Shittim Chest — Chat, Panels and Integrations

Shittim Chest is the user face: chat, panels and multi-channel integrations — the layer that turns model and agent capabilities into something you use every day. It takes models and memory from Arona, and agents, panels and industrial workflows from entelecheia's scepter. This guide is organized by capability: chat, panels, channels, search and auth.

Architecture at a glance

You
 │  desktop app · web UI · 12+ IM channels (discord/slack/matrix/lark/…)
 ▼
Shittim Chest core (node-2:8425)
 ├─ Chat: conversations persisted in Postgres, streamed SSE/WS
 ├─ Panels: data grid / media pipeline / 3D twin (panel engine)
 ├─ Media pipeline: LLM vision review, image generation, 3D gen (stub)
 ├─ Channel bridge: bidirectional IM ↔ core (webhooks + WS)
 └─ RPC bridge → scepter (agents, semantic search, knowledge base)
        │
        ▼
Arona (models, memory) · scepter (agents, panels, search)

1. Chat

• Conversations and messages persist in Postgres (conversations / `messages` tables). The full history is assembled server-side.
• Streaming via SSE with chunk replay; the desktop shell shows thinking chunks and tool calls from scepter agents.
• chat.send supports conversation id, model override and memory flags (see the [Arona guide](./arona.md) for the memory gateway semantics).

2. Panels

Panels are created from a single prompt: the engine generates the layout and widgets, then persists them to scepter's workspace store (.amphoreus/workspace.toml + .noa/views/*.view.toml). Editing is structured, not black-box: the raw data-source binding, widget list and connection state are visible, with a per-row deep-edit dialog and a collapsed natural-language tuning box.

Three panel kinds:

• Data grid — table views with typed fields, sort/filter/group, writes go through `table.*` edit kinds with real validation and rollback.
• Media pipeline — a Dify-style node graph (LLM nodes, image generation, HTTP, knowledge retrieval, branching); pipelines run server-side with streaming progress and can be invoked as tools by agents.
• 3D twin — device model trees with world coordinates, scene configuration and camera bookmarks.

3. Multi-channel integrations

The channel bridge connects core to IM platforms (Discord, Slack, Matrix, Lark, …). Inbound messages become chat turns; outbound responses stream back over the same channel. Each channel requires its OAuth application approved for production use.

4. Semantic search and knowledge

• search.semantic bridges to scepter's vector search (ApoRia workspace index + PhiLia long-term memories fused into one ranked list).
• Knowledge bases (create / add documents / subscribe) persist in Postgres and are searchable through the same RPC surface.
• Agent reports are indexed automatically, so past reports are semantically retrievable.

5. Auth

Authentication is delegated: scepter trusts the caller user id supplied by the chest gateway (X-User-Id / user_id), which chest obtains from Arona or the invitation flow. RBAC roles (admin / operator / viewer / agent) gate write operations on panels and workspaces.

Env reference (subset)