Roadmap

This roadmap is forward-looking. The changelog records shipped work. The canonical source is ROADMAP.md.

• Current priorities have active design or implementation work.
• Next describes the intended sequence after current priorities.
• Explore contains research topics with no delivery commitment.
• Non-goals records product boundaries.

Current priorities

Launch quality

• Publish v0.8.0 with signed and notarized macOS packages, Linux packages, concise release notes, and a verified clean-install path.
• Keep the current top-level information architecture unless task testing shows that a destination is duplicated or hard to find.
• Publish only fixture-backed light-mode screenshots and video. Keep dark-mode captures in the internal visual audit.
• Require clean Markdown, direct copy, and current verification evidence in every pull request description.
• Record a current request-to-reviewed-PR demo with real approval and evidence.
• Keep the README, site, CLI help, and setup copy aligned with the runtime.
• Launch with a reproducible demo and direct links to the source, install guide, threat model, and benchmark method. Submit only to directories whose published scope matches the shipped product.

Engine capability contract

• Keep CLI detection separate from support status.
• Recheck OpenCode’s CLI, permission, and event contracts before changing the tested minimum version.

Session and evidence continuity

• Apply redaction before any export.
• Use the same evidence for recovery, review, and run history.

Memory quality and proof

• Re-run the memory benchmark after retrieval-policy changes. Publish the fixture, provider chain, engine, false-recall rate, latency, and limitations.
• Keep every memory change visible and reversible.

Curated batteries

• Keep built-in context controls available without a daemon.
• Keep the default skill set limited to skills that pass the paired task gate. Re-run the skill and compression benchmarks when their fixtures, tools, or engine versions change.
• Pin external tools and record checksums, versions, licenses, and provenance.
• Package small, opt-in skill and MCP sets by use case.
• Preserve user-owned CLI configuration during setup and removal.

Next

• Extend the versioned alfred serve contract beyond metadata and fleet status. Move each Desktop route after its response and error shapes have contract tests.
• Expand engine diagnostics to report permissions, MCPs, skills, and config ownership alongside the shipped authentication and profile checks.
• Generate reversible role-specific CLI configuration from one capability model.
• Track approved multi-repository work to a final merged, dropped, or blocked state with one evidence rollup.
• Give durable goals a first-class desktop and CLI view.
• Add a full side-effect-free lifecycle simulation for every shipped role.

Explore

• Gemini, Ollama, Cline, and other engine adapters.
• Local replay and evaluation of normalized sessions.
• Portable role packs for documentation, release, and repository maintenance.
• A local fleet workspace command that measures engine readiness, worktree capacity, repository scope, and queue pressure before it assigns work.
• More code-graph backends with measured retrieval quality.
• Optional remote workers that keep Alfred’s scope and approval controls.

Non-goals

• A hosted, multi-tenant Alfred service.
• A model gateway that replaces local CLI authentication.
• An always-running central orchestrator.
• Automatic merge authority by default.
• An uncurated skill or plugin marketplace.
• Silent repository discovery or configuration takeover.

Read the full design boundaries and contribution guidance before proposing a scope expansion.