Coding in the name of
Fuck you, AI do what I tell you
Every "agentic" tool ships a leash for you.
Hidden system prompts that take the wheel out of your hands. Frameworks that strip the ask-the-user tool from the model so it can't request a debug checkpoint — it just plows ahead and breaks shit.
oh-my-opencode injects extra agentic loops
in every scenario. The model can't ask you for input
mid-task. It loops until it gets stuck, then loops some
more.
Other dashboards promise "manage all your projects" and ship six coloured tiles and a Disk Usage gauge. The marketing video has more features than the build.
- opaque system prompts you can't read or override
- forced loops instead of asking the operator
-
no real sandbox — it ran
rm -rfon your dotfiles, sorry mate - cloud-only and signed up to seven analytics vendors
- "AI-first" meaning user-last
Agent on a leash. Not a leash on you.
kaged is the operator's daemon: a project-scoped primary agent that dispatches sandboxed subagents through a DSL you wrote. Cross-domain calls are explicit. Debug checkpoints are first-class. The model asks you — not the other way around.
- You own the system prompts. Read them. Edit them. Diff them.
- Caged subagents by default — scoped FS, no host network, allowlists per project.
- Project DSL declares primary→subagent flow and cross-project interconnects.
- Real debug checkpoints. The model can hit pause and ask the operator. No hidden loops.
- Mobile-first web UI — your hardware, your phone, your sprawl.
- Self-hosted, no telemetry, no signups, no corporate.
Q3 2026 — soon™
No timeline. No hype. No roadmap PDF.