New Bring your own OpenAI, Anthropic or Gemini key

Ship AI agents to production. On your own keys.

Build agents and workflows, run them against real systems, and watch every step, cost and failure live. Your model bill goes straight to your provider at cost — we charge a flat fee for the platform, never a cent on tokens.

Talk to us Read the API docs

0% token markup
No card required
Keys encrypted per workspace

AI Platform / Observe

Runs

Your OpenAI key · sk-…4f2a Live · 4 in flight

Runs today: 1,284; +12%
Success rate: 99.2%; +0.4 pts
p95 latency: 1.2s; −180 ms
Token spend · your key: $48.20; $0 markup

Workflow	Run	State	Duration	Started
Support triage	wf_8fa2c41	Running	1.2s	2 min ago
Lead scoring	wf_8fa1b07	Queued	—	3 min ago
Invoice parse	wf_8f9ed22	Succeeded	2.0s	11 min ago
Contract review	wf_8f9e118	Failed	0.4s	24 min ago
Nightly digest	wf_8f9d904	Canceled	—	1 hr ago

Illustrative interface

OpenAPI 3.1: 105 typed operations at /openapi.json
Tenant isolated: Every query scoped to a workspace
Cost budgets: Hard stops before overspend
Audit trail: Every action, who and when

Bring the models you already pay for

OpenAI
Anthropic
Google Gemini
Mistral
Groq
OpenRouter
Azure OpenAI
Ollama

How it works

From a prompt on a whiteboard to a run you can account for.

Most AI work stalls between the demo and production. Vendor.sh gives each step an owner, a state and a cost, so the work survives contact with real systems.

Build 01

Define agents with instructions, tools and a structured output schema. Chain them on the workflow canvas.

Agents · Tools · Workflow canvas
Run 02

Trigger from the console, a schedule, an embed or one HTTPS call. Every run gets a state the whole team can read.

Queued → Running → Succeeded
Observe 03

Trace each step, its tokens and its latency. Budgets stop spend before the invoice does.

Traces · Costs · Budgets

The platform

Everything between the model and the outcome.

One workspace for the people who design the work and the people who answer for it.

Webhook Trigger

Received

Classify Agent

0.8s · $0.002

Draft reply Agent

Running…

Workflow canvas

Compose triggers, agents, tools and branches visually. Each node keeps its own state, so a failure points at a step, not a mystery.

output_schema:
  type: object
  required: [intent, urgency]
  properties:
    intent:
      enum: [refund, bug, sales]
    urgency:
      type: integer
    note:
      type: [string, "null"]

Structured output

Agents return typed data, not paragraphs. Schemas are made strict for the provider automatically.

run 2.41s

classify 0.52s

search_kb 0.88s

draft_reply 0.94s

guardrail 0.07s

Step-level traces

Every span carries tokens, latency and the exact input it saw. Debug the step that failed, not the whole run.

Support workspace: $148 / $200
Research members: $391 / $400
Nightly batch: $62 / $250

Cost budgets

Set a ceiling per workspace or per member. Runs are refused at the limit and say why, before finance has to ask.

Embeds and API access

Drop a deployed workflow into any site, or call it over HTTPS with a scoped token. Visitors stay anonymous.

The screens on this band are illustrative. Every run id, duration and figure is made up for the example.

Bring your own key

Pay for the platform. Never for the tokens.

Connect the provider accounts you already have. The provider bills those tokens to you at its own price, and a call on your key creates no managed debit here. Our plans meter runs and members — the work we actually do for you.

Model keys — illustrative Example data

Provider	Key	Used by	State
OpenAI	sk-proj-…4f2a	Support triage, Invoice parser	Verified
Anthropic	sk-ant-…9c01	Contract review, Lead scoring	Verified
Google Gemini	AIza…Qm7x	Nightly digest	Verified
OpenRouter	sk-or-…d33e	Fallback route	Disabled

Encrypted at rest, never logged, and never sent to a browser. A screen shows the provider and a masked suffix, nothing more.

Pick the provider per agent

Provider and model are the agent's own

Run a support agent on Anthropic and a parser on OpenAI, each against your own account. Change the model without rebuilding the workflow around it.

Every call in Cost & usage

One record per call

Spend is recorded per call, per provider and per agent, and the route that paid for it is on the row — so you can read the bill before your provider sends it.

Budgets cover either route

Per workspace, or per member

Set a daily or monthly ceiling. A run is refused at the limit instead of surprising finance, whichever route served it.

Keys stay on the server

Encrypted, masked, write-only

A provider key is encrypted at rest and never returned to a browser. After you save it, no screen in the product can show it again.

For developers

One HTTPS call away from any workflow.

Every deployed workflow gets a versioned endpoint, a typed contract and a run you can poll or stream. No SDK lock-in: if it speaks HTTP, it can run your agents.

Scoped tokens per workspace, revocable in one click
OpenAPI 3.1 contract at /openapi.json
Server-sent events for every step as it runs
Idempotency keys, so a retry never runs twice

Explore the API reference

$ curl -X POST https://vendor.sh/api/v1/workflows/support-triage/runs \
    -H "Authorization: Bearer $VENDOR_TOKEN" \
    -H "Idempotency-Key: 7d1f…" \
    -d '{"input": {"ticket": "Refund for #4471"}}'

$run = Http::withToken(config('services.vendor.key'))
    ->withHeaders(['Idempotency-Key' => $ticket->uuid])
    ->post('https://vendor.sh/api/v1/workflows/support-triage/runs', [
        'input' => ['ticket' => $ticket->subject],
    ])->throw()->json();

run = httpx.post(
    "https://vendor.sh/api/v1/workflows/support-triage/runs",
    headers={"Authorization": f"Bearer {token}",
             "Idempotency-Key": ticket.uuid},
    json={"input": {"ticket": ticket.subject}},
).raise_for_status().json()

const run = await fetch(
  "https://vendor.sh/api/v1/workflows/support-triage/runs",
  { method: "POST",
    headers: { Authorization: `Bearer ${token}`,
               "Idempotency-Key": ticket.uuid },
    body: JSON.stringify({ input: { ticket: ticket.subject } }) },
).then((r) => r.json());

Response

{
  "data": {
    "id": "01J9Z4K2QXV7M8T3B6R0YHWCE5",
    "status": "pending",
    "mode": "async",
    "deployment": { "slug": "support-triage", "environment": "production" },
    "workflow":   { "slug": "support-triage", "name": "Support triage" },
    "release":    { "number": 3 },
    "usage": { "cost": "0.0000", "tokens": 0, "duration_ms": null }
  }
}

202 Accepted

Security

We wrote down who we defend against. Then we tested it.

Our threat model names each adversary, the control that answers it and the test that proves the control fires. It also lists what is not covered, because an unlisted gap reads as a solved one.

Read the threat model

The open internet

Rate limits, CSRF, and embed sessions that carry a hashed token.

Covered by a test

Another tenant

Every query is scoped to a workspace. Keys never cross one.

Covered by a test

A member acting badly

Permissions are checked on the server for every action, not in the browser.

Covered by a test

What the model returns

Output is validated against a schema before anything acts on it.

Covered by a test

The supply chain

Pinned packages, audited installs, and no source maps shipped.

Covered by a test

Database access

Settings and provider keys are encrypted at rest and never logged.

Covered by a test

Pricing

Flat platform pricing. Tokens at cost, on your key.

Plans meter runs and members — the work we actually do for you. Every call on your own provider key goes on that provider's invoice, never ours.

Free

USD 0 per month

Build and test a small AI product.

200 workflow executions each month
2 active workflows
1 Project
1 seat
7 days of run retention
1 concurrent run
1 deployment environment
Bring your own provider key
Vendor.sh managed credits
Typed workflow API
Signed webhooks

Talk to us

Builder

USD 19 per month

Deploy AI workflows for one product.

5,000 workflow executions each month
10 active workflows
5 Projects
1 seat
30 days of run retention
3 concurrent runs
1 deployment environment
Bring your own provider key
Vendor.sh managed credits
Typed workflow API
Signed webhooks
Schedules
MCP client

Talk to us

Recommended

Pro

USD 49 per month

Operate production workflows across environments.

25,000 workflow executions each month
50 active workflows
Unlimited Projects
5 seats
90 days of run retention
10 concurrent runs
Development, Staging, and Production environments
Bring your own provider key
Vendor.sh managed credits
Typed workflow API
Signed webhooks
Schedules
MCP client
MCP server
Workflow evaluations
Advanced run logs

Talk to us

Team

USD 149 per month

Run governed AI workflows with a larger team.

100,000 workflow executions each month
Unlimited active workflows
Unlimited Projects
15 seats
180 days of run retention
25 concurrent runs
Development, Staging, and Production environments
Bring your own provider key
Vendor.sh managed credits
Typed workflow API
Signed webhooks
Schedules
MCP client
MCP server
Workflow evaluations
Advanced run logs
Advanced workspace roles
Audit logs
Shared connections
Approval policies

Talk to us

Enterprise

Custom

Use custom limits, security, and support terms.

Unlimited workflow executions each month
Unlimited active workflows
Unlimited Projects
Unlimited seats
Unlimited days of run retention
Unlimited concurrent runs
-1 deployment environment
Bring your own provider key
Vendor.sh managed credits
Typed workflow API
Signed webhooks
Schedules
MCP client
MCP server
Workflow evaluations
Advanced run logs
Advanced workspace roles
Audit logs
Shared connections
Approval policies

Talk to us

Calls on your own key cost nothing here. On every plan.

Your provider bills those tokens at its own price. Credit packs exist only for managed usage you choose to run through Vendor.sh.

Compare every plan

Put your first workflow in the control room today.

Free to start. Connect your own model key, invite your team and ship a run in under ten minutes — with zero markup on a single token.

Talk to us Talk to us

No card required · Cancel anytime