Where protoAgent is headed

an append-only log of everything the model sees, so resume, fork, search and replay all work from one session record.

protoAgent as an Ollama community integration and a Hugging Face "Use this model" local app; the listing PRs are open upstream, awaiting maintainer review.

migrating the remaining bespoke console surfaces onto DS primitives as the design system grows (bottom sheet for the mobile session sheet, dense icon buttons — tracked in protoContent).

management UI, peer rotation, and fleet integration for ADR 0066 tokens.

"Open in New Window" spawns a real second desktop window with its own chat surface.

opt-in install of a plugin's requires_pip packages inside the frozen desktop build: pure-Python wheels onto the host path, or into the managed Python runtime that also gives the desktop app a working execute_code.

per-turn trajectory export from production agents into the lab for downstream training-data collection.

goals, tasks, scheduling, and watches compose into one self-directed OODA loop, with durable task→goal attribution.

plugin tools save generated images/audio/video the chat renders inline (signed URLs, bearer-safe), and can return images a vision model actually sees.

opt-in per-turn trajectory rows (OpenAI chat format + verifiable reward) with a PII-redacting sync pipeline.

import scripts that carry an existing agent's config, memory, and history into protoAgent.

jump a conversation back to an earlier message and branch from there.

installed version, "update if available", and uninstall — from the rail context menu and a settings panel.

a curl | sh bootstrap with an interactive CLI config wizard.