Better model selection and routing
The decision of which model takes on a task gets even more accurate: better assessment of difficulty and risk, finer grading between cheap and strong models, and more traceable reasoning behind every choice.
Argo isn't a finished product — it's a system that gets more reliable, more secure, and easier to use with every release. This page shows, in priority order, what's coming up next, plus the goals that shape the long-term direction — deliberately without fixed dates, because quality comes before speed.
What's in progress now builds on the existing system — it doesn't replace anything, it sharpens what's already running.
The decision of which model takes on a task gets even more accurate: better assessment of difficulty and risk, finer grading between cheap and strong models, and more traceable reasoning behind every choice.
Agent Argo shouldn't stay limited to Windows: iOS and Linux versions bring the same orchestration, control, and traceability to more platforms — adapted to each platform's own sandboxing and file-access characteristics.
The interface should make the system's depth more accessible: a clearer overview of running and past runs, a more understandable view of plans, decisions, and checks — less searching, more understanding at a glance.
Further hardening of the protections around workspace access, stored knowledge, and file operations — more control, finer-grained permissions, and additional safeguards wherever Argo works with sensitive data.
Two goals shape where Argo is headed long-term — in this order of priority.
Project-specific memory grows into a broader system that doesn't just store knowledge, but actively structures, links, and makes it usable for new tasks — beyond individual projects.
An open ecosystem for extending Argo with additional capabilities — controlled, and under the same security and approval principles as the rest of the system.
This roadmap is prioritized, but not set in stone. If one of these priorities matters a lot to you — or something's missing — we'd like to hear about it.