summary system backend implementation

This commit is contained in:
Storme-bit
2026-04-19 06:50:24 -07:00
parent 15c1bec609
commit 2769f436fa
10 changed files with 210 additions and 106 deletions

View File

@@ -98,14 +98,15 @@ src/
│ ├── MessageBubble.jsx # Individual message bubble — renders markdown via react-markdown
│ ├── InfoPanel.jsx # Right panel — model selector and session metadata (slide-in)
│ ├── SessionModal.jsx # Modal for session rename, project assignment, delete
│ ├── ProjectModal.jsx # Modal for project create, edit, delete
│ ├── ProjectModal.jsx # Modal for project create/edit — name, description, colour,
│ │ # system prompt override; delete confirmation
│ ├── AllChatsView.jsx # Paginated session list with project indicator column
│ ├── AllProjectsView.jsx # Project tile grid with create/edit/delete; tile click navigates to ProjectView
│ ├── ProjectView.jsx # Individual project — conversations, new chat input, memory
│ │ # placeholder, user notes, ⋮ edit/delete menu
│ ├── MemoryView.jsx # Paginated, searchable, expandable, deletable episode viewer
│ └── SettingsView.jsx # Settings — Memory limits, Models (inference params, active
│ # model, context window), Service Health, Appearance placeholder
│ └── SettingsView.jsx # Settings — Memory, Models, Behaviour (system prompt),
│ # About, Appearance
├── index.css # Global reset, CSS variables, utility classes
└── main.jsx # React entry point
```
@@ -151,7 +152,7 @@ in the `ChatWindow` header.
| View | Component | Trigger |
|---|---|---|
| `'home'` | `HomeView` | Initial load; going back from chat with no history |
| `'home'` | `HomeView` | Initial load |
| `'chat'` | `ChatWindow` | Selecting a session; new chat; sending from HomeView |
| `'all-chats'` | `AllChatsView` | "All Chats →" or ☰ icon in collapsed rail |
| `'all-projects'` | `AllProjectsView` | "View Projects" button or ⊞ icon |
@@ -178,10 +179,9 @@ leaving `'home'` expands it.
- Centred textarea input — sending creates a new session and navigates to chat
- Quick action pills that populate the input without auto-sending
Sending from HomeView uses `handleHomeSend` in `App.jsx`, which calls
`createSession()` (returns the new session object), then immediately calls
`sendMessage` with the session passed directly as a parameter — avoiding the
React state settling race condition that would cause the message to be dropped.
`handleHomeSend` in `App.jsx` calls `createSession()` (which returns the new
session object), then immediately calls `sendMessage` with the session passed
directly — avoiding the React state settling race condition.
## CSS Architecture
@@ -283,6 +283,7 @@ mode, and delete confirmation in `confirm-delete` mode.
- `useContextMenu` dismisses on a `window` click listener
- Dynamic `updateSession` SQL builds `SET` clause from only the fields passed — prevents accidental overwrites
- `AllChatsView` pagination uses `CLIENT_DEFAULTS.PAGE_SIZE` (not `API_DEFAULTS.PAGE_SIZE` which doesn't exist)
- `Sidebar` groups sessions by project — `key` must be passed directly to `<SessionRow key={...}>`, not included in the props spread object
## Sidebar — Session Grouping
@@ -302,11 +303,27 @@ The isolated toggle has been removed from `ProjectModal`.
`useProjects` fetches the project list from `GET /projects` on mount and
exposes `refreshProjects` for keeping the sidebar in sync after mutations.
`ProjectModal` handles create, edit, and delete confirmation. Fields: name
(required), description (optional), colour picker.
### ProjectModal Fields
Clicking a project tile in `AllProjectsView` calls `onSelectProject` then
navigates to `'project'`.
- **Name** (required)
- **Description** (optional)
- **Colour** — picker from six preset hex values
- **System Prompt** (optional) — overrides the global system prompt for all
conversations in this project. Leave blank to use the global default.
Stored as `system_prompt` (snake_case) matching the SQLite column.
`Enter` key does not submit — textarea fields make it ambiguous. Save button only.
`handleSave` in `ProjectView` destructures `system_prompt` (snake_case) to
match what `ProjectModal` sends. `updateProject` in `orchestration.js` uses
a passthrough pattern — spreads all fields into the request body.
### System Prompt Hierarchy
System prompt resolution in `chat/index.js` (orchestration):
1. `project.system_prompt` — if set on the project (highest priority)
2. `settings.systemPrompt` — global setting from `settings.json`
3. `ORCHESTRATION.SYSTEM_PROMPT` — hardcoded constant in `@nexusai/shared` (last resort)
### ProjectView
@@ -322,9 +339,8 @@ navigates to `'project'`.
saved value (`savedNotes` state tracks the baseline, not `initialNotes`)
`updateProject` in `orchestration.js` uses a passthrough pattern — spreads
all fields directly into the request body, only transforming `isolated` if
present. This allows partial updates like `{ notes }` without clobbering
other fields.
all fields directly into the request body. This allows partial updates like
`{ notes }` or `{ system_prompt }` without clobbering other fields.
For memory isolation behaviour, see `memory-isolation.md`.
@@ -337,9 +353,9 @@ buttons during in-flight requests.
`SettingsView` receives `settings`/`saveSetting`/`saving` from a single
`useSettings()` call at the top level and passes them as props to
`ModelsSection` and `ModelsFolderSetting` — avoiding triple fetch on mount.
`modelProps` (context window, loaded model) is fetched once in `App.jsx` and
passed down as a prop, eliminating a duplicate fetch on every settings open.
`ModelsSection`, `ModelsFolderSetting`, and `SystemPromptSetting` — avoiding
triple fetch on mount. `modelProps` (context window, loaded model) is fetched
once in `App.jsx` and passed down as a prop.
`SettingsView` is organised into sections:
@@ -347,6 +363,9 @@ passed down as a prop, eliminating a duplicate fetch on every settings open.
- **Models** — models folder path, temperature, repeat penalty, Top-P, Top-K,
active model dropdown, read-only model info panel (file, size, context window,
loaded model from llama-server)
- **Behaviour** — global system prompt textarea (`SystemPromptSetting`). Save
button appears only when content differs from `savedPrompt` state. Saving an
empty string sends `null` which reverts to the hardcoded default.
- **About** — service health check panel, version
- **Appearance** — theme (coming soon)

View File

@@ -58,7 +58,7 @@ Six core tables:
- **entities** — named things the system learns about (people, places, concepts)
- **relationships** — directional labeled links between entities
- **summaries** — condensed episode groups for efficient context retrieval
- **projects** — named groupings of sessions with `name`, `description`, `colour`, `icon`, `isolated`, `notes`
- **projects** — named groupings of sessions with `name`, `description`, `colour`, `icon`, `isolated`, `notes`, `system_prompt`
### Migrations
@@ -71,6 +71,7 @@ try { db.exec(`ALTER TABLE sessions ADD COLUMN project_id INTEGER REFERENCES pro
try { db.exec(`CREATE INDEX IF NOT EXISTS idx_sessions_project ON sessions(project_id)`); } catch {}
try { db.exec(`ALTER TABLE projects ADD COLUMN isolated INTEGER NOT NULL DEFAULT 0`); } catch {}
try { db.exec(`ALTER TABLE projects ADD COLUMN notes TEXT`); } catch {}
try { db.exec(`ALTER TABLE projects ADD COLUMN system_prompt TEXT`); } catch {}
```
New migrations are always appended here — never modify the schema file for
@@ -92,29 +93,15 @@ keep the FTS index automatically in sync with the episodes table.
Both `updateSession` and `updateProject` build their `SET` clause dynamically
from only the fields passed — prevents partial updates from overwriting fields
that weren't touched:
that weren't touched.
`updateProject` allowlist:
```js
// updateProject example
function updateProject(id, fields = {}) {
const allowed = ['name', 'description', 'colour', 'icon', 'isolated', 'notes'];
const updates = [];
const values = [];
for (const key of allowed) {
if (fields[key] !== undefined) {
updates.push(`${key} = ?`);
values.push(fields[key] ?? null);
}
}
if (updates.length === 0) return getProject(id);
values.push(id);
db.prepare(`UPDATE projects SET ${updates.join(', ')} WHERE id = ?`).run(...values);
return getProject(id);
}
const allowed = ['name', 'description', 'colour', 'icon', 'isolated', 'notes', 'system_prompt'];
```
This means saving just `{ notes: "..." }` won't touch `name`, `colour`, or
any other field.
This means saving just `{ notes: "..." }` or `{ system_prompt: "..." }` won't
touch any other field.
## Qdrant / Semantic Layer
@@ -183,8 +170,10 @@ After extraction, each entity is:
Qdrant collection with `{ name, type, notes, projectId }` as payload —
`projectId` scopes entities to their project for isolated retrieval
The Qdrant payload stores enough information to reconstruct entity context
at retrieval time without a SQLite roundtrip.
`extractAndStoreEntities` receives `projectId` from `createEpisode`, which
receives it from the episode route, which receives it from orchestration's
`createEpisode` call. This ensures entities are tagged with the correct
project scope at extraction time.
## Project Delete Behaviour

View File

@@ -48,7 +48,7 @@ src/
├── routes/
│ ├── chat.js # POST /chat and POST /chat/stream
│ ├── sessions.js # Session CRUD proxy
│ ├── projects.js # Project CRUD proxy
│ ├── projects.js # Project CRUD proxy — passes req.body straight through
│ ├── episodes.js # Episode list and delete proxy
│ ├── settings.js # GET /settings and PATCH /settings
│ ├── health.js # GET /health — pings all four services
@@ -75,6 +75,7 @@ via `appSettings.load()` — changes apply immediately without a service restart
| `repeatPenalty` | 1.1 | Repeat token penalty |
| `topP` | 0.9 | Nucleus sampling probability mass |
| `topK` | 40 | Top-K token candidates per step |
| `systemPrompt` | *(ORCHESTRATION.SYSTEM_PROMPT)* | Global system prompt. `null` reverts to hardcoded constant. |
Defaults are defined in `config/settings.js` and fall back to constants in
`@nexusai/shared`. Values saved in `settings.json` take precedence.
@@ -91,41 +92,43 @@ difference is how the inference response is delivered to the client.
step needed.
2. **Project context resolution** — if the session has a `project_id`, fetch
the project and all its session IDs. Used to scope semantic search. See
`memory-isolation.md` for full behaviour.
the project and all its session IDs. Used to scope semantic search. The
project's `system_prompt` is also read at this step if set.
3. **Recent episode retrieval** — fetch the most recent episodes for the
3. **System prompt resolution** — three-tier hierarchy:
- `project.system_prompt` — if the session is in a project and it's set (highest priority)
- `settings.systemPrompt` — global setting from `settings.json`
- `ORCHESTRATION.SYSTEM_PROMPT` — hardcoded constant in `@nexusai/shared` (last resort)
4. **Recent episode retrieval** — fetch the most recent episodes for the
session (`recentEpisodeLimit`, default 5).
4. **Semantic search** — embed the user message, query Qdrant for the top
5. **Semantic search** — embed the user message, query Qdrant for the top
most similar past episodes (`semanticLimit`, `scoreThreshold`). Deduplicated
against recent episodes. Non-critical — if it fails, pipeline continues with
recency-only context.
5. **Entity search**reuse the embedded user message vector to query the
`entities` Qdrant collection (score threshold 0.6, limit 5). Returns
entity payloads (`name`, `type`, `notes`) directly — no SQLite roundtrip
needed. Non-critical — if it fails, pipeline continues without entity context.
6. **Entity search**query the `entities` Qdrant collection filtered by
`projectId`. Non-project sessions receive no entity context. Non-critical.
6. **Prompt assembly** — combine system prompt, entity context, semantic
episodes, recent episodes, and user message.
7. **Prompt assembly** — combine resolved system prompt, entity context,
semantic episodes, recent episodes, and user message.
7. **Inference** — send to inference service with settings-derived parameters
8. **Inference** — send to inference service with settings-derived parameters
(temperature, topP, topK, repeatPenalty). `/chat` awaits full response;
`/chat/stream` pipes SSE chunks to the client.
8. **Episode write** — write the exchange back to memory. Fire-and-forget
for `/chat`; awaited for `/chat/stream` to ensure the full text is
accumulated before saving.
9. **Episode write** — write the exchange back to memory with `projectId`.
Fire-and-forget for `/chat`; awaited for `/chat/stream`.
9. **Auto-naming** — on `isFirstMessage && !session.name`, fire a secondary
inference call with a naming prompt (max 20 tokens, temperature 0.3) and
write the result back as `session.name`. Fully fire-and-forget.
10. **Auto-naming** — on `isFirstMessage && !session.name`, fire a secondary
inference call with a naming prompt (max 20 tokens, temperature 0.3) and
write the result back as `session.name`. Fully fire-and-forget.
### Prompt Structure
```
[System prompt]
[Resolved system prompt]
Here is what you know about entities relevant to this conversation:
- {name} ({type}): {notes}
@@ -175,9 +178,9 @@ is terminated by `res.end()` after the done event.
folder for richer metadata (label, description). Returns file size in GB.
`GET /models/props` fetches directly from llama-server via `LLAMA_SERVER_URL`.
Returns `{ contextWindow, modelAlias }`. Used by the client to display
read-only context window size and the currently loaded model in the settings
panel. Returns `503` if llama-server is unreachable.
Returns `{ contextWindow, modelAlias }`. `n_ctx` is at
`data.default_generation_settings.n_ctx` in the llama-server response.
Returns `503` if llama-server is unreachable.
## Sessions Route Behaviour