Skip to content

SYSTEM Cited by 2 sources

Vercel Knowledge Agent Template

Knowledge Agent Template (vercel.com/templates) is Vercel's open-source, filesystem-based knowledge- agent architecture for deployment on Vercel. The template ships a production-ready composition of:

  • a snapshot repository populated by Vercel Workflow from Postgres-stored source configuration;
  • a per-request Vercel Sandbox that loads the snapshot and exposes bash / bash_batch tools to the agent;
  • a Chat SDK-based adapter layer fronting Slack, Discord, Microsoft Teams, Google Chat, GitHub, and web chat with one shared agent pipeline;
  • a complexity router dispatching simple vs hard questions to fast/cheap vs slow/powerful models over Vercel AI Gateway;
  • an admin interface with usage stats, error logs, user management, source config, and sync controls;
  • an AI-powered admin agent with internal tools (query_stats, query_errors, run_sql, chart) so operators can "debug your agent with an agent".

(Source: sources/2026-04-21-vercel-build-knowledge-agents-without-embeddings)

Why it shows up on this wiki

Canonical wiki instantiation of agent retrieval without embeddings. The template replaces the vector- DB / chunking / embedding-model stack with a filesystem plus bash tools, on the architectural claim that LLMs already understand filesystems from code training, and that filesystem retrieval is traceable while vector retrieval is opaque.

Canonical composition of five distinct wiki patterns in one ready-to-deploy artifact:

Pipeline (verbatim)

From the post:

"(1) You add sources through the admin interface, and they're stored in Postgres. (2) Content syncs to a snapshot repository via Vercel Workflow. (3) When the agent needs to search, a Vercel Sandbox loads the snapshot. (4) The agent's bash and bash_batch tools execute file-system commands. (5) The agent returns an answer with optional references."

Five layers; each layer is a distinct wiki substrate.

Source shapes supported

"Plug any source: GitHub repos, YouTube transcripts, documents (e.g., markdown files), or custom APIs." The template is source-shape-agnostic — the common pipe is the snapshot repository, not any particular ingestion mechanism.

Distribution shapes supported

"Ship it as a web chat app, a GitHub bot, a Discord bot, or all three at once." Template-shipped adapters: GitHub + Discord. Chat-SDK-supported adapters also named: Slack, Microsoft Teams, Google Chat. Plus the public adapter directory for official + community adapters.

Consumer SDK

@savoir/sdk — package exposing tools that any AI- SDK-powered agent can import and call:

const savoir = createSavoir({
  apiUrl: process.env.SAVOIR_API_URL!,
  apiKey: process.env.SAVOIR_API_KEY,
})
const { text } = await generateText({
  // ...
  prompt: 'How do I configure authentication?',
})

Post explicitly notes this is a rename-before-ship placeholder: "If you plan to extend the SDK's capabilities and publish it, customize the package name from @savoir/sdk to your own."

Operational datum

~\$1.00 → ~\$0.25 per call on Vercel's internal sales-call summarisation agent after replacing the vector pipeline with this template's filesystem + bash retrieval shape. "The output quality improved." The 4× cost reduction + quality improvement is the load-bearing production evidence for the architectural pitch.

Seen in

Last updated · 476 distilled / 1,218 read