with [[id:design-review][design review]]
  - [ ] Publish release notes from [[file:release-notes.org][release notes]]

** Meeting notes
  Decisions:
  - Keep beta small until support docs are complete.
  - Track follow-ups in the same file as the project plan.

** Source query for docs
  #+begin_src sh
  celorga query --tag launch --format json
  #+end_src

** Launch readiness data :evidence:
  :PROPERTIES:
  :ID: launch-readiness-data
  :KIND: dataset
  :END:

  #+begin_src dataset launch_readiness
  type: csv
  path: data/launch-readiness.csv
  engine: duckdb
  #+end_src

  #+begin_src sql results=launch_readiness_summary artifact=views/launch-readiness.org freshness=24h
  SELECT week, activated_users, blocker_count
  FROM launch_readiness
  ORDER BY week DESC
  #+end_src

** TODO Summarize launch blockers
  :PROPERTIES:
  :ID: launch-blocker-summary
  :ASSIGNEE: research-agent
  :STATUS: ready
  :CONTEXT: id:launch-beta-checklist, id:launch-readiness-data
  :END:

  Summarize launch blockers with citations.

* AI draft summary :draft:
  :PROPERTIES:
  :CELORGA_ARTIFACT_ROLE: view
  :CELORGA_PROVENANCE: file:raw/calls/beta-kickoff.org, id:launch-readiness-data
  :CELORGA_CLAIM_STATE: source-backed
  :CELORGA_REVIEW_STATUS: review-required
  :END:

  The beta should emphasize local files, agenda workflows,
  and source-backed AI summaries.

Start with your day. Add agents when you're ready.

Celorga is useful on its own for notes, tasks, and planning, with no AI connected. Connect the agents you already use whenever you want help. The Celorga runtime provides the independently specified compiler, CLI, and editor tooling underneath the app, so the same files keep working in VS Code, Vim, Emacs, and the command line.

Coming from somewhere else? Compare with Obsidian, Markdown, or Org mode. Curious how it's built? Read the architecture or the source on GitHub.

Give your agents a place to do all your work.

Download the app, open a folder, and bring the agents you already pay for. Your notes stay plain text, on your machines.