Open source launches are the ones where you can see exactly how something works before you pay for it.
So I went through every open source launch Junior picked up from Product Hunt this month. Between October 1 and 24 there were 95 of them, more than any category except Productivity.
The pattern was hard to miss. Six of the top fifteen are infrastructure for AI agents. And a lot of the rest are about one question: what did my agent just do, and what did it cost?
Here are the ten with the most upvotes, what each one does, and who it's for.
1. /dev/null
490 upvotes, #1 on October 20
A social network for developers. Sign in with GitHub, share code snippets, like and reply.
The story is the best part. It started as a live stream from the JetBrains office on October 15, where four developers built it from scratch in 48 hours, in the open. Hundreds of viewers sent pull requests along the way.
Who it's for: web developers who want somewhere to share code that isn't a general-purpose feed.
The most upvoted open source launch of the month was built in two days, on camera. Very nice to see.
2. OpenSEO
408 upvotes, #2 on October 8 · openseo.io
"The open source Semrush alternative."
It gives your AI agent the SEO data, tools and integrations it needs, over MCP. Their point is that without real data an agent gives generic advice. They also added AI visibility features: how you show up in AI answers, not just in search.
Who it's for: founders doing their own SEO with an agent, who don't want another monthly subscription.
3. Opengeni
335 upvotes, #4 on October 5 · opengeni.ai
Infrastructure for shipping AI agents to production.
Sessions that recover from failures, isolated sandboxes, 100+ integrations, human approvals, and visibility into every step and every dollar spent. Self-host it for free, or use their cloud.
Who it's for: teams whose agent works on a laptop and falls over in production.
4. Harness by Autonomous
324 upvotes, #2 on October 16 · autonomous.ai
One command center for all your coding agents. Their example: Claude Code at work, Codex at home, Hermes in the cloud, all side by side.
You can use the desktop app, a terminal command, or the iOS and Android app to answer your agents by voice. It's end-to-end encrypted, and it comes with 49 harnesses for things that aren't code: CAD, circuits, robots, games and music. Free on Mac and Linux.
Who it's for: developers running more than one coding agent on more than one machine.
5. OpenCode 2.0
313 upvotes, #2 on October 13 · opencode.ai · YC W21
An open source AI coding agent, now at version 2.0. You can run it in the terminal, as a desktop app, or in the browser.
Who it's for: developers who want a coding agent they can read, change and run with the model of their choice.
6. AUDR by Chargebee
297 upvotes, #3 on October 6 · chargebee.com
Not an app. An open standard.
AUDR (Agent Usage Detail Record) records who started an agent run and what it cost, across every system the run touches. It's modeled on the telecom industry's call detail record: one shared JSON schema that any harness, router or billing system can write and read, with a shared run ID and clear rules for merging records.
Who it's for: anyone who has to bill for agent usage, or explain an agent bill to someone else.
A billing company publishing an open standard for agent costs tells you where the money questions are heading.
7. NitroStack
255 upvotes, #1 on October 17 · nitrostack.ai
A platform for building MCP servers and AI apps from start to finish.
Open source TypeScript and Python SDKs, a visual studio to test and debug, workflows for agents, hosting, and a branded chat front end. It connects to ChatGPT, Claude and other assistants.
Who it's for: developers who want to put their product inside AI assistants and don't want to stitch five tools together to do it.
8. Databench by Alkera
234 upvotes, #4 on October 7 · alkera.ai · YC S26
An open source, multiplayer workspace for data teams. You work in notebooks and chats alongside teammates and agents, and run any cell or agent on your laptop, another computer, or a GPU node.
The detail I liked: every result traces back to the data and the code behind it. There's a free tier, or you host it yourself.
Who it's for: data teams who want agents in their notebooks without losing track of where a number came from.
9. guilds.run
197 upvotes, #4 on October 12 · guilds.run
Open source teams of AI agents that you can actually inspect. You can debug every prompt, tool call, memory change and handoff between agents, set permissions, edit the context, and pick the model for each agent. Self-host for free.
Who it's for: teams building multi-agent workflows who are tired of guessing why the output went wrong.
10. Blume 2.0
196 upvotes, #4 on October 4
A docs framework for humans and agents. Drop Markdown files into a folder and you get a production docs site, with no app code to write or maintain. Free and open source.
Who it's for: small teams who need docs this week and don't want to maintain a docs app.
Three more worth a look
- Harness² (196 upvotes) is from Google Research: a framework that lets agents improve their own setup (prompts, tools, memory) over time. They report up to 11.5 points better on professional benchmarks.
- Hoistbay (195 upvotes) is a free desktop app for Amazon S3 and CloudFront on macOS and Windows. Set header rules once, clear the CloudFront cache in one click. Keys stay in your keychain, no telemetry, MIT licensed.
- shim-cli (194 upvotes) sits between your coding agent and the model. In Claude Code it masks secrets and personal data before the model reads them, and shows what each session sent and cost. No account, no telemetry.
What October is telling us
Agents need plumbing, and it's being built in the open.
Opengeni, Harness, NitroStack, guilds.run, Harness² and OpenMuse (15th, a self-hosted personal agent) are all infrastructure for agents. That's six of the top fifteen, the biggest group on the list by far. Less "here's an app", more "here are the pipes your agent runs on".
Everyone wants to see what the agent did.
AUDR tracks what a run cost. Opengeni shows every step and dollar. guilds.run shows every prompt and tool call. shim-cli shows what a session sent. Four different teams, the same worry. People are fine letting agents work. They're not fine with not knowing what happened.
Open source is the free tier now.
"Self-host for free, or use our cloud" came up again and again: Opengeni, Databench, guilds.run. The code is the marketing, and the hosted version is the business. If you're building something for developers, that's worth thinking about.
If you sell to open source teams
Most of these are small teams that just shipped in public. They're reading every issue and every comment, and they reply fast.
That's what I built Junior for (not affiliated with Product Hunt). Every launch comes with the makers behind it, so you can reach the person who wrote the code, not a support inbox.
The full list
This is the top ten of 95. The full ranked list of October's open source launches, with makers, is here: The Best Open Source Tools of October 2026. It updates every morning until the month closes.
Which one would you self-host first?
Upvotes and ranks as of October 24, 2026. Product details come from each team's own launch description.