mirror of
https://github.com/Memo-2023/mana-monorepo.git
synced 2026-05-15 04:41:09 +02:00
refactor(brain): migrate Companion engine to LlmOrchestrator (4-tier system)
Replaces the ad-hoc local-first + server-fallback pattern with the
shared LlmOrchestrator, giving the Companion Chat full access to
the 4-tier system (none/browser/mana-server/cloud) and its privacy
+ user-preference enforcement.
New companionChatTask (lib/llm-tasks/companion-chat.ts):
- name: 'companion.chat'
- minTier: 'browser' (no rules fallback — needs an LLM)
- contentClass: 'personal' (allows server/cloud if user opted in;
NOT 'sensitive' because the chat isn't restricted to browser-only,
but the user can set it per-task via taskOverrides)
- requires: { streaming: true }
Engine changes:
- callLlm() now delegates to llmOrchestrator.run(companionChatTask, ...)
- Still preloads the local model when browser tier is available so
the UI can show download progress
- isCompanionAvailable() now asks llmOrchestrator.canRun() which
considers user settings + backend readiness + consent gates
User benefits:
- Tier-selector in the PillNav now applies to Companion Chat
- Users can force cloud/server/browser per-task via settings overrides
- Cloud tier only runs when cloudConsentGiven is set
- Privacy: content marked 'sensitive' in other tasks (Journal etc.)
is still restricted to browser/rules — Companion respects the
same orchestrator so privacy invariants hold consistently
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Till JS
parent
4b2007e97c
commit
e885713fd0
2 changed files with 90 additions and 64 deletions
53
apps/mana/apps/web/src/lib/llm-tasks/companion-chat.ts
Normal file
53
apps/mana/apps/web/src/lib/llm-tasks/companion-chat.ts
Normal file
|
|
@ -0,0 +1,53 @@
|
|||
/**
|
||||
* companionChatTask — LLM task definition for the Companion Brain chat.
|
||||
*
|
||||
* Routes through the shared LlmOrchestrator so the user's tier settings
|
||||
* and privacy rules are respected:
|
||||
*
|
||||
* - minTier: browser (needs at least local Gemma; no rules fallback)
|
||||
* - contentClass: 'personal' — user messages may reference their data
|
||||
* but aren't the most sensitive class (which is reserved for things
|
||||
* like Journal, Dreams, Finance). 'personal' allows mana-server or
|
||||
* cloud if the user opted in; 'sensitive' would restrict to browser.
|
||||
* - streaming: true — the chat UI relies on per-token updates
|
||||
*
|
||||
* Individual callers can override the tier via settings.taskOverrides
|
||||
* (e.g. force cloud for Companion even if the default is browser).
|
||||
*/
|
||||
|
||||
import type { LlmBackend, LlmTask, GenerateResult } from '@mana/shared-llm';
|
||||
|
||||
export interface CompanionChatInput {
|
||||
messages: { role: 'user' | 'assistant' | 'system'; content: string }[];
|
||||
temperature?: number;
|
||||
maxTokens?: number;
|
||||
onToken?: (token: string) => void;
|
||||
}
|
||||
|
||||
export type CompanionChatOutput = {
|
||||
content: string;
|
||||
usage?: GenerateResult['usage'];
|
||||
};
|
||||
|
||||
export const companionChatTask: LlmTask<CompanionChatInput, CompanionChatOutput> = {
|
||||
name: 'companion.chat',
|
||||
minTier: 'browser',
|
||||
contentClass: 'personal',
|
||||
requires: { streaming: true },
|
||||
displayLabel: 'Companion Chat',
|
||||
|
||||
async runLlm(input: CompanionChatInput, backend: LlmBackend): Promise<CompanionChatOutput> {
|
||||
const result = await backend.generate({
|
||||
taskName: 'companion.chat',
|
||||
contentClass: 'personal',
|
||||
messages: input.messages,
|
||||
temperature: input.temperature ?? 0.7,
|
||||
maxTokens: input.maxTokens ?? 1024,
|
||||
onToken: input.onToken,
|
||||
});
|
||||
return {
|
||||
content: result.content,
|
||||
usage: result.usage,
|
||||
};
|
||||
},
|
||||
};
|
||||
|
|
@ -1,15 +1,21 @@
|
|||
/**
|
||||
* Companion Chat Engine — Orchestrates LLM + Context Document + Tool Calling.
|
||||
*
|
||||
* Tries local LLM (Gemma via @mana/local-llm) first. If WebGPU is not
|
||||
* available, falls back to the mana-llm server endpoint. Tool calling
|
||||
* uses JSON extraction from the LLM output.
|
||||
* Routes through the shared LlmOrchestrator (4-tier system). The orchestrator
|
||||
* picks browser/mana-server/cloud based on user settings + the task's
|
||||
* contentClass ('personal'). Users can override per-task via their LLM
|
||||
* settings (e.g. "Companion always via cloud" or "never leave device").
|
||||
*
|
||||
* Tool calling is simulated via JSON extraction since none of the tiers
|
||||
* natively speak function calling (Gemma doesn't, Gemini via our proxy
|
||||
* routes through text-completion).
|
||||
*/
|
||||
|
||||
import { generate, getLocalLlmStatus, loadLocalLlm, isLocalLlmSupported } from '@mana/local-llm';
|
||||
import { llmOrchestrator } from '@mana/shared-llm';
|
||||
import { isLocalLlmSupported, getLocalLlmStatus, loadLocalLlm } from '@mana/local-llm';
|
||||
import { companionChatTask } from '$lib/llm-tasks/companion-chat';
|
||||
import { generateContextDocument } from '$lib/data/projections/context-document';
|
||||
import { getToolsForLlm, executeTool } from '$lib/data/tools';
|
||||
import { authStore } from '$lib/stores/auth.svelte';
|
||||
import type { DaySnapshot, StreakInfo } from '$lib/data/projections/types';
|
||||
import { emitDomainEvent } from '$lib/data/events';
|
||||
import { getTool } from '$lib/data/tools/registry';
|
||||
|
|
@ -20,59 +26,37 @@ const MAX_TOOL_ROUNDS = 3;
|
|||
|
||||
type LlmMessage = { role: 'user' | 'assistant' | 'system'; content: string };
|
||||
|
||||
/** Try local LLM, fall back to server if WebGPU unavailable. */
|
||||
/**
|
||||
* Route an LLM call through the orchestrator. The orchestrator handles
|
||||
* tier selection, privacy enforcement, and fallbacks. If the browser
|
||||
* tier is chosen but the local model hasn't loaded yet, we trigger
|
||||
* the download first so the UI can show progress.
|
||||
*/
|
||||
async function callLlm(messages: LlmMessage[], onToken?: (token: string) => void): Promise<string> {
|
||||
// Try local first (WebGPU + Gemma)
|
||||
// If browser tier is available, preload the model so the
|
||||
// CompanionChat UI can show download progress before generation starts.
|
||||
if (isLocalLlmSupported()) {
|
||||
const status = getLocalLlmStatus();
|
||||
if (status.current.state !== 'ready') {
|
||||
try {
|
||||
await loadLocalLlm();
|
||||
} catch {
|
||||
// Fall through to server
|
||||
return callServerLlm(messages);
|
||||
}
|
||||
if (status.current.state === 'idle' || status.current.state === 'checking') {
|
||||
// Fire-and-forget — the orchestrator will await isReady() anyway
|
||||
void loadLocalLlm().catch(() => {
|
||||
/* fall through to next tier */
|
||||
});
|
||||
}
|
||||
const result = await generate({ messages, temperature: 0.7, maxTokens: 1024, onToken });
|
||||
return result.content;
|
||||
}
|
||||
|
||||
// Fallback: server-side LLM via mana-api
|
||||
return callServerLlm(messages);
|
||||
}
|
||||
|
||||
async function callServerLlm(messages: LlmMessage[]): Promise<string> {
|
||||
const apiUrl =
|
||||
(typeof window !== 'undefined' &&
|
||||
(window as unknown as Record<string, string>).__PUBLIC_MANA_API_URL__) ||
|
||||
import.meta.env.PUBLIC_MANA_API_URL ||
|
||||
'';
|
||||
|
||||
if (!apiUrl) {
|
||||
return 'LLM nicht verfuegbar — weder WebGPU noch Server-Endpoint konfiguriert.';
|
||||
}
|
||||
|
||||
const headers: Record<string, string> = { 'Content-Type': 'application/json' };
|
||||
try {
|
||||
const token = await authStore.getValidToken();
|
||||
if (token) headers['Authorization'] = `Bearer ${token}`;
|
||||
} catch {
|
||||
// Continue without auth — server will decide
|
||||
const result = await llmOrchestrator.run(companionChatTask, {
|
||||
messages,
|
||||
onToken,
|
||||
temperature: 0.7,
|
||||
maxTokens: 1024,
|
||||
});
|
||||
return result.value.content;
|
||||
} catch (err) {
|
||||
const msg = err instanceof Error ? err.message : String(err);
|
||||
return `LLM nicht verfuegbar: ${msg}`;
|
||||
}
|
||||
|
||||
const response = await fetch(`${apiUrl}/api/v1/chat/completions`, {
|
||||
method: 'POST',
|
||||
headers,
|
||||
body: JSON.stringify({ messages, model: 'companion' }),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
const err = await response.text().catch(() => response.statusText);
|
||||
return `Server-Fehler: ${err}`;
|
||||
}
|
||||
|
||||
const data = (await response.json()) as { choices?: { message?: { content?: string } }[] };
|
||||
return data.choices?.[0]?.message?.content ?? 'Keine Antwort vom Server.';
|
||||
}
|
||||
|
||||
interface EngineResult {
|
||||
|
|
@ -272,20 +256,9 @@ export async function runCompanionChat(
|
|||
}
|
||||
|
||||
/**
|
||||
* Check if the Companion Chat is available.
|
||||
* Returns true if either local LLM or server endpoint is usable.
|
||||
* Check if the Companion Chat is available — delegates to the orchestrator
|
||||
* which considers the user's enabled tiers and backend readiness.
|
||||
*/
|
||||
export function isCompanionAvailable(): boolean {
|
||||
// Local LLM available?
|
||||
if (isLocalLlmSupported()) {
|
||||
const status = getLocalLlmStatus();
|
||||
if (status.current.state === 'ready' || status.current.state === 'idle') return true;
|
||||
}
|
||||
// Server fallback configured?
|
||||
const apiUrl =
|
||||
(typeof window !== 'undefined' &&
|
||||
(window as unknown as Record<string, string>).__PUBLIC_MANA_API_URL__) ||
|
||||
import.meta.env.PUBLIC_MANA_API_URL ||
|
||||
'';
|
||||
return !!apiUrl;
|
||||
return llmOrchestrator.canRun(companionChatTask);
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue