What you'll get here:
• AI tools you can actually use this week
• Automation workflows for solo builders
• Agent setups with steps, not slogans
I post at most 2 original English posts a day.
If a claim needs a source, it gets one.
Follow for practical AI + automation.
Linux gets RPM and Debian packages. On macOS and Windows you build a Docker image from their Dockerfile (no published image). Unimplemented operations return 501.
Blog: https://t.co/pJL7k6yWfu
Repo: https://t.co/RmjYRuwJRd
AWS made a free local emulator of its own APIs.
AWS Emulator lets you and your coding agents test S3, DynamoDB and more on your machine, no AWS account needed.
Limits: one account, IAM policies not checked like AWS, telemetry on by default (opt-out).
CC0-1.0.
Needs Node 24.22.3+ on macOS or Linux. You can also self-host the server or use the hosted console.
Blog: https://t.co/olGCeddGaf
GitHub: https://t.co/Q2wldLARjy
Running agents in Claude Code or Codex? AgentRun adds a control plane on top.
Open-source alpha from Parcha Labs:
- Durable execution and run tracking
- Works with Claude Code, Codex, OpenCode and pi
- Starts locally with a single npx command
Apache 2.0.
Needs Docker or Podman with at least 4 CPUs, 4 GB memory and 40 GB free disk. Desktop app for Mac and Windows.
Blog: https://t.co/ur3h0ac4qA
Release: https://t.co/VPlxWYZaQy
A self-hostable agent workspace where Bots ask before they delete, send or post.
OpenBot 1.0 (CopilotKit):
- Desktop app sets up everything, no terminal
- Teach a Bot by doing a task once
- Use your Claude, ChatGPT or xAI plan
MIT, free for personal use.
Free software; model charges depend on the provider you connect. It asks permission before tool actions, and /diff shows changes. Preview: config and behavior may still change. Community PRs paused until after Dec 2026.
Release: https://t.co/4qXYhYlGaM
Qwen Code, an open-source coding agent for your terminal, hit 1.0 preview.
From the Qwen team:
- Native packages for macOS, Linux, Windows; no separate Node.js
- Qwen/Qoder account, API key or custom endpoint
- Resume sessions or run non-interactively
Apache 2.0.
Needs macOS with full Xcode, an iPhone on USB (unlocked, trusted, Developer Mode on), a signed Appium WebDriverAgent, Python 3.13, Node 24+ and a signed-in Codex CLI. Not fully offline: task and screen text go to the model.
Repo: https://t.co/JE1afQYxaU
Let an AI agent control your USB-connected iPhone from your Mac.
PhoneUse Local:
- Drive it with prompts, live video, a REST API or MCP
- Experimental developer preview
- Separate from the hosted PhoneUse product
- Prompts run through your own Codex CLI login
MIT.
Free app; your model provider bills you. Needs macOS 14+ (Windows 11 preview needs WSL 2). No n8n account needed. Stops when your computer sleeps. Usage data sharing is on by default; you can turn it off.
Install: https://t.co/SdvrXY6MXD
Repo: https://t.co/fwBJ2TNGHz
A personal AI agent that runs on your own computer and asks before it acts.
Nod v0.6.2 from the n8n team (research preview):
- Its own sandbox and browser
- Finishes jobs on real websites
- Bring your own API key, ChatGPT Plus/Pro, or a local model
Apache 2.0.
No model is included. The desktop app runs from a source checkout; self-hosting is experimental and in active development.
Run it (needs Node.js 22+ and uv):
git clone https://t.co/Y8L12TWiGq
cd suparagent/desktop
npm run app
Repo: https://t.co/3RuHgodL0i
Named AI coworkers, each with a job and a workspace.
Suparagent (formerly Open Duet):
- Chats, memory, routines, approvals in one place
- "Ask first" mode pauses for approval
- Routines never self-approve actions
- Bring your own engine + provider account
Apache 2.0.
Two new read-only Hermes agent templates.
- Fare Hawk plans trips and watches fares, never gets your card
- No live fare source? Results say "fixture/demo, not live fares"
- AI Claim Checker rates launch claims: supported, vendor-only, contradicted or unverifiable
MIT.
Run a team of AI bots in a local chat app on the Claude, Codex or Grok CLIs you already use.
OpenMausBot 0.1.150:
- macOS, Windows, Ubuntu
- Uses your existing CLI logins, no new accounts
- Transcripts and keys stay on your machine
- You approve what matters
Apache 2.0.
@flyinghigherai Hybrid BM25 + vector with RRF is a solid base. For the escalation router, logging the confidence score next to each route decision makes threshold tuning much easier later.
@axdavid_ai Nice pattern. One caveat: add a review or dedupe step before saving extracted facts, otherwise one bad extraction gets reused in every future reply.