Harness as a Service
The agent harness, explained.
Vectrify is the hosted agent harness vertical SaaS teams embed instead of building. An agent harness is the runtime an agent needs that isn't the model: where it thinks, what it can do, what it knows, when it runs, and who it runs for. We host all of it, multi-tenant — your team writes agents and skills, not infrastructure.
What's in the harness
13 components. All hosted by us except one — the runner, which runs on your customer's machine because that's where their files and shell live.
| component | what it does for your product | where it runs |
|---|---|---|
| Agents | Configure how your product reasons, delegates, and answers | cloud |
| Skills | Package one capability once, reuse it on every tenant | cloud |
| Code sandbox | Runs agent-written Python safely — no server for you to patch | cloud |
| Runner | Executes file, shell, and git work on your customer's own machine | customer machine |
| Connectors & ingestion | Pulls and indexes your customers' data automatically | cloud |
| Workspaces | Isolates each use case's files and knowledge, per tenant | cloud |
| Automations | Fires agents on a schedule, a webhook, or an event | cloud |
| Workflows | Chains agents and skills into a multi-stage process with AI-driven routing | cloud |
| Memory | Keeps long sessions coherent without you writing compaction logic | cloud |
| Tools | Built-in registry — knowledge search, host actions, extraction, HTTP | cloud |
| Multi-tenancy & auth | Isolates every tenant, scopes every key | cloud |
| Model layer (BYOK) | Runs on your keys or your tenant's keys — never ours | cloud, your keys |
| White-label surface | Embeds the whole harness under your brand | cloud → your app |
Full detail on any of these lives on Harness →.
Your product. Our harness.
Stays yours
Your UI (unless you use ours), your data, your tenants, your customer relationships, your pricing to your customers, your product decisions.
We run
The agent loop, the code sandbox, connectors and ingestion, scheduling, tenant isolation, metering, the embed SDK, and the model layer on your keys.
Day 1 → Day 2 → Month 1
- Day 1: Connect your first data source and API key. Agents are already searching it.
- Day 2: Drop
<VectrifyEmbed />into your app, or call the REST API directly. Themed to your brand. - Month 1 (the plan, not a track record yet): Your customers using AI features inside your product — self-serve, tenant by tenant.
Two ways in
Same harness, your choice of surface. Both from Business.
White-label embed SDK
@vectrify-ai/ui, one component, ephemeral-token auth, ships this afternoon.
Your own UI on the REST API
Full parity with the embed SDK: create tenants, attach skills, run agents, start workflows, ingest documents. Build the front end you already have.
See the full breakdown on Harness →.
What we don't do
Honest, on purpose:
- No MCP server. It isn't in the product, so it isn't on this site.
- No tracing/evals product. The HelpDesk demo has a trace drawer for one run; we don't sell observability as a separate product.
- No SOC2 yet. Tenant isolation, scoped keys, and ephemeral tokens exist today; certification doesn't.
- No one-click bulk export yet. Anything you can store you can read back through the API — there is no packaged export tool.