Skip to content

Feature status

This inventory reflects the source implementations reviewed on 7 October 2026. “Implemented” means executable source exists; it does not imply a packaged release. Subconscious is currently available from source, with Windows desktop as the first release target.

At a glance

Component Implemented today Boundary
Local engine Background/tray host, CLI, authenticated API, agent graphs, storage, model/tool configuration, workspace files, local retrieval Source-only; tray client launch actions and AG-UI are placeholders
Windows desktop Full adaptive workspace, chat, file, model, tool, account, and settings UI First release target; no installer yet
Subconscious Code Shared chat, IDE context, coding tools, approvals, MCP, environment profiles VS Code/Open VSX-compatible; publication is not yet verified
Flutter web and Android Main adaptive UI and remote engine connection Source target; not first-release packaging
iOS, macOS, Linux Flutter platform scaffolds Not currently release-verified
Dart TUI Connect, browse workspaces/threads/messages, refresh, and send a streaming turn Basic client; no administration or approval UI
Chromium extension Connect, workspace/thread CRUD, transcript, chat, root listing Unpacked popup; no deep browsing, editing, models, tools, or RAG controls
Cloud service Identity, OAuth/API-key connections, and durable webhooks Separate from the local engine and its data

Shared product capabilities

  • One local source of truth: desktop and coding clients share engine workspaces, threads, messages, models, and tools.
  • Streaming agent chat: model selection, thread defaults, cancellation, tool approval, active/tagged files, and bounded attachments in the full client.
  • Workspace control: multiple absolute roots, scoped file operations, per-operation approvals, per-root retrieval, and built-in/configured tool policies.
  • Local retrieval: changed-file indexing, keyword/vector/hybrid search, line-aware chunks, deletion pruning, daily refresh, and explicit reindexing.
  • Flexible models: Amazon Bedrock plus OpenAI and OpenAI-compatible endpoints for OpenRouter, Ollama, LM Studio, vLLM, and local services.
  • Local security: random bearer credentials outside development mode, loopback-first binding, encrypted credential storage, sensitive-file content exclusion, and restricted origins.
  • Multi-client UI: adaptive Flutter layouts, native Windows integration, web/mobile connection paths, a basic terminal UI, Chromium popup, and an IDE-native coding surface.

Known limitations

Do not treat source targets as released products

There are currently no packaged installation directions. Windows desktop is first; other Flutter platform folders indicate build targets, not equal release support.

  • subconscious desktop, subconscious tui, subconscious web, and subconscious extension are reserved CLI stubs. Launch each source client directly.
  • POST /api/v1/agui is a placeholder and should not be used as an integration contract.
  • The Flutter model form displays more provider labels than the engine accepts directly. Use the providers listed in Local engine.
  • The engine provides HTTP and WebSocket transport, not TLS termination. Use LAN mode only on a trusted network and protect the joining token.
  • Semantic embedding may need to acquire its pinned local model on first use; offline operation from a completely empty cache is not guaranteed.
  • The Flutter Skills settings destination is currently a placeholder.

What comes next

Packaging and verified release channels remain the major gap: a Windows desktop installer, a published VS Code/Open VSX extension, connected CLI/tray launch actions, and explicit verification for additional platforms. This page will move capabilities from source target to released only when distribution and verification exist.