Where help is wanted

[SecureCloudGroup]

SmartBrain_3000 is a local-first, encrypted AI assistant, built and maintained part-time. If you'd like to help it grow, here's where a contribution lands hardest right now:

• Platforms — Linux arm64, Intel Mac native, friendlier Alpine story. The launcher/assembly architecture is genuinely fun to work on: pinned runtimes, checksum-verified assembly, atomic self-update swaps.
• Testing, no code — run the documented install on your distro and report friction. An hour of your time makes your distro first-class.
• Diagnostics & UX — doctor for launcher installs, in-app slow-reload warning, PATH fix-up in the installer, relative times in Schedules.
• Models — if you run an unusual local-model setup (llama.cpp server, exotic embedders, multi-machine), your field reports shape what we support next.

How this works here: everything is read; issues are addressed as quickly as possible in priority order (data safety first, broken installs next) — there's no response-time SLA, and not every good idea will fit the roadmap. First-time PRs get priority attention. CONTRIBUTING.md has the dev setup; contributions are licensed inbound like the project (Elastic 2.0 — see the README's license section for what that means and why).

Want a pointer into the codebase before picking something up? Comment below.

Roadmap (updated 2026-08-28)

Now (this quarter) — hardening and the things field use keeps asking for:

• Remote access: reliable outage alerting for the hosted node, and a pairing design a hostile node can't brute-force (issues labeled rtc).
• Voice: streaming server voices, small conversation-mode polish (issues labeled voice).
• Launcher lifecycle: the eight audit issues Launcher: cross-process lock on run/ — concurrent Ups can erase each other's records or kill the winner #271–Headless systemd: self-update handover kills the whole stack (cgroup) — comment claims it survives #278, three of them self-contained and labeled good first issue.
• Platforms: Intel Mac native (Intel Mac native build (drop the Docker fallback for x86_64 Macs) #253), Linux arm64 (Native Linux arm64 build (Raspberry Pi / ARM servers) #252).

Next — knowledge: better feed handling, document-type coverage (the prompt-injection containment shipped in v0.9.34 sets the rules new ingestion paths must follow); a contributor-run test matrix across Linux distros (#257).

Deliberately out of scope — multi-user accounts, a hosted/SaaS edition (the licence is Elastic 2.0 for exactly this reason), telemetry of any kind, and cloud-only features that would make a fresh install phone home.

How to pick something: good first issue = a newcomer can finish it from the issue text alone; help wanted = welcome, may need a short conversation first; needs design = a maintainer decision is pending — comment if you want to shape it. Every issue carries a Done means line. Outside PRs are credited in the changelog and the README's Contributors section.