neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
翻譯準備中
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證Podium is an ADE with built-in cross-harness subagents and agent communication. It runs on desktop, mobile and VPS.
翻譯準備中
取自儲存庫 README
正在載入完整 README。 以下為已儲存的摘錄。
Think bigger. Your agents will keep track.
An open-source agent development environment. Work through an idea with your coding agent, then ask it to organize the tasks and coordinate the team. Keep the whole effort visible and under your direction.
Your projects, their tasks, and the agents doing the work, all in one place.
More agents shouldn't mean more project management檢視完整 README ↗You have agents that can write code, investigate a problem, review an approach, and test the result. But someone still has to split up the job, pass along findings, explain dependencies, and remember what hasn't finished. Usually, that's you.
Podium gives your agents a shared task system they can operate themselves. The agent you're talking to can become the coordinator: create subtasks, bring in other agents, link dependencies, and track their results. You see the same work in the board and team views, and can join any conversation along the way.
Start with a rough idea. Work out the approach together. Then give your agent a bigger job.
For a feature you've already discussed, try a prompt like this:
Act as the coordinator. Break this into phases and subtasks, and link their dependencies. Ask Claude Code to propose the architecture and Codex to review it. After we've agreed on the approach, delegate implementation into separate tasks that can run in parallel. Have Grok test the result against our original plan. Keep the tasks updated and bring decisions back to me.
Pick the agents and models you want for each part. Podium supports cross-agent delegation and communication across Claude Code, Codex, Grok, OpenCode, and Cursor. Use the agents you've set up on your machines; their accounts and usage limits still apply.
Be explicit about coordination. Ask for phases, subtasks, dependencies, and reviewers. Those instructions shape how the agent uses Podium's tools. You can change the plan in conversation as you learn more.
An issue in Podium holds the goal, the breakdown, dependencies, and progress. Agents create and update that structure directly through Podium's CLI and MCP tools. The same issues are what you use to understand and steer the work.
That gives coordination somewhere to live beyond a single conversation. A session can end while the task, its brief, and its recorded progress remain available to the next agent. A follow-up can wait on a prerequisite without you keeping it in your head.
Work is organized around tasks. Each issue can own its branch and workspace, with several agents contributing to it. Separate implementation tasks can have separate workspaces; agents sharing one workspace still need clear ownership of their changes.
We used Podium to build a competitor research and comparison system. We asked the initial chat agent to coordinate the effort, establish a research contract, and organize the phases. More than 50 research agents investigated competitors in parallel. Prototype and production tasks for hundreds of comparison pages were already in the task system, waiting on the research they needed.
We could discuss the approach and follow the work without manually handing every result to the next agent. That's the experience we're building Podium around: taking on work that would otherwise be too much coordination to enjoy.
Run locally or install Podium on your own VPS. Agents on that server can keep working when your laptop is off. Reconnect from your desktop, browser, or phone.
Podium is pre-1.0 and moving quickly. This README describes current development; check the release notes for what's in a packaged build.
See the update guide for stable and edge channels.
The download page picks the build for your machine. macOS is supported on Apple silicon and Intel; the Linux desktop AppImage is experimental, and Windows is coming soon. See the installation guide for details.
Open Podium, set up the coding agents you want to use, and start with a repository and a task. You can begin with one agent, then ask it to coordinate a larger effort when you're ready.
Install the headless server on Linux x86_64 or ARM64:
curl -fsSL https://podium.do/install.sh | sh
Follow the setup prompts. If setup didn't run during installation, run podium setup. Run podium to start the instance; it prints the browser URL. If your shell doesn't yet find the command, open a new login shell so the installer's PATH change takes effect.
The installation and networking guide covers remote access through Tailscale, Cloudflare, or your own reverse proxy. Connect from the desktop app or a browser, including your phone's browser.
To add another machine, copy the pairing command from Settings → Machines → Add machine and run it there. Choose where your agents work while keeping the task system in one place.
Follow CONTRIBUTING.md for prerequisites and setup, and ARCHITECTURE.md for the system design.
Tell us about the job you tried to hand to your agents and where you still had to do the coordination yourself. Those are the problems we want to hear about.
Apache License 2.0. © 2026 Michael Wirth, Till Felippi and the Podium contributors. See third-party notices.