# Armature — Roadmap ## Current: v0.9.x — Workflow Redesign + Multi-Surface Packages Self-hosted extensible platform kernel. Auth, identity, packages, Starlark sandbox, storage, realtime, and ops are kernel primitives. Everything else is an extension. **Kernel capabilities:** Auth (builtin/mTLS/OIDC) · Users/teams/groups/RBAC · Surfaces/extensions/libraries/workflows · Starlark sandbox (capability-gated) · Object storage (PVC/S3) + ext_data tables · WebSocket hub + realtime pub/sub · Audit log · Notifications · Scheduled tasks · Cluster registry + HA · Extension composability (slots/contributes/cross-package calls) --- ## Completed — v0.6.x through v0.8.x All completed work is documented in `CHANGELOG.md`. ### v0.6.x — MVP + Hardening | Version | Title | |---------|-------| | v0.6.0 | Cluster Registry + HA | | v0.6.1 | Backup/Restore + Docs | | v0.6.2 | Docs Polish + Dynamic OpenAPI | | v0.6.3 | Dead Code Sweep + Registry Fix | | v0.6.4 | Admin Health/Metrics + Cluster Merge | | v0.6.5 | Renderer Pipeline + Docs Rewrite | | v0.6.6 | Final Hardening | | v0.6.7 | Native mTLS | | v0.6.8 | Rebrand + Cookie Fix | | v0.6.9 | Session Lifetime Config | | v0.6.10 | Viewport Foundation | | v0.6.11 | CSS Deduplication | | v0.6.12 | Extension CSS Isolation | | v0.6.13 | Responsive & Spacing | | v0.6.14 | Visual Polish | | v0.6.15 | User Display Audit | | v0.6.16 | Usability Survey Gate | | v0.6.17 | Bug Fixes & Welcome | | v0.6.18 | CI Bundle Wiring | ### v0.7.x — Test Infrastructure + Quality Gate | Version | Title | |---------|-------| | v0.7.0 | Shell Contract + Surface Audit + Rebrand | | v0.7.1 | Surface Runner Framework | | v0.7.2 | Package Runners + CI Gate | | v0.7.3 | Extension Shell Migration | | v0.7.4 | Documentation + Surface Work | | v0.7.5 | Headless E2E + CI Gate | | v0.7.6 | Code Hygiene + Test Coverage | | v0.7.7 | API Tokens + Extension Permissions | | v0.7.8 | Bug Fixes & Admin Gaps | | v0.7.9 | Workflow Independence Audit | | v0.7.10 | Workflow Handoff + Assignment UI | | v0.7.11 | Query & HTTP Ergonomics | | v0.7.12 | Concurrent Execution Primitive | ### v0.8.x — Storage Primitives + Composability (Kernel Complete) | Version | Title | |---------|-------| | v0.8.0 | `files` Module | | v0.8.1 | `workspace` Module | | v0.8.2 | Capability Negotiation | | v0.8.3 | Vector Column Type | | v0.8.4 | Documentation Refresh + Surface Sizing Fix | | v0.8.5 | Extension Composability | --- ## Planned ### v0.9.x — Multi-Surface Packages + Workflow Redesign **v0.9.0 — Multi-Surface Packages** *(completed)* Packages declare a `surfaces` array with per-path access controls, titles, and layouts. Unified route tree dispatches between surface rendering and ext API calls. `sw.navigate()` for client-side sub-path routing. Design doc: `docs/DESIGN-multi-surface.md`. **v0.9.1 — Server-Side Sub-Path Routing** *(completed)* Consolidated root and catch-all route handlers into a unified dispatcher. Added `aggregateAccess()` for early auth short-circuit on all-authenticated packages. SDK seeds initial history state for back-button resilience. 8 handler integration tests + 5 aggregateAccess unit tests. **v0.9.2 — Starlark Converter Consolidation + Snapshot Cleanup** *(completed)* Consolidated duplicate Go↔Starlark converters into `sandbox/convert.go` (4 exported functions) and snapshot parsers into `models/snapshot.go`. Standardized on wrapped snapshot format. ~350 lines of duplication removed. Design doc: `docs/DESIGN-workflow-redesign.md`. **v0.9.3 — Team User Roles** *(completed)* Many-to-many `team_user_roles` table. `RequireRole()` middleware. Manifest `requires_roles` field (advisory). Starlark `teams` module with `get_member_roles()` and `has_role()`. Team-admin UI with role badge chips and assignment dropdown. 10 new tests. **v0.9.4 — Package Adoption + Roles** *(completed)* `adoptable` manifest field + `team_role_catalog` table. When a team adopts an adoptable package, the package's `requires_roles` auto-populate into the team's role catalog. Adopted packages reference the original via `adopted_from` column (shared assets, no disk duplication). `AdoptTeamWorkflow` deprecated in favor of package-level adoption. 4 new endpoints, migration 017, 11 new tests. **v0.9.5 — Typed Forms → SDK Primitive** *(completed)* Extracted `TypedFormTemplate`, `FormField`, `FormFieldset`, etc. from `models/workflow.go` into a standalone `forms` package. REST endpoint `POST /api/v1/forms/validate`. Starlark `forms.validate()` module. FE SDK: `sw.forms.render()`, `sw.forms.validate()`, `sw.forms.validateRemote()`. Manifest `form_template` accepted at package level. 16 new tests. **v0.9.6 — Deprecate `stage_type`, Collapse `stage_mode`** *(completed)* `stage_type` deprecated (no longer validated, defaults to "simple"). `stage_mode` collapsed from 4→3 values: form / delegated / automated. "review" mapped to "form" on input; review surface removed (~110 lines). Migration 018. 4 package manifests updated. **v0.9.7 — Full Read/Write Workflow Starlark Module** *(completed)* `WorkflowEngine` interface extracted in sandbox package to break circular import. Four write builtins added: `workflow.start()`, `workflow.advance()`, `workflow.cancel()`, `workflow.submit_signoff()`. `instanceToDict` and `signoffToDict` helpers shared by read+write paths. 6 new tests. **v0.9.8 — Conditional Routing → SDK Primitive** *(completed)* `routing.evaluate(rules, data)` Starlark builtin — a generic decision engine reusable by any extension. 10 operators (exists, not_exists, eq, neq, gt, lt, gte, lte, in, contains), first-match-wins, returns target string or None. Always available (pure computation, no permission). 8 new tests. **v0.9.9 — Surface Access via Roles** *(completed)* `role:ROLENAME` surface access level. User must hold the role in any team (any-team semantics, no URL context needed). `evaluateAccess` promoted to Engine method for store access. `HasRoleInAnyTeam` store method queries both primary and additional roles. Admin bypass, fail- closed on nil store. 10 new tests. --- ### v0.10.x — Panels + Composable Layout Panels are a new kernel rendering tier between surfaces (full-page) and block renderers (inline). They solve composable companion views — e.g., a notes reference panel inside chat. Design doc: `docs/DESIGN-panels.md`. | Version | Title | |---------|-------| | v0.10.0 | Panel Manifest + Lifecycle | | v0.10.1 | FloatingPanel Primitive | | v0.10.2 | Docked Panels + Mode Transitions | | v0.10.3 | Panel Communication Patterns | | v0.10.4 | Reference Panel: Notes (basic) | --- ### v0.11.x — Notes Reference Extension Notes becomes the first reference extension — a production-quality knowledge base that exercises every kernel primitive. The UI/UX redesign is front-loaded as v0.11.0 so every subsequent feature version builds on a clean visual foundation. Design doc: `docs/DESIGN-notes-v011x.md`. | Version | Title | |---------|-------| | v0.11.0 | UI/UX Foundation — visual redesign | | v0.11.1 | Deep Folders + Navigation | | v0.11.2 | Wikilinks + Backlinks | | v0.11.3 | Live Preview + Rich Editing | | v0.11.4 | Note Sharing + Permissions | | v0.11.5 | Graph + Outline Hardening | | v0.11.6 | Quick Switcher + Commands | | v0.11.7 | Daily Notes + Templates | | v0.11.8 | Transclusion + Embeds | | v0.11.9 | Composability: Slots + Actions | | v0.11.10 | Panel Enhancement + Quality Gate | --- ### v0.12.x — Chat Reference Extension Chat becomes the second reference extension — human-to-human messaging built entirely on Armature's extension architecture. Where notes proved surfaces, panels, and storage, chat proves **realtime**, **cross-package composition**, and **extensible data models**. Chat is human-to-human first; AI participants arrive via `llm-bridge` (v0.13.x) extending chat through folder attributes and slot contributions. Design doc: `docs/DESIGN-chat-v012x.md`. | Version | Title | |---------|-------| | v0.12.0 | UI/UX Foundation | | v0.12.1 | Conversation Folders + Attributes | | v0.12.2 | Reactions + Threads + Pins | | v0.12.3 | Rich Compose + Attachments | | v0.12.4 | @Mentions + Notifications | | v0.12.5 | Link Previews + Message Formatting | | v0.12.6 | Conversation Themes + Personality | | v0.12.7 | Composability: Slots + Actions | | v0.12.8 | Panels + Quality Gate | --- ### v0.13.x — Reference Libraries + Extensions Custom public root surface unblocks everything — anonymous visitors, landing pages, and the package registry. `llm-bridge` is the key library delivery — extends both notes and chat through composability primitives, and introduces the **tool meta-tool** pattern: `sw.actions.list()` becomes the LLM tool registry, so installing an extension = granting AI a new capability. Zero configuration. `armature.run` deployment is the dogfood gate at the end of the series. | Version | Title | |---------|-------| | v0.13.0 | Custom Public Root Surface | | v0.13.1 | `vector-store` Library | | v0.13.2 | `llm-bridge` Core Library | | v0.13.3 | `llm-bridge` → Chat: Multi-Persona Context + Tool Meta-Tool | | v0.13.4 | `llm-bridge` → Notes: AI Toolbar Actions | | v0.13.5 | `file-share` Extension | | v0.13.6 | Package Registry Extension | | v0.13.7 | `code-workspace` Extension | | v0.13.8 | `image-gen` + `image-edit` Extensions | | v0.13.9 | Tool Meta-Tool Hardening + Scoping | | v0.13.10 | `armature.run` Deployment | | v0.13.11 | Integration Quality Gate | --- ### v0.14.x — Sidecar Tier Connect-inward model: sidecars connect TO the kernel (no K8s RBAC, no service mesh, no DNS discovery). Instance sidecars (shared infrastructure) and user sidecars (personal local tools). Design doc: `docs/DESIGN-sidecar-v014x.md`. | Version | Title | |---------|-------| | v0.14.0 | Sidecar Registry + Auth | | v0.14.1 | Capability Registration + Execution | | v0.14.2 | Kernel API Access + Event Bus | | v0.14.3 | Manifest Integration + Admin Polish | | v0.14.4 | Reference Sidecar (`armature-embed`) + Instance Gate | | v0.14.5 | User Sidecars + Reference (`user-bridge`) + User Gate | --- ### v0.15.x — Polish + Stability | Version | Title | |---------|-------| | v0.15.0 | Native Dialog Audit | | v0.15.1 | Versioned Migrations + `armature migrate` CLI | | v0.15.2 | Pre-1.0 Schema Freeze + Upgrade Path Validation | --- ### v1.0.0 — Stable Release Kernel API surface frozen. Extensions are the product. Gate criteria: - All kernel Starlark modules documented with examples - All `api_routes` covered by OpenAPI spec - Upgrade path tested from v0.8.0 → v1.0.0 - Notes reference extension shipped with all v0.11.x features - Notes UI/UX reviewed against v0.11.0 design principles — no placeholder UI - Chat reference extension shipped with all v0.12.x features - `llm-bridge` extends both notes and chat through composability - Tool meta-tool demonstrated: AI uses 3+ extension actions in a single conversation turn - Multi-persona context archetypes demonstrated - Admin safety rails validated - At least 2 panels consumed cross-package - At least 2 slot contributions per host surface demonstrated - Note and conversation sharing functional end-to-end - Headless E2E green on PG + SQLite - `armature-ca.sh` + mTLS deployment guide - Single-binary + Docker + K8s deployment paths documented - Instance sidecar (`armature-embed`) deployed and functional - User sidecar (`user-bridge`) functional on macOS/Linux/Windows - Token + mTLS sidecar auth both tested - `armature migrate` CLI functional --- ## Post-1.0 Horizon These are candidates, not commitments. Each requires a design doc. - **Federation** — cross-instance package sharing, identity federation - **Package marketplace** — signing, review, discovery registry - **Desktop app** — Tauri wrapper for local-first deployment - **Offline/sync** — SQLite-first with PG sync for field deployments - **Multi-tenant SaaS mode** — tenant isolation at the team boundary --- ## Design Principles | Principle | Implication | |-----------|-------------| | Extensions are the product | Chat, tasks, LLM, vector search, file sharing — all extensions. Zero kernel awareness of domain logic. | | Kernel stays thin | New kernel primitives require justification. If it can be an extension, it must be. | | Progressive enhancement | Every feature works on SQLite. PG adds performance. pgvector adds native vectors. S3 adds scalable storage. | | KISS-first | No unnecessary dependencies. Preact+htm (3KB), single binary, dual-DB from one codebase. | | Changeset discipline | Each CS independently CI-green. Design docs as implementation contracts. | | Pre-1.0 migration freedom | Schema changes fold into existing migrations. Post-1.0: proper versioned migrations. | --- ## Design Decisions Log | Decision | Rationale | |----------|-----------| | Tasks → extension | Three trigger primitives replace the monolithic scheduler. | | Sessions removed | Workflow instances with dedicated storage replace kernel sessions. | | `custom` stage mode | Delegates to a surface package, proving extension composability. | | Providers removed from kernel | Connections + Starlark `provider.complete` as the interface. | | Kernel permissions simplified | 6 platform permissions. Extensions define their own. | | Preact+htm retained | 3KB runtime, no build step, KISS. | | Single Docker image | Go binary + assets + migrations. | | Admin → RBAC group | Grant check replaces role check. | | Settings cascade | Scope auth + `user_overridable`. Two orthogonal axes. | | Chat as extension, not kernel | Zero kernel awareness. Proves extensibility thesis. | | PG as consensus layer | UNLOGGED node_registry + LISTEN/NOTIFY. No etcd/Consul/Redis. | | Two trigger tiers | Extension-declared (full sandbox) vs user ad-hoc (restricted). | | Two-slot topbar model | Left slot (title/branding) + center slot (`flex: 1`, tabs/pickers). | | `db` module is the structured store | No separate KV primitive. Extensions declare tables. | | `files` module rides ObjectStore | No new kernel tables. Metadata as companion objects. | | `workspace` for real filesystem | Flat blob store can't serve git/compilers/ffmpeg. Managed disk paths. | | Capability negotiation at install | Fail loud with actionable message, not silently at runtime. | | Vector column with three-tier fallback | Works everywhere, works fast with pgvector. | | Sidecar deferred to v0.14.x | HTTP module covers external APIs. Sidecars connect inward (no k8s RBAC needed). | | Workflow redesign before reference extensions | Clean up debt and promote primitives before building on top. | | Panels as kernel primitive (v0.10.x) | Z-index coordination, drag/resize, layout negotiation are kernel concerns. | | UI/UX redesign as first version in each reference series | Every feature builds on the visual foundation. Front-loading avoids double work. | | Notes as dedicated v0.11.x series (11 versions) | Complex enough to need changeset discipline. Each version independently shippable. | | Notes before chat | Notes is simpler (no realtime) and proves storage/rendering/composability. Chat adds realtime + llm-bridge story. | | Chat human-to-human first (v0.12.x) | AI is an extension concern. Keeps chat testable and usable standalone. | | Chat as dedicated v0.12.x series (9 versions) | Second reference extension. Proves realtime, extensible data models, cross-package composition. | | Folder attributes as extension bridge | Chat stores attributes it doesn't understand. llm-bridge contributes definitions. Zero coupling. | | Action registry as tool registry (meta-tool) | Dynamic, zero-config AI tool-use. Installed extensions = AI capabilities. Uniquely Armature. | | Layered prompt architecture (6 layers) | Admin safety not overridable. Clear separation: platform → space → character → context → tools → user. | | Personas in llm-bridge, not chat | Chat sees AI as just another participant_type. Persona identity is llm-bridge's concern. | | llm-bridge after notes + chat (v0.13.x) | Proves the composability hooks work without being designed for a specific consumer. | | Custom public root first in v0.13.x | Unblocks registry, landing page, and armature.run. Small kernel change, high leverage. | | Package registry as reference extension | Proves public surfaces work. Needs files module for .pkg storage. Self-referential: Armature's registry runs on Armature. | | armature.run as dogfood gate | If armature.run can't run on Armature, the platform isn't ready. Issues feed into quality gate fixes. | | Sidecar connect-inward | No K8s RBAC, no service mesh, no DNS discovery. Works on any deployment target. | | Sidecar HTTP/JSON, not gRPC | KISS. 4-endpoint contract. Every language has HTTP. | | User sidecars in v0.14.5 | Prove instance contract first, extend to users with same mechanism. | | Three-layer user sidecar RBAC | Admin controls who + what, user controls visibility. Defense in depth. | | Separate polish (v0.15.x) | Security-critical sidecar infra and quality-of-life polish shouldn't share focus. |