Skip to content
WORKSI AI Control
Sign inStart trialStart free trial

One loop for every AI call.

Connect once. Every model call, tool call and agent run is observed, checked, explained when it fails, controlled when it must be, and measured against what it costs.

Start free trialRead the quickstart
The gateway sits on the request path. Everything else happens beside it. Your apps send requests to the Worksi gateway. The gateway calls your provider with your key. The gateway sends spans to the Worksi control plane asynchronously, off the request path. The SDKs, MCP and OpenTelemetry send spans to the Worksi control plane. The control plane sends alerts to your team.

The management loop

Six stages, one product.

Each stage feeds the next. The loop closes when what you measure changes what you control.

  1. 01

    Connect

    Swap the base URL, add an SDK, connect over MCP or send OpenTelemetry.

    base_url="https://api.openai.com/v1"base_url="https://gateway.worksi.ai/v1"
    Integrations
  2. 02

    Observe

    Live activity, traces, runs and cost per call, plus an inventory of every agent and model.

    trace
    2bef9e35…6da1
    spans
    9 · 1 LLM · 4 tool
    duration
    1.5 s
    Observability
  3. 03

    Detect

    Eleven detectors open deduplicated incidents for loops, outages, spikes and stuck runs.

    detector
    tool_failure
    signal
    lookup_order failed 3× in 10m
    incident
    INC‑5 · critical
    Incidents
  4. 04

    Understand

    Evidence packs, the failure location and a diagnosis that cites its evidence.

    root cause
    orders-db returns 503
    confidence
    78% · high
    cites
    ev-1 ev-2 ev-3 ev-6
    Diagnosis
  5. 05

    Control

    Policies, approvals, end‑user quotas, pause and stop, enforced at the gateway.

    HTTP/1.1 403 Forbidden"code": "model_not_allowed"
    Governance
  6. 06

    Measure

    Spend by agent, team and model, outcomes and ROI in management language.

    AI spend (24 h)
    $7.16
    cost per outcome
    $0.07
    value multiple
    166×
    ROI & cost

Readouts show sample data from the built‑in demo organization.

In the path of every call, never in the way.

The gateway authenticates, governs and forwards each request to your own provider. Recording and analysis happen beside the request path, so telemetry can never fail or slow a call.

Gateway on the hot path
Checks the key, your plan, policies, end‑user quotas and pause or stop, then forwards to your provider with timeouts, retries and your fallback models. It adds 0.82 ms at p95, measured — how we measured gateway overhead.
Telemetry, asynchronous
Each call becomes a span after the response is sent. If the telemetry pipeline is unavailable, the gateway buffers spans and your requests continue.
Your key, your provider
Any of 24 provider presets, from Anthropic and OpenAI to DeepSeek, Qwen and a local Ollama, or any OpenAI‑compatible server, with your own key encrypted at rest. Your provider bills you for tokens, and Worksi never marks them up.
Tenant isolation
Every query is scoped to your organization. The organization, workspace and environment of each call come from its API key, never from the client.

No feature walls around the core loop.

Plans scale volume, retention, team size and a few advanced governance features. Everything here is included from Starter up.

  • OpenAI‑compatible gateway plus Anthropic’s Messages API: chat, Responses, embeddings, streaming
  • Bring your own key for 24 provider presets or any OpenAI‑compatible server, with no token mark‑up
  • Built‑in simulator provider for tests and demos
  • Python and TypeScript SDKs, OpenTelemetry ingestion and the Worksi MCP server
  • Live activity, traces with waterfall, runs and dashboards
  • Eleven incident detectors, evidence packs and deterministic diagnosis
  • Pause agents, stop runs, runbooks and recovery actions
  • Capture modes and redaction of sensitive values
  • Role-based access control and a tamper-evident audit log
Compare plans

Point one client at the gateway.

Change the base URL and the key. Everything on this page starts from that one request.

Start free trialRead the quickstart
app.py
import osfrom openai import OpenAI client = OpenAI(    base_url="https://gateway.worksi.ai/v1",    api_key=os.environ["AIW_API_KEY"],)