neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
For individuals and enterprises, take control of the computer like a human, using visual perception and cross‑application capabilities to achieve automated workflows.
AI translation
面向个人与企业,像人一样接管电脑,依靠视觉感知与跨应用能力实现自动化工作流。
Developer toolsRust · TypeScriptGitHub @chinphingOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
No updates detected yet. GitHub releases will appear here.
Owners of public repositories under a personal GitHub account can sign in to verify ownership and manage the introduction themselves. No administrator approval is needed.
Sign in, generate a verification file, and commit it to this repository’s default branch within 30 minutes. Files on other branches or forks do not count. Then check the file below to manage the project without administrator approval. Keep the file to retain your access.
If needed, you will go to sign-in. After signing in, return here and press your chosen verification button again.
Run /nomorevibe verify in your project folder to verify ownership and manage the introduction yourself.
Verify in one minuteAsk about other verification options
An out-of-the-box, ultra-low-resource desktop agent foundation for building digital employees. Built for developers and the enterprise.
Taken from the repository README
Loading the full README. The saved excerpt is shown below.
An out-of-the-box, ultra-low-resource desktop agent foundation for building digital employees. Built for developers and the enterprise.
Build and run digital employees on the computer people already use, and keep them running stably on a server. Both the agent and the model can stay inside your own environment, with no per-seat fees. The desktop app and the web app share the same capabilities, on Windows, macOS, and Linux.Read the full README ↗An out-of-the-box, ultra-low-resource desktop agent foundation for building digital employees. Built for developers and the enterprise.
Build and run digital employees on the computer people already use, and keep them running stably on a server. Both the agent and the model can stay inside your own environment, with no per-seat fees. The desktop app and the web app share the same capabilities, on Windows, macOS, and Linux.
| You want | Get it here |
|---|---|
| Ready to use, out of the box | Official download |
| Community installers | GitHub Releases |
| Deploy inside your company | Deployment guide |
A digital employee has to work inside a real office setup: a license that allows commercial use, a desktop client and a web app, a small footprint, eyes and a mouse on the screen, and one conversation that can keep going.
Pointer is more than a runtime for Skills. It builds the business system, helps turn manual work into a Skill that drives it through its UI and HTTP interfaces, and carries what manual runs teach back into the Skill and the system itself.
| Capability | What you get |
|---|---|
| Agent system | Built-in general, coder, computer, and explore. Sub-agents can run in the background so the main conversation can continue |
| Chat engine | Streaming, prompt assembly, and tool calling; long conversations, context compression, and milestone recall |
| Tools | Terminal, file read/write/search, web search and fetch, task board; vision computer control for GUI apps (still being improved) |
| Skills and extensions | Skills, plugins, and user rules; external tools via MCP |
| Dispatch and triggers | Scheduled tasks, webhooks, and IM inbound events enter chat through one dispatcher |
| Entry points | Desktop, web, and Feishu, DingTalk, WeCom, and WeChat |
| Model access | OpenAI-compatible APIs. Qwen and DeepSeek are built in; Doubao, Kimi, Zhipu, OpenRouter, and others can be added—no single-vendor lock-in |
After a role is written as a reusable Skill, you need a shared format, an easy install path, editing in the conversation, and a way to validate it at scale.
| What you need | What Pointer provides |
|---|---|
| Write in a shared format | Skills use a common skill format, compatible with Codex and similar directories. Instructions, references, scripts, and assets stay separate |
| Install, then turn it on | Enable skills in the library, including zip import. Existing skills can be imported from Codex, Claude, OpenClaw, and Hermes |
| Author and edit in the conversation | Creating, editing, reviewing, and packaging goes to the coding sub-agent. A skill can also ship inside a plugin |
| Skill validation | Run the Skill under high-concurrency background sub-agents in isolated environments, and sweep many test samples quickly to check stability and correctness |
| Fill in a missing runtime | If Node, Python, or another runtime is missing, install it in the conversation and keep testing the skill |
Start with getting started.
The desktop app, web app, and IM channels share one pointer-core. The UI handles interaction; orchestration, tools, Skills, and sessions live in the core.
Layer notes: docs/en/developer/architecture.md.
| Audience | Start here |
|---|---|
| Users | docs/user/ |
| Developers | docs/developer/ |
| Contributors | CONTRIBUTING.md · docs/en/DEVELOPMENT.md |
| Security | SECURITY.md |
| Changes | CHANGELOG.md |
Full index: docs/README.md.
The desktop app is Tauri 2 and Vue 3. The web UI is the same interface served by pointer-server. Chat, tools, Skills, and agents live in crates/pointer-core.
npm install
npm run tauri:dev
The desktop window opens with no control plane bound when pointer.local.env is absent. The first compile can take a few minutes.
Use two terminals:
# terminal 1
npm install
npm run server:dev
# terminal 2
npm run web:dev
Then open http://127.0.0.1:1420/. Details: docs/en/DEVELOPMENT.md.
Users start the desktop client with the script. It does not call npm. An existing target/release/pointer-app or an installed app opens directly; --build is what recompiles locally:
./scripts/quick-deploy.sh start standalone-client --runtime bin
Flavour (POINTER_EDITION) and runtime form (client / server) matter when you are building an installer:
| What you are doing | Start here |
|---|---|
| Run a client locally, or hand a self-built client to colleagues | standalone × client |
| Stand up a server for your team | standalone × server |
| Enterprise: client bound to your internal control plane | managed × client |
| Enterprise: deploy the control-plane server | managed × server |
All four — build command, variables, artifacts, verification, dependencies: docs/en/deploy/README.md. Platform packaging (Windows / macOS / Linux): docs/en/deploy/platforms.md.
Bump the root VERSION file, sync, then push a v*.*.* tag. CI builds Windows / macOS / Linux and opens a draft on GitHub Releases. CI produces the standalone client package only — managed (official / enterprise) packages are built locally or on your own CI.
# edit VERSION, then:
npm run version:sync
npm run version:check
git tag v0.1.2
git push origin v0.1.2
# or build packages on this machine:
npm run tauri:build
npm run server:build
Details: docs/en/contributing/cross-platform-build.md · docs/en/contributing/versioning.md.
Pointer learned from and was inspired by:
Apache License 2.0. Copyright 2026 刘新星 (https://pointer.readflowai.com). Trademarks and official domains are described in NOTICE.