Frink

So today, I've finally launched my AI harness - open sourced and free. If you're like me and hate long posts, TLDR: Built an AI harness, started working on it when few other apps existed and am still working on it today. Launched the app with unique features (mainly around automations) and would love feedback to see if it's worth pursuing. Backstory: I am a software engineer day-to-day and started working on my own application at the start of the year to fill gaps that no app at the time supported and make my life easier. Fast forward to today, I still work on it whilst putting in many, many hours per day - completely solo. Would love some feedback on it to see if it's worth me putting down or if it's something worth keeping, as you all enjoy it. About the app: As I said, I built it for myself first (as most people should IMO), and after developing it over time, I felt like I had a somewhat gap and carried on. It's an AI harness like most others nowadays, but as the ecosystem is changing so quickly, I thought it would still be worth sharing. Features: - Import MCPs, rules, permissions, hooks, etc. into the app, irrespective of the provider; they will all rely on the app's credentials. This means if you import your MCPs from Claude into the app and then switch to Codex, they will work natively in the app. Similar to how other apps do this, e.g., importing Claude into Codex and vice versa, this app doesn't restrict you to one provider. - "flows". Essentially graph-based automations; think of n8n or any other agentic graph workflow that's popping up recently. What's different with the app here is: -- You can get your agent to create custom nodes, so if you want an automation to say, connect to your smart home every day on a cron job - you can. Get it to build a custom node and use that node indefinitely. -- You can connect the graph to third-party plugins. The app's 'plugins' come with the native agentic plugin, plus they allow you to use those tools in flows logically instead of through an agent. -- You can spawn any number of agents under a specific flow, chained together to complete entire backlogs. Or, more technically, a DAG. -- The flow can be set up in a UI yourself or simply get your AI to build it for you. - Real code editor (not many harnesses have editors nowadays). - Better UI, allows split-screen like tmux. There are numerous other USPs of the app, but those are a chosen few. Note: Currently on macOS only, but I'm planning on launching it within the next week or so on Windows. As I said, I'm here for guidance and also to show off my application. Honest feedback is great; let me know if I should drop it and am getting carried away with the development, or if it is genuinely worth carrying on :) https://frink.dev/ Or full repo, on MIT: https://github.com/benord-labs/frink