DoraBot Desktop Assistant
A Mac-based AI assistant that remembers ongoing work, runs scheduled tasks, and stays in touch through messaging.
- Source repo
- suitedaces/dorabot
- Stars
- ★ 248
- Last updated
- 4d ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 45/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code · OpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Individuals who move between a desktop and Telegram can keep preferences, decisions, and work context in shared assistant memory.
- Not a fit if
- Users who need a Windows or Linux desktop app
- Website
- dora.so ↗ set on GitHub
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-16latest release v0.3.9 (2026-11-15)
- Source review
- 45/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
DoraBot is a macOS desktop app that combines an IDE-style workspace with an AI assistant designed to run continuously. The workspace includes a file explorer, a Monaco editor with edit mode and autosave, a Git panel, and a PTY terminal with tabs and split panes. The assistant keeps journals and curated memory, tracks goals and research, and runs scheduled or recurring tasks. You can reach the same assistant from the desktop app, Telegram, or WhatsApp, with shared memory and context across sessions. It runs locally while making model calls to an AI provider; documented options include Claude, OpenAI Codex, and MiniMax.
During setup, the built-in onboard skill interviews you and creates SOUL.md, USER.md, and MEMORY.md in ~/.dorabot/workspace/. The assistant can keep daily journals, search past conversations, maintain research topics, and propose goals, write plans, and execute work; users can approve goals in the desktop app or Telegram. It can also run cron jobs, scheduled pulses, and recurring tasks, then message you with results. From Telegram or WhatsApp, users can send text, photos, voice, and documents. Browser automation offers more than 40 actions through an existing Chrome profile. The desktop workspace also lets users browse and edit files, inspect Git status, use a terminal, and view diffs.
- Individuals who move between a desktop and Telegram can keep preferences, decisions, and work context in shared assistant memory.
- Users with recurring research or work routines can schedule tasks and receive message updates.
- People managing longer projects can ask the assistant to propose goals and plans, then approve execution from the desktop app or Telegram.
- Users who exchange photos, voice messages, and documents with an assistant on WhatsApp or Telegram can continue the same context from mobile.
- Developers who want AI workflows alongside a coding workspace in one macOS app can use the editor, Git panel, and terminal.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npm install && npm run build && npm link
- Install notes cover
- macOS
How do you install or deploy this agent?
Download the macOS DMG, open it, drag the app to Applications, and follow onboarding. You need macOS and either a Claude Code or OpenAI Codex subscription; Claude, OpenAI, or MiniMax API keys are also supported. To build from source:
git clone https://github.com/suitedaces/dorabot.git && cd dorabot
npm install && npm run build && npm linkHow do you use this agent?
Start the gateway and desktop app in development mode, or use the documented production gateway, interactive terminal, and one-off question commands:
npm run dev
dorabot -g
dorabot -i
dorabot -m "message"Follow onboarding on first launch. The onboard skill creates personalization files that you can edit directly or let the assistant manage.
What are this agent's strengths and limitations?
- Keeps journals, curated memory, and searchable conversation history across sessions, with shared context between desktop, Telegram, and WhatsApp.
- Provides a goals workflow with planning, approval, and execution; approval is available in the desktop app or Telegram.
- Combines a Monaco editor, file explorer, Git panel, and PTY terminal with the assistant's workflows in one desktop app.
- Includes browser automation using an existing Chrome profile, plus Apple Calendar sync and iCal RRULE scheduling.
- The desktop app targets macOS; the README does not document Windows or Linux desktop support.
- Use requires a Claude Code or OpenAI Codex subscription, or an API key for Claude, OpenAI, or MiniMax; model costs are separate.
- The app runs locally, but model calls need a network connection to the chosen AI provider.
- The README does not give detailed setup steps or account permission requirements for Telegram, WhatsApp, or browser automation.
How does this agent compare with similar options?
The README compares DoraBot with Claude Code and Cursor, describing those primarily as coding tools while DoraBot combines memory, goals, scheduling, and messaging workflows with a desktop IDE-style workspace. It provides no feature or performance benchmarks.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| DoraBot Desktop Assistant This agent | 45 · Major gaps | Desktop appFree + model costs | ★ 248 | 4d ago | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
| Markus AI Workforce | 51 · Major gaps | Desktop appFree + model costs | ★ 219 | today | TypeScript | OpenAI API · Claude API |
| Abu Desktop Assistant | 70 · Some gaps | Desktop appFree + model costs | ★ 424 | 20d ago | TypeScript | OpenAI API · Claude API |
| LobsterAI | 60 · Some gaps | Desktop appFree + model costs | ★ 6.2k | 1d ago | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
AGENTS.md describes path allowlists, denied sensitive directories, gateway-token authentication, and tiered tool approvals. However, the sandbox is disabled by default, the default allowed paths are broad, and the README advertises use of a real Chrome profile, cross-channel messaging, and autonomous scheduled work without showing per-capability permission boundaries or rollback. Least privilege, confirmation, data-flow transparency, and external-effects controls are therefore only partly evidenced. Exact dependency pins and stated audit, vulnerability-scanning, and package-age checks support a dependency-security score of 2. The packet does not explain encryption, retention, or deletion of personal data, and it gives little clear author or contributor attribution, so sensitive-data handling and source attribution remain weak.
The documentation gives substantial architecture and data-flow detail, but README, AGENTS.md, and CLAUDE.md conflict on providers, SDKs, and defaults, including whether Codex or Claude is the default. This reduces self-consistency. package.json pins direct dependency versions and requires Node.js 22 or newer, supporting ordinary installation availability. The CLI prints runtime errors and channel status can expose lastError, but the packet does not show comprehensive failure categories, recovery guidance, or offline fallback behavior.
The README clearly targets users seeking a persistent personal agent on macOS with an IDE, memory, scheduling, and messaging, and describes both subscription and API-key onboarding. Configuration supports provider, skill, and channel choices, but the boundaries and default triggers for browser scripts, message sending, and system control are not fully explained. The macOS requirement is explicit and source-build commands are provided, supporting an ordinary environment-fit score.
The README, AGENTS.md, and skills directory provide clear feature sections, build entry points, configuration locations, examples, and an FAQ. The MIT license text is complete, package.json has a version, and the release workflow derives release notes from commits. Inconsistent provider and SDK naming across the documents weakens naming stability. Limitations are scattered, including Accessibility permission and external CLI credentials in some skills; maintenance responsibility is only lightly indicated through issue/PR links and the copyright name, with no clear support or update commitment.
The product combines an IDE, persistent memory, scheduled tasks, goals, browser actions, and messaging in one local workspace, which could add practical value for ongoing personal-agent workflows. CLI commands, skill examples, and interface descriptions help users understand the available outputs. The score is limited because the packet mostly provides feature claims and demo links, without measurable results for typical tasks or evidence about resource costs and always-on operation.
Some README claims can be traced to architecture descriptions, CLI entry points, dependencies, and release workflows, but most product and security claims are not demonstrated feature by feature in the supplied implementation excerpts. AGENTS.md and CLAUDE.md conflict on SDKs and defaults, and the packet offers little independent corroboration. The README separates product claims from setup information, but the materials do not consistently distinguish source-supported facts from promotional summaries, so traceability and fact/inference separation are only partial.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- AGENTS.md and CLAUDE.md materially disagree about the default model, SDK, and provider; do not rely on the docs alone to determine runtime configuration.
- The packet says the sandbox is disabled by default and default file access is broad. Review permissions and approval settings before enabling browser, messaging, or macOS automation.
- The README claims data stays local and only AI-provider network calls occur, but the supplied materials do not include enough implementation detail to verify those data-flow claims.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.3.9 0.3.9
Release notes excerpt
d5e951d v0.3.9 aa456d2 feat(desktop): send-now and move-to-background buttons, sweeps, tinted tool input d7cb026 feat(gateway): close idle cli processes, send queued messages now, move a call to the background 9a90ac4 fix(browser): restart only the browser holding the dorabot profile 87dfee4 fix(desktop): scrolling up lets go of the bottom on the first tick 2c5cedb fix(desktop): visible table lines, headers and code blocks in every theme fcba769 fix(desktop): readable secondary text in every theme 33a7176 fix(desktop): more contrast in the chatView on GitHub ↗ -
v0.3.8 0.3.8
Release notes excerpt
c8c6144 v0.3.8 9f15974 fix(desktop): keep the clock time in the turn footer 5d663e3 fix(desktop): apply the less-is-more rule across the app b971920 fix(desktop): back-to-latest is an arrow icon; add the less-is-more ui rule to CLAUDE.md 20e1bce fix(desktop): strip the remaining labels 42b968d fix(desktop): thought rows are just "Thought", the text is inside 8e42e51 fix: quieter tool rows, thin panes, queued messages, "working" only while working 371af84 fix(desktop): tool row actions are icons, labels are tooltipsView on GitHub ↗ -
v0.3.7 0.3.7
Release notes excerpt
1c8f4df v0.3.7 996bba3 feat: show the rest of what the sdk reportsView on GitHub ↗
FAQ
Will I pay model costs in addition to using DoraBot?
Does my data leave my Mac?
Can I approve goals without opening the desktop app?
Can the assistant work automatically in the background?
Is DoraBot Desktop Assistant open source?
Is DoraBot Desktop Assistant safe to use?
Is DoraBot Desktop Assistant still maintained?
Who makes DoraBot Desktop Assistant?
What is DoraBot Desktop Assistant's official website?
Thanks — we use this to improve the page.