Repository navigation
Dev Companion v0.8.0 — local desktop companion preview
Pre-releaseDev Companion v0.8.0 — a calmer desktop companion
This Linux x86_64 preview keeps the existing local project ledger, approvals and offline model path. It adds a forest-toned quick panel and work desk with bundled offline fonts, 16 original companion characters plus no pet, pixel/soft presentation, goal starters, a downloaded local-model picker, persistent hide/show, and focused push-to-talk. An optional session-only global shortcut asks through the desktop permission dialog. Speech fills a draft for review; it never approves work.
The project-work flow still separates an inspected source range, a model interpretation, a proposed diff, approved application and an explicitly authorized check. Source explanations now carry inspected excerpts between local-model steps. When a requested citation remains unusable, the result is marked needs input and shows the inspected ranges and unverified interpretation.
Validation: 122 ordinary tests passed (two pre-existing installed-runtime checks excluded); formatting, Clippy, Python syntax and staged known-pattern scan passed. The exact static binary passed a controlled-model embedded GTK journey covering model selection, personalization, hide/show, goal presets, review, approval and continuity. An external-network-isolated continuing session used downloaded Qwen on a narrow source explanation and retained the saved goal. The archive contains only the binary, MIT license, third-party notices and quickstart.
Known limits: Eight retained Qwen attempts on the larger pinned Offline-GPT repository were not reliable. The last attempt read relevant implementation and tests but misstated test coverage and supplied unusable citations; it ended as needs input. Physical microphone comfort and portal shortcut permission have not been tested with a person. Independent usefulness and returning-user adoption remain unmeasured. The VS Code adapter asset is unchanged from v0.7.1.
Initial Ollama/model and optional voice downloads require internet. Runtime inference and speech have no cloud fallback. Projects and background investigation need explicit enablement; edits and checks still need separate approval. Run dot update --check and dot update from an installed v0.7.1 build, or use the archive and setup guide.