SI 精選 AI Agents BoardIssue #2 · 2026-11-03
Method: candidates gathered from GitHub, Hacker News, Product Hunt, Hugging Face, Reddit and tech news (TechCrunch, The Verge, Ars Technica); only AI agents that complete tasks on their own are kept (open-source and commercial), each checked by hand. Combined score = each source's signal, log-scaled and divided by this week's best, times its weight (GitHub 25, Hacker News 20, Product Hunt 15, Hugging Face 10, Reddit 15, News 15), plus 5 points for each extra source. Signals read 2026-11-03; they show attention, not quality. Badges show where each agent appeared; click one to see the source.
Who it's for: Developers and engineering teams who want to hand whole coding tasks to an AI.
What it does: Describe what you want and it plans, writes code, runs tests and opens pull requests; it can work on several tasks at once and runs in auto mode by default.
Example: ask it to fix the login page bug and add tests; it edits the code, runs the tests and opens a PR for you to review.
Who it's for: Engineers on a ChatGPT plan who need bugs fixed, refactors and new features.
What it does: OpenAI's coding agent runs many agents in parallel, verifies work with computer and browser tools, keeps going in the cloud after you close your laptop, and reviews pull requests.
Example: hand a hard-to-find bug to Codex cloud and check progress and the fix later from another device.
Who it's for: Developers and Docker users who want to assemble AI agent teams without code.
What it does: Docker's CLI plugin: define agents in YAML, give them tools (including MCP servers) and let teams of specialized agents delegate tasks automatically; works with many model providers.
Example: write an agent.yaml with a DuckDuckGo search tool and run it with docker agent run to research and answer.
Who it's for: People who want a private AI assistant whose data stays in their own cloud.
What it does: A personal AI agent that runs entirely in your own Cloudflare account: it chats, remembers what matters, manages tasks, notes and projects, and runs reminders and routines on a schedule.
Example: deploy it with one npx create-talorys command and have it run a morning routine over your tasks.
Who it's for: Teams who want a self-hosted managed-agent platform instead of Claude Managed Agents.
What it does: An open-source managed agent platform: run AI agents on any model and any machine through an API, with tools, secrets, permissions and streaming built in.
Example: launch an agent.yaml with npx omnara agents launch and ask it to report the machine's OS and free disk space.
Who it's for: Teams researching autonomous coding agents (the README says it is a research prototype, not an official product).
What it does: A proactive coding agent for one GitHub repo: it listens to GitHub (optionally Feishu), takes work from allowed people, works in its own session and replies in the original thread, e.g. with a PR; a web panel shows what needs a person.
Example: ask it in an issue to fix a bug; it opens a work item and replies with a pull request.
Who it's for: People and companies who want a general-purpose AI agent to carry out whole tasks.
What it does: A general-purpose AI agent that carries out the tasks you describe; news reports say Manus raised over $500 million this month and partnered with OPPO to bring its agent to smartphones.
Example: news reports say OPPO's Find X10 series will ship a Manus agent mode that gets things done on the phone for you.
Who it's for: Developers who want to automate websites with code.
What it does: Give it a goal and the agent picks actions (click, type, select, scroll) from a table of page elements; the README cites a Google Flights search done in 7.1 seconds.
Example: ask “find flights Zurich to London” and it fills in Google Flights and searches.
Who it's for: Developers who want an AI coding workbench (README mainly in Simplified Chinese).
What it does: An AI coding workbench with a desktop app, browser UI and terminal agent; the repo holds the client, backend and agent CLI source.
Example: the start of the README only covers install and build (pnpm bootstrap); no concrete usage example.
Who it's for: Android 11+ phone users.
What it does: A phone AI assistant: say what you need and it operates apps and the web on a hidden background screen until done, while you keep using your phone.
Example: say “send a message to File Transfer” and it opens WeChat, finds the contact and sends it (README demo).
Who it's for: Makers and developers who want an AI to operate a real iPhone.
What it does: Lets an AI model use a real iPhone with a $7 ESP32: it sees the screen over AirPlay and taps through USB input, with no computer and no jailbreak.
Example: plug in the ESP32 and let the AI open apps and tap through a task on your iPhone.
Who it's for: Users who want an AI to use any app on their phone or Mac the way they do.
What it does: An agent for Android and Mac that uses any app like you do, with no root, no custom ROM and no per-app integrations.
Example: install the APK, grant overlay and accessibility, tap the floating ball and say what you need; it works the apps for you.
Who it's for: Teams who want a self-hosted proactive work assistant in Slack or Teams.
What it does: An open-source proactive agent that starts on your work without being asked and keeps you in the loop; each agent has its own computer and works on web, mobile, Slack or Teams.
Example: self-host it, add it to Slack, and it starts your day with a summary and reports progress on its own.
Who it's for: Podcasters, creators and teams who record conversations or talking-head videos.
What it does: Four video-only agents: Producer preps the session, Editor turns raw recordings into an episode, Clipper cuts social clips, Marketer writes captions and schedules posts; nothing is published until you approve.
Example: after an interview, the agents remove pauses and filler words, cut highlight clips and schedule posts for your approval.
Who it's for: Software teams and QA who want bugs caught before release.
What it does: Mo is an AI QA engineer that runs autonomous test sessions in the cloud on web or hosted iOS/Android; a separate agent reproduces each bug before Mo reports it.
Example: start Mo from a PR comment; it tests the preview deployment, posts its findings and sets a commit check.
One email a week, free. Enter your email on beehiiv to subscribe; unsubscribe anytime.