How to read this
- Shipped — "Now shipping." Wired, tested, and in the current pre-release build.
- In flight — "Shipping with v0.x." Scoped, designed, and under active development.
- Planned — "On the public roadmap." Scoped but not yet under active development. Subject to change.
We don't publish dates because we don't want to make promises we can't keep.
Shipped
Foundation
- Tauri 2 + React + TypeScript monorepo (Phase 0)
- Chat-first first-run, no wizard (Phase 0.5)
- Onboarding v2 — Welcome frame, Import Hub, Provider Intelligence, contextual tour (Phase 1.6)
Agent core
- Chat & agent core, streaming, tool calling, context assembly (Phase 1)
- Provider expansion & MCP (Phase 3)
- Sub-agents in git worktrees (Phase 4)
- Five routing profiles & smart routing (Phase 2)
- Token optimisation, caching, condensation, Budget Bridge (Phase 2)
- Project intelligence, milestones, drift detection (Phase 5)
- Session recovery & stall detection (Phase 1.5)
- Editor & LSP integration (Phase 1.5)
Surfaces (the four lenses)
- Project surface — code editor, file tree, terminal, debugger (Developer)
- Canvas surface — visual no-code builder (Maker) (Phase 2.5)
- Document surface — long-form writing (Writer) (Phase 6.5)
- Workflow surface — visual automation builder (Power user) (Phase 6.5)
Polish & quality
- UX polish & accessibility pass (Phase 6)
- Repo hardening, linter consolidation, CI gates (Phase 9.6)
- Source-available under the FSL (with Apache 2.0 Change Date 2028-06-12)
In flight — Phase 1.7 (BYOK, Free Gateway, Provider Intelligence)
The goal: replace the "connect a model" chat-first flow with a proper Provider & Models panel. Phase 1.7 answers "what model do I talk to?" — Phase 1.6 answered "where is my project?".
- Free multi-provider gateway (zero Miton spend)
- Per-provider free-tier limits (canonical source-of-truth)
- OpenCode Go subscription (goodwill refer)
- First-message consent screen (explicit user acknowledgement)
- Provider & Models panel in Settings
- Model discovery and selection UI
- Session cost indicator and model health
- Fallback toast and key rotation
- Rust provider key validation command
- Offline provider key lookup (prefix + fuzzy)
- Provider catalogue DB (SQLite, models.dev)
- Top-bar deployment picker
- Model dashboard (model-quality and health)
- Budget banner (visible free-tier limit warning)
- Local identity and paid-plan referral
- Free provider signup links & marketing consent
Planned
v0.x — closing out the pre-release cycle
- Memory store (cross-session memory)
- Project intelligence enrichment (deeper knowledge graph)
- Visual workflow builder (first-class node editor)
- OpenRouter fusion server tool (server-side model fusion)
v1.0 — the public launch line
- Pro and Team pricing live (ADR-0007)
- Managed proxy (Mistral) with 15% markup
- Skills, commands, and sub-agents (open standard, agentskills.io)
- App Builder (no-code application builder for Makers)
- Hermes (cross-surface integration surface)
v1.x — post-launch
- Collaborative editing (multi-user document and project editing)
- Plugin marketplace (curated skills, sub-agents, MCP server presets)
What's not on the public roadmap
These are public commitments to users about what Miton will never be:
- No cloud agents. Miton never executes your work on a vendor VM. Everything runs on your machine.
- No telemetry on user data. No chat, code, keys, or files ever leave your machine without explicit, in-app user action.
- No proprietary model lock-in. Miton does not ship its own model. The model layer is a uniform adapter over external providers.