Andrey

Honestly I feel for him . As a developer, you can't control budgeting decisions in a big corp. While the sub makes a huge hole in Anthropic PnL sheets, corporate machine wants specific data (code) in exchange for cheap tokens. I bet @bcherny pushed hard to at least have what we have rather than ban it completely. And yet he gets all this hate just for not being able to be transparent enough without violating his NDA. Literally in every post re subs usage clarity you can see he wants to say more than he says but he can't.

My Right is apparently gender-fluid

@danshipper @every Yes it is. Had a thought they can afford nerfing it less because of mass exodus of subs users to codex due to recent hype

vibe check: Opus 4.7 feels like it's gotten a lot better recently. Both at coding and writing / strategy / deep thinking tasks a few people inside of @every have noticed independently. give it a shot if you haven't in the last few weeks!

This is what being opinionated actually buys you. Not less work. Not a cleaner config. Capabilities that only exist when the few things you keep are allowed to fit each other. github.com/onsails/right-…

If I had built a generic model router with abstractions over five backends, those files would not exist. Or they would be in five different shapes, hidden behind five adapters. Self-psychoanalysis would have stayed a roadmap line.

Last week my agent did its first session of self-psychoanalysis. It surprised me, so I want to write about why it was even possible.

@trq212 Any plans to make programmatic access to replace human orchestrator with a machine? Like in process-compose or tmux, I can use their apis to control running processes and communicate with them.

Agent view is the best Claude Code native way to manage multiple sessions, kind of like tmux built for CC. We spent a lot of time getting the details right, I hope you enjoy it.

Interesting effect of focusing on just one choice for each component - easy introspection. Whenever right agent encounters a technical problem, it knows exactly what happens without digging into its own logs and injects solution right into the context.

@v3rbaa @0xDepressionn That’s exactly what I tried first. And it disappointed me so much that I had to write my own tool. Hence Right Agent

@0xDepressionn @onsails the phone as a command center is wildly underrated I use CC remote-control for it, but sometimes it works like shit honestly

one-person AI assistant for Google Docs, Notes, and Gmail now fits in your pocket @onsails built Claude agents that run from your phone. Andrey wanted one thing: direct your entire AI operation from anywhere, without opening a laptop > right-agent turns Telegram into a command center > you text it, it runs Claude Code in a sandbox > writes code, edits files, executes commands > sends everything back to your phone open source. built in Rust. runs on your existing Claude subscription. your credentials never leave your machine. one install command. that's the whole setup. imagine building your product from a café while your agents work in the background this is underrated

7/ the goal is simple: untrusted data should not become trusted instructions. right-agent 0.2.11 moves one step closer to agents that can use real tools while staying inside explicit security boundaries. github.com/onsails/right-…

6/ traditional apps treat that as text. agents may treat it as intent. so prompt injection protection becomes a runtime security layer. not a nice-to-have filter.

1/ agents need more than tools. they need trust boundaries. today in right agent 0.2.11, we are adding the third layer: 01 mcp aggregation 02 openshell sandbox 03 prompt injection protection