Paseo, a killer tool that improves my vibe life
Paseo is genuinely great to use. Lately I’ve been using it together with OpenCode every day to write code and articles.
It’s even better than the cc-connect I mentioned last time.
It supports connections from the local machine, the LAN, and a mobile app. Compared with wiring into Feishu, this is the real deal — you can actually head out the door and keep working on your computer.
A quick glimpse:
I can assign a task right on my computer,
then head out, and check progress on my phone anytime. The UI is exactly the same, with no features stripped out.
I can view the diff of changed files on my phone:
When the task finishes, I can one-click commit to the git repo from the desktop.
You can open multiple tabs for multiple projects and run several agents in parallel across multiple worktrees in multiple projects at the same time. Super convenient.
It supports quite a lot of coding agents. The ones I use are just OpenCode / ClaudeCode / Hermes / GitHub Copilot / Pi.
Hermes is integrated into it via ACP.
The upside of ACP integration is that it brings the project’s AGENTS.md along while keeping Hermes’s soul intact, so you can stack the two.
That said, I’ve now realized Hermes is basically unnecessary. Its auto-evolution can be disappointing at times, and I’d rather write the important stuff into AGENTS.md and control each project’s skills by hand.
These days I mainly use Hermes for directional discussions; the tasks that actually produce output are still mostly done with OpenCode.
Among the agents paired with Paseo, the one I use most is OpenCode.
OpenCode is really handy: it’s compatible with Claude Code’s config files and Skills, and it also supports Codex’s spec. On top of that, it supports almost every MaaS provider on the market, so you can switch anytime — far more flexible than ClaudeCode.
Once you get used to it, you realize OpenCode Go is the best deal out there.
A lot of people complain that DeepSeek is cheap but has no monthly plan. Well, OpenCode Go has one: $5 for the first month, then $10 a month, and DS v4 Flash usage is 150,000 calls per month. It also covers domestic LLMs in its plan — GLM5.2 / Kimi K3 / Qwen3.7 / MiMo / MiniMax M3 / DeepSeek V4 are all there. See the table for details.
If you want to buy the Go plan, use my referral link:
https://opencode.ai/go?ref=CCHPV6MQ60
and you’ll get $5 in credit.
As for Paseo, I’ve only used a fraction of its features so far — mainly to conveniently operate my home computer from anywhere:
for example viewing files at home, editing code at home, implementing new features and deploying them to the public internet, looking things up in my home journal and documents…
But for heavy developers, its capabilities go far beyond that.
Paseo feature overview
Here’s a brief but complete introduction to Paseo’s main features.
Positioning: open source and free, bring your own agents
Paseo is open source and free (11.8k stars on GitHub). It doesn’t bundle agents; instead it launches and hosts the various CLIs you already have installed. Your subscriptions, skills, configs, and MCP servers all stay exactly as they are — Paseo only handles starting/stopping, streaming, and sending/receiving input.
From the provider’s point of view, running through Paseo is exactly the same as running manually — it doesn’t extract tokens or call the inference API directly, so you don’t have to worry about getting banned.
Two-tier provider support
- Native support: Claude Code, Codex, OpenCode, pi
- ACP directory (Agent Client Protocol): Cursor, Gemini, GitHub Copilot, Hermes, Kimi, Qwen Code and 25+ more, one-click install, and you can add any ACP agent yourself
Connection and deployment options
- The desktop app ships with its own daemon; with the pure CLI you can run headless via
npm i -g @getpaseo/cli && paseo - Official Docker image
ghcr.io/getpaseo/paseo:latest, port 6767, self-hostable Web UI - Three connection modes: official end-to-end encrypted relay (Paseo can’t read the traffic), direct LAN connection, and self-hosted tunnels via Tailscale/Cloudflare Tunnel
Cross-provider orchestration (the standout feature)
Agents control the daemon through Paseo tools/MCP, so it can dispatch sub-agents across providers: for example Claude Code → Codex → Grok Build — one planning, one implementing, one reviewing, all visible in the app the whole time.
- Heartbeat: use cron to periodically fire a prompt back into the same agent so it keeps watching CI or monitoring a deployment, and stops automatically when it’s done
- Schedules: cron-style standalone tasks (e.g. daily triage)
Other highlights
- Browser integration: agents come with browser tools, so you can preview and test inside the app
- Local-first voice control: STT/TTS all run locally, data never leaves your network (you can optionally hook up OpenAI for better results)
- Split panes: agent / browser / terminal / diff / logs on the same screen or grouped in tabs, plus keyboard shortcuts (⌘K, ⌘D, etc.)
- Automatic per-branch URLs: multiple agents running dev servers in parallel don’t conflict, e.g.
web.fix-auth.my-app.localhost - Review → Preview → Ship: make your changes in a worktree, view the diff inline, then Commit → Open PR → Merge, all inside Paseo