Roadmap

Vibe Factory is young. The list below is ordered by priority, not by date.

0.1 — foundation (released)

• Layered crate architecture with documented extension points
• Anthropic and OpenAI-compatible providers, mock provider
• Built-in tools with shell security policy and path containment
• Git worktree isolation and merge
• Agent runtime with parallel tools, hooks, context budgeting, continuation
• Spec → plan → build → QA/fix pipeline with file persistence and resume
• Plugin protocol (JSON-RPC over stdio, MCP-compatible tools) and Rust plugin helper
• vibe CLI
• User guide, design book, ADRs, API docs

0.2 — reliability (released)

• Required commands before ready/merge, persistent results and replay on resume
• Feed deterministic validation failures into the bounded QA/fix loop
• Validate the integration candidate, including assisted conflict resolution, before publication
• Ten reproducible Rust evaluation cases with independent oracles, a suite runner and a summary of success, time and tokens
• Worktree per subtask, ordered integration and conflict tests
• Container workspace provider with explicit network, mounts and resource limits
• Persistent token and duration budgets across resumes
• Cancellation-safe plugin writer task
• Cross-platform release qualification and migration guide

Carried over: a published real-model baseline, done in 0.4.0 (Claude Sonnet 5, 2026-09-27: 30/30 runs over the ten cases).

0.3 — interaction foundations and terminal UI (released in 0.4.0)

The engine learns to stream, to wait for a human and to outlive the terminal that started it, then gets a terminal UI. See ADR-007.

• Real-model evaluation baseline, published with the framework commit and model (Claude Sonnet 5: 30/30 runs over the ten cases)
• Streaming completions and live agent text
• Events v2: sequence numbers, schema version, validation, integration, budget and artefact events, replay and follow
• Human approvals after spec, plan or before merge
• Run manager and cross-process locking of .vibe/
• Terminal UI: task board and live run view

0.4 — server, web UI and integrations (released)

• vibe serve: HTTP API and server-sent events over the run manager
• Local web UI: board, activity, plan, spec, QA, changes, approvals, evaluation dashboard
• GitHub and GitLab issue import and pull request creation
• Web fetch and search tools
• Persistent project memory across tasks
• VS Code extension on vibe serve

0.5 — visibility: history, global activity, full trace, macOS app (released)

See everything that was done, from every interface (ADR-008).

• Events that tell the whole story: call ids, exit codes and output references on tool calls, the model of each session, committed and merged events, run totals on run_finished (schema stays 2)
• Trace store: the complete output of every tool call, local and git-ignored
• Read layer: incremental and project-wide event reading with a stable cursor, history rebuilt from events and git, traces of tool calls
• vibe history, vibe trace, vibe events across every task, optional [pricing]
• Server routes for the global event stream, history and trace; web views Activity, History and Trace
• Terminal UI screens Activity and History and a Trace tab
• Native macOS app on vibe serve (SwiftUI, macOS 14+): signed, notarized DMG with Sparkle auto-update

Later

• OpenAPI description of the server API
• Editing the plan from the web UI before approving it
• Linear and Jira importers

Ideas

• Semantic merge assistance for parallel tasks touching the same files
• Prompt caching statistics and optimisation hints
• A gallery of community agents and plugins

Contributions to any item are welcome — see Contributing.