The agent platform for vertical SaaS

Not a search API.
The agent harness for vertical SaaS teams.

Agents. Skills. Workflows. Sessions. Scheduled tasks. Injectable prompts. A white-label embed SDK. Everything your vertical SaaS product needs to ship AI features—already built.

Start free trialSee pricing

Embed It

Ship inside your product

Two ways in, same harness. Drop in our white-label UI and you can typically ship in an afternoon, or build your own UI on the REST API — every agent, workflow, session, tenant, and document is an endpoint. Either way it ships as part of your app, themed to your brand, authenticated to your users.

REST API — everything, not just embed

Create tenants, attach skills, run agents, start workflows, schedule tasks, ingest documents, read results — over REST. Build your own front-end or wire agents into your backend with no UI at all.

  • Scoped vapi_ keys (read / write / admin)
  • Ephemeral tokens for the browser
  • Webhooks and WebSocket for results

Embed SDK

@vectrify-ai/ui drops the harness into your product as one component: <VectrifyEmbed /> for React, a vanilla JS loader for everything else.

One npm install, themed to your brand.

  • npm install @vectrify-ai/ui
  • Ephemeral token auth — no long-lived keys in the browser
  • Context object carries injectable prompts

Full theming API

Override every CSS variable at the account, tenant, or runtime level. Light and dark mode, per-tenant themes.

Every CSS variable, per tenant.

  • CSS variable overrides
  • Light + dark mode
  • Per-tenant theme overrides
  • Runtime override via embed context

Injectable prompts

Versioned prompt templates with @mention expansion — user name, account tier, date, feature flags — passed in as values from your own code.

  • Passed via API or embed SDK context
  • 50 versions retained per prompt

Scoped API keys

Issue read, write, or admin keys per integration. Shown once, then stored only as a SHA-256 hash.

Make It Act

Agents that act, not chat

Configure an agent once — it answers, executes, delegates, and proposes actions your app can commit, without you writing orchestration code.

Configurable agents

Set reasoning depth, tone, context window, and persona per agent.

  • Subagents spawn, delegate, and aggregate results
  • Departments group agents

Propose / commit actions

Agents don't just answer — they propose an action and your app commits it. The agent decides what to do; your code decides what actually happens.

Skills

Packaged, versioned units of capability. Write one, attach it to any agent across any tenant; update once, every agent picks it up.

  • Custom skills: SKILL.md manifest + reference docs + runnable scripts, Python runtime
  • First-party library: extraction, summarization, structured output, web search, CRM lookups

Workflows

Chain agents, skills, and data operations into projects with named stages. An LLM routes between stages; every step streams to your UI over WebSocket.

Keep It Running

Work while you sleep

Sessions and scheduled tasks give agents a place to live between requests — persistent, monitored, and driven on a clock, not just a chat turn.

Sessions

The managed execution context for an agent: tool calls, retries, context management, result persistence.

  • Per-tenant run controls and limits
  • Streamed to WebSocket or webhook
  • Full execution log per run

Scheduled tasks

Point a cron expression at any agent and prompt. Vectrify fires it on schedule, stores the result, calls your webhook when done.

  • Per-tenant schedules, each customer's own cadence
  • Webhook on completion
  • Pause, skip, or run now

Local runner

A Go daemon installed on the customer's own machine — Windows, Linux, or macOS — for work that has to happen where the data lives: file, shell, and git operations.

  • Outbound-only WebSocket — no inbound ports, no firewall changes
  • File operations confined to a configured workspace root; paths outside it are rejected
  • Shell access is opt-in and off by default
  • Runs unprivileged on Linux and macOS, as the installing user, not root

Isolate & Meter Per Tenant

Built for multi-tenant resale

Every tenant is isolated by default and metered on tenants and sessions — never seats, never per run — so you can build your own pricing on top.

Workspaces

Each workspace is an isolated knowledge namespace, isolated per tenant. Connect multiple sources; assign one or many workspaces to any agent.

Connectors & ingestion

Pull data from where your customers already keep it. Parsing, chunking, embedding, and indexing run automatically, with live status.

Live today

  • Confluence (OAuth)
  • SharePoint (OAuth)
  • Website crawl
  • Direct upload — PDF, DOCX, HTML, MD

Coming soon

  • Google DriveComing soon
  • DropboxComing soon
  • NotionComing soon
  • SlackComing soon
  • GitHubComing soon
  • JiraComing soon

Don't see yours? We build connectors on request, usually about a day.

  • Hybrid search — vector + BM25 — per workspace
  • Auto re-ingestion via polling sync; crawls respect robots.txt
  • Tenant-isolated vector search — every query is scoped to the tenant

Per-tenant token and cost caps

Set a token or cost limit per tenant; requests are gated (HTTP 402) once it's hit. Build your own tier structure on top of Vectrify's.

BYOK per tenant

Bring your own model keys, for your account or for each of your tenants individually. Your customers can run on their own model accounts.

One-call provisioning

Provision a new, fully isolated tenant with a single API call. No infrastructure work on your end.

About six months of engineering. Available today.

14-day free trial. All features. No credit card required.

Start free trialView pricing