Dataembed

Dataembed is a pnpm/Turborepo monorepo whose deployable unit is one Next.js app — the eve agent runs inside it, so there is no separate agent service to operate. You need Node.js ≥ 24, pnpm, and a Postgres database.

Setup

git clone https://github.com/kyh/dataembed.git
cd dataembed
pnpm install

cp .env.example .env   # fill in the required variables below

pnpm db:push           # create tables (Drizzle)
pnpm dev               # app + eve agent on one dev server

Environment reference

Configuration lives in .env at the repo root.

Required

Variable	Purpose
`POSTGRES_URL`	Postgres connection string (any Postgres works)
`BETTER_AUTH_SECRET`	Secret for better-auth session signing (also salts anonymous IP hashing)
`AI_GATEWAY_API_KEY`	Vercel AI Gateway key — powers the agent, verification, and the default search-grounded provider

Search providers (optional)

Each key activates its provider on every query; ranked lists are merged with reciprocal rank fusion. See Search providers.

Variable	Provider
`EXA_API_KEY`	Exa — semantic search, page text inline
`PARALLEL_API_KEY`	Parallel — agent-oriented search, extended excerpts
`FIRECRAWL_API_KEY`	Firecrawl — keyword SERP, pages scraped to markdown

Model overrides (optional)

Gateway model ids; defaults shown. Details in Search providers.

Variable	Default
`SEARCH_AGENT_MODEL`	`anthropic/claude-sonnet-5`
`SEARCH_VERIFICATION_MODEL`	`openai/gpt-5.4-mini`
`SEARCH_QUERY_MODEL`	`perplexity/sonar`

Caps (optional)

Defaults shown; see the caps table for how they apply.

Variable	Default
`SEARCH_MAX_QUERIES`	`6`
`SEARCH_MAX_RESULTS`	`25`
`SEARCH_ANON_MAX_QUERIES`	`3`
`SEARCH_ANON_MAX_RESULTS`	`10`
`SEARCH_ANON_SESSIONS_PER_HOUR`	`3`

Auth and email (optional)

Email/password auth works out of the box at /auth/login, /auth/register, and /auth/password-reset.

Variable	Purpose
`GITHUB_CLIENT_ID`, `GITHUB_CLIENT_SECRET`	Enable GitHub OAuth — the sign-in button appears only when both are set
`RESEND_API_KEY`	Send password-reset emails via Resend; without it, the reset URL is logged to the server console (fine for dev)
`EMAIL_FROM`	From address for outgoing email
`BETTER_AUTH_URL`	Public origin of your deployment, for auth callbacks on non-Vercel hosts

Eve (optional, non-Vercel only)

Variable	Purpose
`EVE_CHANNEL_SECRET`	Shared secret authenticating server → eve calls. Not needed on Vercel (OIDC) or in local dev
`EVE_ORIGIN`	Origin serving eve's HTTP channel, when it differs from the app's own origin

Deployment

Vercel is the primary target: deploy the repo as a single project and the eve agent ships alongside the Next.js app with zero extra configuration — server-to-agent calls authenticate via Vercel OIDC, and runs are durable via Vercel Workflow.

Elsewhere, the same app runs on any Node host with Postgres. Set EVE_CHANNEL_SECRET so the server can authenticate to the eve channel (local pnpm dev doesn't need it), and EVE_ORIGIN only if you serve /eve/* from a different origin than the app.

Database commands

pnpm db:push          # push Drizzle schema to POSTGRES_URL
pnpm db:push-remote   # push schema to prod (reads .env.production.local)
pnpm db:studio        # browse data in Drizzle Studio

Self-hosting