@startsimpli/integrations 0.1.4 → 0.1.6

package/package.json

@@ -1,13 +1,15 @@
 {
   "name": "@startsimpli/integrations",
-  "version": "0.1.4",
+  "version": "0.1.6",
   "description": "Provider-agnostic types, API clients, hooks, and UI primitives for email + calendar integrations across StartSimpli apps.",
   "main": "./src/index.ts",
   "types": "./src/index.ts",
   "exports": {
     ".": "./src/index.ts",
     "./types": "./src/types/index.ts",
-    "./api": "./src/api/index.ts"
+    "./api": "./src/api/index.ts",
+    "./corpus": "./src/corpus/index.ts",
+    "./corpus-llm": "./src/corpus/llm.ts"
   },
   "files": [
     "src",
@@ -19,7 +21,8 @@
   "dependencies": {},
   "peerDependencies": {
     "react": ">=18.0.0",
-    "@tanstack/react-query": ">=5.0.0"
+    "@tanstack/react-query": ">=5.0.0",
+    "@startsimpli/llm": ">=0.4.0"
   },
   "peerDependenciesMeta": {
     "react": {
@@ -27,6 +30,9 @@
     },
     "@tanstack/react-query": {
       "optional": true
+    },
+    "@startsimpli/llm": {
+      "optional": true
     }
   },
   "devDependencies": {
@@ -41,7 +47,8 @@
     "react": "^19.0.0",
     "react-dom": "^19.0.0",
     "typescript": "^6.0.0",
-    "vitest": "^3.0.0"
+    "vitest": "^3.0.0",
+    "@startsimpli/llm": "0.4.14"
   },
   "scripts": {
     "type-check": "tsc --noEmit",

package/src/__tests__/corpus-ask.test.ts

@@ -0,0 +1,279 @@
+/**
+ * The Ask pipeline: citations, batching, and the honesty rules
+ * (bd startsim-768w.38.7).
+ *
+ * THE THING THIS SUITE PROTECTS: "an answer he cannot trace is an answer he
+ * cannot trust with a contract". So the assertions are not about the answer
+ * text — a model writes that — they are about whether the surface can ever
+ * present an untraceable answer as a traceable one. Every dishonest rendering
+ * has a test that forbids it.
+ *
+ * Scope isolation has its own file (`corpus-scope.test.ts`) because it is a
+ * different failure with a different blast radius.
+ */
+import { describe, expect, it } from 'vitest';
+
+import {
+  askCorpus,
+  buildAnswerRequest,
+  documentText,
+  renderAnswerPrompt,
+  resolveCitations,
+} from '../corpus/ask';
+import {
+  createFixtureCorpusApi,
+  createFixtureLlm,
+  FIXTURE_DOCUMENTS,
+  FIXTURE_ORG,
+} from '../corpus/fixtures';
+import { rankDocuments } from '../corpus/retrieval';
+import type { AnswerDraft, AskRequest, CorpusLlm, RetrievedDocument } from '../corpus/types';
+
+const SCOPE = { orgId: FIXTURE_ORG };
+const ask = (question: string): AskRequest => ({ question, mode: 'ask', scope: SCOPE });
+
+/** A model that returns exactly what the test tells it to. */
+function scriptedLlm(draft: AnswerDraft, calls: unknown[] = []): CorpusLlm {
+  return {
+    describe: () => ({ provider: 'scripted', model: 'scripted-1', live: false }),
+    async answer(input) {
+      calls.push(input);
+      return draft;
+    },
+  };
+}
+
+function retrievedFor(question: string): RetrievedDocument[] {
+  const own = FIXTURE_DOCUMENTS.filter((document) => document.scope.orgId === FIXTURE_ORG);
+  return rankDocuments(own, question, { limit: 5 });
+}
+
+// ---------------------------------------------------------------------------
+
+describe('batching — N documents, ONE call', () => {
+  it('sends every retrieved document in a single request', async () => {
+    const calls: unknown[] = [];
+    const corpus = createFixtureCorpusApi({ scope: SCOPE });
+    const llm = scriptedLlm({ answer: 'ok', citations: [] }, calls);
+
+    await askCorpus(ask('Acme renewal rate and seats'), { corpus, llm });
+
+    expect(calls).toHaveLength(1);
+    const input = calls[0] as { documents: unknown[] };
+    expect(input.documents.length).toBeGreaterThan(1);
+  });
+
+  it('gives the model an id per document and repeats it in the prompt', () => {
+    const retrieved = retrievedFor('Acme renewal');
+    const request = buildAnswerRequest(ask('Acme renewal'), retrieved, 4000);
+    const prompt = renderAnswerPrompt(request);
+
+    for (const document of request.documents) {
+      // The echoed id is the whole citation mechanism. Without it in the prompt
+      // there is nothing the model could return that resolves to a record.
+      expect(prompt).toContain(`<document id="${document.id}">`);
+    }
+    expect(request.documents.map((d) => d.id)).toEqual(retrieved.map((r) => r.document.id));
+  });
+
+  it('carries the SOURCE into the prompt, so the model can say where it learned it', () => {
+    const retrieved = retrievedFor('Acme renewal contract');
+    const prompt = renderAnswerPrompt(buildAnswerRequest(ask('Acme renewal contract'), retrieved, 4000));
+    expect(prompt).toMatch(/source: (Granola|Dropbox|Google Drive|Stripe)/);
+  });
+
+  it('truncates a long document rather than dropping it from the batch', () => {
+    const long = {
+      ...FIXTURE_DOCUMENTS[0],
+      sections: [{ key: 'transcript', label: 'Transcript', text: 'x'.repeat(50_000) }],
+    };
+    const text = documentText(long, 500);
+    expect(text.length).toBeLessThan(700);
+    expect(text).toContain('[truncated]');
+  });
+});
+
+describe('citations resolve to records retrieval actually supplied', () => {
+  it('resolves a good citation to the record itself', () => {
+    const retrieved = retrievedFor('Acme renewal');
+    const id = retrieved[0].document.id;
+    const { citations, unresolved } = resolveCitations(
+      { citations: [{ documentId: id, quote: 'twelve months' }] },
+      retrieved,
+    );
+    expect(unresolved).toEqual([]);
+    expect(citations).toHaveLength(1);
+    expect(citations[0].document.id).toBe(id);
+    expect(citations[0].quote).toBe('twelve months');
+  });
+
+  it('drops an invented id and REPORTS it', () => {
+    const retrieved = retrievedFor('Acme renewal');
+    const { citations, unresolved } = resolveCitations(
+      { citations: [{ documentId: 'not_this_never_existed' }] },
+      retrieved,
+    );
+    expect(citations).toHaveLength(0);
+    // Swallowing this would render a clean answer over a hallucinated source.
+    expect(unresolved).toEqual(['not_this_never_existed']);
+  });
+
+  it('deduplicates a repeated citation', () => {
+    const retrieved = retrievedFor('Acme renewal');
+    const id = retrieved[0].document.id;
+    const { citations } = resolveCitations(
+      { citations: [{ documentId: id }, { documentId: id }] },
+      retrieved,
+    );
+    expect(citations).toHaveLength(1);
+  });
+
+  it('substitutes the record’s own text when the model returns no quote', () => {
+    const retrieved = retrievedFor('Acme renewal');
+    const { citations } = resolveCitations(
+      { citations: [{ documentId: retrieved[0].document.id }] },
+      retrieved,
+    );
+    expect(citations[0].quote.length).toBeGreaterThan(0);
+  });
+
+  it('accepts the snake_case spelling a model may emit', () => {
+    const retrieved = retrievedFor('Acme renewal');
+    const id = retrieved[0].document.id;
+    const { citations } = resolveCitations(
+      { citations: [{ document_id: id } as unknown as { documentId: string }] },
+      retrieved,
+    );
+    expect(citations).toHaveLength(1);
+  });
+});
+
+describe('the surface can never present an untraceable answer as traceable', () => {
+  const corpus = () => createFixtureCorpusApi({ scope: SCOPE });
+
+  it('marks a cited answer as answered', async () => {
+    const result = await askCorpus(ask('Acme renewal rate'), {
+      corpus: corpus(),
+      llm: createFixtureLlm(),
+    });
+    expect(result.status).toBe('answered');
+    expect(result.citations.length).toBeGreaterThan(0);
+  });
+
+  it('marks an answer with NO resolvable citation as uncited, not answered', async () => {
+    const result = await askCorpus(ask('Acme renewal rate'), {
+      corpus: corpus(),
+      llm: scriptedLlm({ answer: 'You agreed four thousand a month.', citations: [] }),
+    });
+    // The text is still returned — hiding it would be its own kind of lying —
+    // but the status is what the UI keys its warning off.
+    expect(result.status).toBe('uncited');
+    expect(result.answer).toContain('four thousand');
+    expect(result.citations).toHaveLength(0);
+  });
+
+  it('marks an answer whose every citation was invented as uncited, and names them', async () => {
+    const result = await askCorpus(ask('Acme renewal rate'), {
+      corpus: corpus(),
+      llm: scriptedLlm({
+        answer: 'You agreed four thousand a month.',
+        citations: [{ documentId: 'doc_that_does_not_exist' }],
+      }),
+    });
+    expect(result.status).toBe('uncited');
+    expect(result.unresolvedCitations).toEqual(['doc_that_does_not_exist']);
+  });
+
+  it('reports no_evidence when the corpus has nothing, without calling the model', async () => {
+    const calls: unknown[] = [];
+    const result = await askCorpus(ask('zzzznothingmatchesthis'), {
+      corpus: corpus(),
+      llm: scriptedLlm({ answer: 'I could make something up.', citations: [] }, calls),
+    });
+    expect(result.status).toBe('no_evidence');
+    // Not calling the model at all is the only way to guarantee it cannot
+    // improvise an answer over an empty document list.
+    expect(calls).toHaveLength(0);
+  });
+
+  it('names the provider that actually answered, and whether it was live', async () => {
+    const result = await askCorpus(ask('Acme renewal'), {
+      corpus: corpus(),
+      llm: createFixtureLlm(),
+    });
+    expect(result.provider.live).toBe(false);
+    expect(result.provider.provider).toBe('fixture');
+  });
+
+  it('carries freshness onto the answer so a stale corpus is visible', async () => {
+    const result = await askCorpus(ask('Acme renewal'), {
+      corpus: corpus(),
+      llm: createFixtureLlm(),
+    });
+    expect(result.freshness?.some((source) => source.status !== 'ok')).toBe(true);
+  });
+});
+
+describe('retrieval spans documents and money, and stays generic', () => {
+  it('retrieves a Stripe money event alongside meeting notes', async () => {
+    const calls: unknown[] = [];
+    await askCorpus(ask('Acme Robotics charge refunded'), {
+      corpus: createFixtureCorpusApi({ scope: SCOPE }),
+      llm: scriptedLlm({ answer: 'ok', citations: [] }, calls),
+    });
+    const ids = (calls[0] as { documents: Array<{ id: string }> }).documents.map((d) => d.id);
+    // "What did we agree with this company" spans a meeting note and an invoice,
+    // and it cannot span them if they are ranked in separate universes.
+    expect(ids).toContain('ch_RedSuiteJpy0001');
+  });
+
+  it('still answers when the tenant has NO money connector at all', async () => {
+    const corpus = createFixtureCorpusApi({ scope: SCOPE });
+    delete (corpus as { listMoneyEvents?: unknown }).listMoneyEvents;
+
+    const result = await askCorpus(ask('Acme renewal rate'), { corpus, llm: createFixtureLlm() });
+    expect(result.status).toBe('answered');
+  });
+
+  it('answers a Dropbox-only question with a Dropbox citation', async () => {
+    // The standing generic test: nothing in the pipeline may be better at
+    // meeting notes than at files.
+    const result = await askCorpus(ask('master services agreement seats cap'), {
+      corpus: createFixtureCorpusApi({ scope: SCOPE }),
+      llm: createFixtureLlm(),
+    });
+    expect(result.citations.some((citation) => citation.document.source === 'dropbox')).toBe(true);
+  });
+
+  it('survives a money endpoint that fails, rather than failing the whole answer', async () => {
+    const corpus = createFixtureCorpusApi({ scope: SCOPE });
+    corpus.listMoneyEvents = async () => { throw new Error('money service is down'); };
+
+    const result = await askCorpus(ask('Acme renewal rate'), { corpus, llm: createFixtureLlm() });
+    expect(result.status).toBe('answered');
+  });
+});
+
+describe('generate mode', () => {
+  it('asks the model for a draft and still requires citations', async () => {
+    const calls: unknown[] = [];
+    const result = await askCorpus(
+      {
+        question: 'Draft a renewal letter for Acme Robotics',
+        mode: 'generate',
+        scope: SCOPE,
+        instructions: 'Formal, one page',
+      },
+      { corpus: createFixtureCorpusApi({ scope: SCOPE }), llm: scriptedLlm({
+        answer: 'Dear Acme…',
+        citations: [{ documentId: 'dbx_9812f0', quote: 'twelve (12) months' }],
+      }, calls) },
+    );
+
+    const input = calls[0] as { mode: string; instructions?: string };
+    expect(input.mode).toBe('generate');
+    expect(input.instructions).toBe('Formal, one page');
+    expect(result.status).toBe('answered');
+    expect(result.citations[0].document.source).toBe('dropbox');
+  });
+});

package/src/__tests__/corpus-copy.test.tsx

@@ -0,0 +1,196 @@
+/**
+ * What the corpus surfaces SAY, as opposed to what they fetch.
+ *
+ * Three defects the pre-demo rehearsal (bd startsim-768w.38.10) put on a
+ * screenshot, all of them the kind that survives a green test suite because no
+ * assertion ever read the sentence:
+ *
+ *   • Markdown rendered literally — `## Decisions` and `**$4,800**` as text
+ *     (bd startsim-768w.38.10, MINOR);
+ *   • `Answered by a none stub ((provider default))` — not English, and
+ *     `(provider default)` where the model's name belongs
+ *     (bd startsim-768w.38.21);
+ *   • `1 more record were searched`.
+ */
+import { describe, expect, it } from 'vitest';
+import { render, screen, within } from '@testing-library/react';
+import userEvent from '@testing-library/user-event';
+
+import { Markdown, stripMarkdown } from '../components/Markdown';
+import { AskPage } from '../components/pages/AskPage';
+import { LibraryPage } from '../components/pages/LibraryPage';
+import { configuredModel } from '../corpus/llm';
+import { createFixtureCorpusApi, FIXTURE_DOCUMENTS, FIXTURE_ORG } from '../corpus/fixtures';
+import type { AskResult, CorpusAskApi, SourceDocument } from '../corpus/types';
+
+const SCOPE = { orgId: FIXTURE_ORG };
+
+const MARKDOWN_BODY = [
+  '## Decisions',
+  '- Globex pricing agreed at **$4,800 per month**, 20 seats included.',
+  '- Additional seats are `$95` each per month.',
+  '',
+  'See the [rate card](https://docs.example/rate-card) for the rest.',
+].join('\n');
+
+function resultWith(overrides: Partial<AskResult>): AskResult {
+  const document = FIXTURE_DOCUMENTS[0];
+  return {
+    status: 'answered',
+    answer: MARKDOWN_BODY,
+    citations: [{ documentId: document.id, document, quote: '**SAML SSO has not shipped**' }],
+    considered: [{ document, score: 1, matchedTerms: ['globex'], snippet: '' }],
+    unresolvedCitations: [],
+    droppedOutOfScope: 0,
+    provider: { provider: 'openai', model: 'gpt-4o-mini', live: true },
+    ...overrides,
+  };
+}
+
+function askApi(result: AskResult): CorpusAskApi {
+  return { ask: async () => result };
+}
+
+async function ask(result: AskResult) {
+  const user = userEvent.setup();
+  render(<AskPage api={askApi(result)} scope={SCOPE} />);
+  await user.type(screen.getByRole('textbox'), 'What did we agree?');
+  await user.click(screen.getByRole('button', { name: 'Ask' }));
+  return screen.findByRole('region', { name: 'Answer' });
+}
+
+// ---------------------------------------------------------------------------
+
+describe('Markdown is rendered, not printed', () => {
+  it('turns a heading into a heading rather than two hashes', () => {
+    render(<Markdown text={MARKDOWN_BODY} />);
+    expect(screen.getByText('Decisions')).toBeInTheDocument();
+    expect(screen.queryByText(/##/)).not.toBeInTheDocument();
+  });
+
+  it('emphasises an amount instead of wrapping it in asterisks', () => {
+    render(<Markdown text={MARKDOWN_BODY} />);
+    const strong = screen.getByText('$4,800 per month');
+    expect(strong.tagName).toBe('STRONG');
+    expect(screen.queryByText(/\*\*/)).not.toBeInTheDocument();
+  });
+
+  it('makes a list a list, and a link a link', () => {
+    render(<Markdown text={MARKDOWN_BODY} />);
+    expect(screen.getAllByRole('listitem')).toHaveLength(2);
+    expect(screen.getByRole('link', { name: 'rate card' })).toHaveAttribute(
+      'href',
+      'https://docs.example/rate-card',
+    );
+  });
+
+  it('keeps the label but drops the link for a scheme it does not trust', () => {
+    render(<Markdown text={'[click me](javascript:doSomething)'} />);
+    expect(screen.getByText('click me')).toBeInTheDocument();
+    expect(screen.queryByRole('link')).not.toBeInTheDocument();
+  });
+
+  it('leaves prose containing a lone asterisk exactly as written', () => {
+    // A false positive here silently edits somebody's contract.
+    render(<Markdown text={'Fees are 2.5% * the invoice total.'} />);
+    expect(screen.getByText('Fees are 2.5% * the invoice total.')).toBeInTheDocument();
+  });
+
+  it('flattens to plain text when a string is what is needed', () => {
+    expect(stripMarkdown('## Blocker\n- **SAML SSO** has not shipped')).toBe(
+      'Blocker\nSAML SSO has not shipped',
+    );
+  });
+});
+
+describe('LibraryPage renders a document’s own Markdown', () => {
+  it('shows the summary as prose, not as source', async () => {
+    const user = userEvent.setup();
+    const document: SourceDocument = { ...FIXTURE_DOCUMENTS[0], body: MARKDOWN_BODY };
+    const api = createFixtureCorpusApi({ scope: SCOPE, documents: [document], moneyEvents: [] });
+    render(<LibraryPage api={api} scope={SCOPE} />);
+
+    await user.click(await screen.findByText(document.title));
+    const detail = await screen.findByRole('region', { name: 'Record detail' });
+    expect(within(detail).getByText('Decisions')).toBeInTheDocument();
+    expect(within(detail).getByText('$4,800 per month').tagName).toBe('STRONG');
+  });
+});
+
+describe('AskPage says who answered, in English', () => {
+  it('names the model when one answered', async () => {
+    const panel = await ask(resultWith({}));
+    expect(within(panel).getByText('Answered by openai · gpt-4o-mini')).toBeInTheDocument();
+  });
+
+  it('says no model is configured rather than "a none stub ((provider default))"', async () => {
+    const panel = await ask(
+      resultWith({ provider: { provider: 'none', model: '', live: false } }),
+    );
+    expect(within(panel).getByText(/No model is configured for this workspace/)).toBeInTheDocument();
+    // The honesty is kept: it still says nothing live ran.
+    expect(within(panel).getByText(/no live model was called/)).toBeInTheDocument();
+    expect(within(panel).queryByText(/\(\(/)).not.toBeInTheDocument();
+    expect(within(panel).queryByText(/none stub/)).not.toBeInTheDocument();
+  });
+
+  it('still calls a stub a stub', async () => {
+    const panel = await ask(
+      resultWith({ provider: { provider: 'fixture', model: 'fixture-echo', live: false } }),
+    );
+    expect(
+      within(panel).getByText('Answered by a fixture stub (fixture-echo) — no live model was called'),
+    ).toBeInTheDocument();
+  });
+
+  it('renders the answer’s Markdown and flattens the citation quote', async () => {
+    const panel = await ask(resultWith({}));
+    expect(within(panel).getByText('Decisions')).toBeInTheDocument();
+    expect(within(panel).getByText('“SAML SSO has not shipped”')).toBeInTheDocument();
+  });
+
+  it('agrees in number about how many records went uncited', async () => {
+    const document = FIXTURE_DOCUMENTS[1];
+    const panel = await ask(
+      resultWith({
+        considered: [
+          { document: FIXTURE_DOCUMENTS[0], score: 1, matchedTerms: [], snippet: '' },
+          { document, score: 0.5, matchedTerms: [], snippet: '' },
+        ],
+      }),
+    );
+    expect(within(panel).getByText('1 more record was searched but not cited')).toBeInTheDocument();
+  });
+
+  it('uses the plural when more than one was passed over', async () => {
+    const panel = await ask(
+      resultWith({
+        considered: [
+          { document: FIXTURE_DOCUMENTS[0], score: 1, matchedTerms: [], snippet: '' },
+          { document: FIXTURE_DOCUMENTS[1], score: 0.5, matchedTerms: [], snippet: '' },
+          { document: FIXTURE_DOCUMENTS[2], score: 0.4, matchedTerms: [], snippet: '' },
+        ],
+      }),
+    );
+    expect(within(panel).getByText('2 more records were searched but not cited')).toBeInTheDocument();
+  });
+});
+
+describe('the model named before the first call is the CONFIGURED one', () => {
+  it('reads the provider’s own variable', () => {
+    expect(configuredModel('openai', { OPENAI_MODEL: 'gpt-4o-mini' })).toBe('gpt-4o-mini');
+    expect(configuredModel('anthropic', { ANTHROPIC_MODEL: 'claude-x' })).toBe('claude-x');
+  });
+
+  it('prefers the specific variable over the generic one', () => {
+    expect(configuredModel('openai', { OPENAI_MODEL: 'gpt-4o-mini', LLM_MODEL: 'other' })).toBe(
+      'gpt-4o-mini',
+    );
+  });
+
+  it('falls back to LLM_MODEL, and to nothing at all', () => {
+    expect(configuredModel('openai', { LLM_MODEL: 'gpt-5' })).toBe('gpt-5');
+    expect(configuredModel('openai', {})).toBe('');
+    expect(configuredModel('none', { OPENAI_MODEL: 'gpt-4o-mini' })).toBe('');
+  });
+});