Server Functions

Typed RPC between the browser and the Worker.

All server logic lives in *.functions.ts files using createServerFn from @tanstack/react-start.

createAndRunBrew

File: src/lib/brews.functions.ts Creates a brew row, seeds five stage rows, runs the full pipeline sequentially, and returns the new brewId.

const run = useServerFn(createAndRunBrew);
const { brewId } = await run({
  data: {
    title: "Edge AI in 2026",
    brew_type: "research",
    goal: "Briefing for product team",
    context: "...notes...",
    tone: "calm, precise",
    output_format: "report",
  },
});

refineBrew

Takes the current final_output plus user feedback and produces a refined output. Appends a row to brew_outputs and updates brews.final_output.

await refine({ data: { brewId, feedback: "Tighter intro, more examples." } });

rerunBrew

Deletes existing brew_steps, reseeds them, and runs the pipeline from scratch using the brew’s original inputs.

archiveAsRecipe

Snapshots the brew’s inputs into a new recipes row.

await archive({
  data: { brewId, name: "Weekly competitive scan", description: "..." },
});

Cellar functions

File: src/lib/cellar.functions.ts

• listCellar() — return the user’s entries grouped by kind.
• addCellarEntry({ kind, content, weight? }) — manual add.
• updateCellarEntry({ id, content?, kind?, weight? }) — edit in place.
• deleteCellarEntry({ id }) — forget an entry.
• getCellarSettings() / updateCellarSettings({ enabled?, max_entries?, manual_pack? }).

The Cellar is read automatically by brews.functions.ts on every stage — you don’t need to wire it manually.

Ingest functions

File: src/lib/ingest.functions.ts

scrapeUrl({ url })

Scrapes any public URL via Firecrawl, returns { title, sourceUrl, markdown }. Large pages are truncated to 20k characters.

const scraped = await scrape({ data: { url: "https://example.com/post" } });

searchKeyword({ query, sources, limit, afterReddit?, pageHN? })

Searches Reddit and/or Hacker News for conversations around a keyword.

Param	Type	Default	Description
query	string	required	Keyword or phrase
sources	("reddit" | "hn")[]	["reddit", "hn"]	Which sources to query
limit	number	8	Results per source (max 25)
afterReddit	string?	—	Reddit pagination cursor
pageHN	number	0	HN page number via Algolia

Returns { query, count, markdown, afterReddit, hasMoreHN, nextPageHN }. Reddit uses cursor-based pagination; HN uses numbered pages. The markdown block includes titles, scores, comment counts, and post snippets grouped by source.

const res = await search({
  data: { query: "agentic workflow", sources: ["reddit", "hn"], limit: 10 },
});
// res.afterReddit → cursor for next Reddit batch
// res.hasMoreHN → true if more HN pages exist

Share functions

File: src/lib/shares.functions.ts

• getMyShare({ brewId }) — returns the existing slug, if any.
• createShare({ brewId }) — mints a new nanoid slug (idempotent if not revoked).
• revokeShare({ brewId }) — sets revoked_at; the public route returns 404 afterwards.

Anonymous reads go through the server route below.

Wallet auth functions

File: src/lib/wallet-auth.functions.ts

• getWalletChallenge({ publicKey }) — issues a one-time nonce to sign.
• verifyWalletSignature({ publicKey, signature, nonce }) — verifies the Ed25519 signature and returns a Supabase session keyed to the wallet’s auth.users row.

Public API route

File: src/routes/api/public/brew.$slug.ts

GET /api/public/brew/:slug returns a safe DTO for the shared brew (no user_id, no context, no internal step payloads beyond stage names + statuses). Backed by supabaseAdmin; protected by slug entropy and revoked_at.

Authoring rules

• Use .middleware([requireSupabaseAuth]) for any function that touches user data.
• Validate input with Zod via .inputValidator().
• Read secrets like TEAGENTIC_AI_KEY inside .handler(), never at module scope.
• Don’t import server-only modules from client code (the *.server.ts suffix blocks this).