disclosure-bureau/web/lib/chat/index.ts

157 lines
4.5 KiB
TypeScript
Raw Normal View History

/**
* Chat orchestrator.
*
* Exports:
* sendChat(req) non-streaming, no tools (used by tests, fallback paths)
* streamChat(req, cb) streaming + tool calling via OpenRouter (Pattern C)
*
* CHAT_PROVIDER controls which path:
* 'openrouter' (default for Pattern C) full tools + streaming
* 'claude-code' simple Q&A via OAuth subprocess, NO tools
* 'auto' claude-code first; on rate-limit/error fall back to OpenRouter (no tools)
*/
import { claudeCodeProvider } from "./claude-code";
import { sendOnce, streamWithTools } from "./openrouter";
import { createEventStream } from "./agui";
import type { ToolHandlerContext } from "./tools";
export type Provider = "claude-code" | "openrouter";
const MODE = (process.env.CHAT_PROVIDER || "openrouter") as Provider | "auto";
/* ─── Non-streaming (legacy/fallback) ───────────────────────────────────── */
export interface SendChatReq {
system: string;
messages: Array<{ role: "user" | "assistant" | "system"; content: string }>;
maxTokens?: number;
}
export interface SendChatResp {
provider: Provider;
model: string;
content: string;
tokensIn?: number;
tokensOut?: number;
costUsd?: number;
durationMs: number;
}
export async function sendChat(req: SendChatReq): Promise<SendChatResp> {
const t0 = Date.now();
async function viaOpenRouter(): Promise<SendChatResp> {
const r = await sendOnce({
system: req.system,
messages: req.messages,
maxTokens: req.maxTokens,
});
return {
provider: "openrouter",
model: r.model,
content: r.content,
tokensIn: r.tokensIn,
tokensOut: r.tokensOut,
costUsd: 0,
durationMs: Date.now() - t0,
};
}
if (MODE === "openrouter") return viaOpenRouter();
if (MODE === "claude-code") {
if (!claudeCodeProvider.isAvailable()) {
throw new Error("claude-code mode but CLAUDE_CODE_OAUTH_TOKEN not set");
}
const r = await claudeCodeProvider.send({
system: req.system,
messages: req.messages,
maxTokens: req.maxTokens,
});
return { ...r, durationMs: Date.now() - t0 };
}
// auto
if (claudeCodeProvider.isAvailable()) {
try {
const r = await claudeCodeProvider.send({
system: req.system,
messages: req.messages,
maxTokens: req.maxTokens,
});
return { ...r, durationMs: Date.now() - t0 };
} catch (e) {
const isRate = (e as Error & { isRateLimit?: boolean }).isRateLimit;
if (isRate || /401|403|oauth|token/i.test((e as Error).message)) {
return viaOpenRouter();
}
throw e;
}
}
return viaOpenRouter();
}
/* ─── Streaming + tool calling (Pattern C) ──────────────────────────────── */
export interface StreamChatReq {
system: string;
history: Array<{ role: "user" | "assistant"; content: string }>;
userTurn: string;
ctx: ToolHandlerContext;
}
export interface StreamChatResult {
stream: ReadableStream<Uint8Array>;
/** Resolves AFTER the stream completes — usable in a deferred persist step. */
done: Promise<{
content: string;
model: string;
tokensIn: number;
tokensOut: number;
toolCalls: Array<{ name: string; args: Record<string, unknown>; result: unknown }>;
ship: synthesize 158 entities, AG-UI artifacts, chat persistence, auth flow Fase 3 onda 2 — entity synthesis at scale: - scripts/synthesize/20_entity_summary.py: queries DB for entities with total_mentions ≥ threshold + top-K verbatim chunk snippets via entity_mentions JOIN, prompts Sonnet (Holmes-Watson voice, bilingual), writes narrative_summary EN+PT-BR + summary_status=synthesized. Ran on 187 candidates (mentions ≥ 20) → 158 OK · 1 err · 29 skipped (no snippets). Combined with anchor curation: 20 curated + 158 synthesized = 178 entities with real narrative (vs 0 a day ago). Fase 4 — chat with typed artifacts + persistence: - lib/chat/agui.ts: AG-UI v1 typed Artifact union (citation, crop_image, entity_card, evidence_card, hypothesis_card, case_card, navigation_offer) alongside the existing event types. - lib/chat/tools.ts + openrouter.ts: hybrid_search emits up to 6 citation + crop_image artifacts per query. Provider collects them and returns in done.artifacts so the route can persist. - api/sessions/[id]/messages: persist artifacts to messages.citations. - components/chat-bubble.tsx: ArtifactCard renders inline cards (citation, crop_image, entity_card, navigation_offer) for streamed and persisted messages. activeId now persisted in localStorage so navigation between pages keeps the same conversation. New sessions are lazy (only when user has zero). loadMessages hydrates tools + artifacts from server. CRUD UI: rename (✎) + archive (🗑) buttons per session in the list. Home search: - doc-list-filters: input now fires hybrid_search (rerank=0 for speed) in parallel with the local title filter; chunk hits render above the doc grid with snippet + score + classification. - api/search/hybrid: accept ?rerank=0 to skip the cross-encoder (1.3s vs 60s). Auth flow: - infra: SMTP_HOST=mail.spacemail.com:587 + DMARC published; mail now lands in inbox. GOTRUE_MAILER_AUTOCONFIRM=false (real email verification). - kong.yml: proxy /auth/callback on api.disclosure.top → web:3000 so PKCE email links don't 404 at the gateway. - web/app/auth/callback: handle both ?code= (OAuth) and ?token=&type= (PKCE); redirect to the public site host before verifyOtp so the session cookie lands on the right domain. Audit deliverables: - .nirvana/outputs/disclosure-bureau/.../systems-atelier/: 5 docs (code analysis, tech debt, discovery brief, system arch, 5 ADRs) authored by sa-principal that produced this roadmap. Kept in-tree for traceability.
2026-05-18 06:52:59 +00:00
artifacts: import("./agui").Artifact[];
}>;
}
/**
* Returns immediately with a ReadableStream the caller can pipe to Response.
* The `done` promise resolves when the full conversation (including all tool
* rounds) is finished so the caller can then persist the assistant message
* to the database.
*/
export function streamChat(req: StreamChatReq): StreamChatResult {
const { stream, emit, close } = createEventStream();
const done = (async () => {
try {
const result = await streamWithTools(
{
system: req.system,
history: req.history,
userTurn: req.userTurn,
ctx: req.ctx,
},
{ emit },
);
emit({
type: "done",
provider: "openrouter",
model: result.model,
usage: {
tokens_in: result.tokensIn,
tokens_out: result.tokensOut,
tool_calls: result.toolCalls.length,
},
});
close();
return result;
} catch (e) {
emit({ type: "error", message: e instanceof Error ? e.message : String(e) });
close();
throw e;
}
})();
return { stream, done };
}