OpenAI LiveConnectedBring your own key
01 Inference
Every model,one governed door.
Connect puts one API in front of 50+ model providers, routes every request by your rules, fails over when a provider is down and stops spending at the caps you set.
[ 01 ] InferenceRoute. Fail over. Cap spend.
01 In this layer
Models and the gateway in front of them.
- Routing rules per workspaceAvailable
- Failover across providersAvailable
- Usage caps and audit eventsAvailable
- Open-weight models on your infrastructureDesigned for
03 SEE IT WORK
See Connect at work.
Universal keys
A key per person. Every provider they're allowed.
Issue keys from the Access page. By default a key reaches every provider your workspace has connected; narrow it when you need to. Secrets are shown once, rotate without downtime, and disappear when the person leaves.
02 [ 01.A ]
A change of model is a setting, not a project.
One OpenAI-compatible API in front of 50+ model providers. Switch, route and fall back without rewriting the application.
Available
03 [ 01.B ]
When a provider fails, the work moves on.
Fallback chains move a request to the next model when a provider is down or rate-limited.
Available
04 [ 01.C ]
Open-weight models on infrastructure you control.
Run open and private models in your own environment behind the same gateway. Self-deploy is on request.
Designed for
[ 01.D – 01.O ] INSIDE CONNECT
Inside Connect, screen by screen.
Twelve chapters, each a real screen from the console, assembling as you scroll.
D1 [ 01.D ]Universal keys
A key per person. Every provider they're allowed.
Issue keys from the Access page. By default a key reaches every provider your workspace has connected; narrow it when you need to. Secrets are shown once, rotate without downtime, and disappear when the person leaves.
Available- 01Issue
- 02Narrow
- 03Reveal once
- 04Rotate and revoke
D2 [ 01.E ]Providers and gateway
Bring your own keys. Or don't.
Connect providers with your own credentials, or run on Swfte credits. One base URL speaks the OpenAI, Anthropic and Responses protocols, and only routes you've allowed are exposed. Your own keys are never billed twice.
Available- 01Connect a provider
- 02One endpoint, three protocols
- 03Who pays, on every request
- 04Only what you allow
D3 [ 01.F ]Models and fleet
Text, image, speech, video. Hosted or yours.
Filter the catalogue by what a model can do, deploy open models on your own GPU pools, and wire in MCP servers. Everything your workspace runs sits in one fleet view.
Available- 01Every modality
- 02Your own GPUs
- 03Tools via MCP
- 04One fleet
D4 [ 01.G ]Routing
Decide where every request goes.
Draw the path on a canvas: conditions on spend, tokens, error rate or time of day, and an ordered chain of models for each branch. Then test a request against your draft and see exactly which rule fires, before anything is spent.
Available- 01Request stream
- 02Conditions
- 03Model chains
- 04Test request
D5 [ 01.H ]Controls
Caps that hold. Labels that don't lie.
Cap spend and tokens for the workspace, a model, a person or a single key. Block, downgrade to a cheaper model, or just alert. Every setting says whether the gateway enforces it. If rules can't be read, credit-funded requests stop before a provider call.
Available- 01Set a cap
- 02Hit the line
- 03Say what's enforced
- 04Fail closed
D6 [ 01.I ]People and seats
Licensed seats. A console for every role.
Team and Enterprise workspaces share licensed seats. Invite people with a role, watch the pool fill, add seats when needed. Owners get the whole console; members get a simple read-only view of their own keys, usage and limits.
Available- 01Invite
- 02Seat pool
- 03Roles
- 04My access
D7 [ 01.J ]Analytics and reports
Slice usage by anything you call a team.
Define your own dimensions (Team, Department, Cost centre, Client) and put people in groups. Then pivot spend and tokens by person, group, key, provider, model or tool, drill in, compare periods, and export the exact view as CSV, Excel or an offline report.
Available- 01Your dimensions
- 02Pivot anything
- 03Drill in
- 04Export and schedule
D8 [ 01.K ]Evaluation
Try models side by side before routing.
Send the same prompt, image or audio to two models and compare the answers, cost and latency. Save the cases that matter to you as test suites and rerun them whenever you change a rule or a model.
Available- 01Compare
- 02Any input
- 03Suites
- 04Rerun
D9 [ 01.L ]Coding agents and SDKs
Point your tools at one URL.
Copy a setup snippet for Claude Code, Codex or your SDK, swap in your Swfte key, and requests go through your rules. Connect tags requests by tool, so you can see usage by tool, by person and by team.
- 01Paste
- 02Run
- 03Which tool works where
- 04See it by tool
D10 [ 01.M ]Built to be trusted
Safe defaults, honest screens.
Only the routes you allow are exposed. Secrets are shown once. When something can't be loaded, Connect says so instead of showing a number it made up, and it stops you acting on the wrong workspace.
- 01Deny by default
- 02Honest when it breaks
- 03Right workspace, every time
- 04Provider health
D11 [ 01.N ]Across industries
Same console. Your rules.
Whatever you build, the building blocks are the same: keys per person, caps per group, routes per workload, and reports in the shape your business already uses.
D12 [ 01.O ]Why Connect
Everything teams need to use AI with confidence.
One gateway, one console, the controls finance and security ask for, and a view people actually understand.
- 01Eight capabilities, one view
- 02A connected system
- 03Explore each piece
Frequently asked questions
범용 API는 어떻게 작동하나요?
Connect는 OpenAI 형식과 호환되는 단일 API 엔드포인트를 제공합니다. 단일 매개변수를 변경하여 50개 이상의 모델 간에 전환할 수 있습니다.
스마트 라우팅은 어떻게 비용을 절감하나요?
Connect는 품질 요구 사항을 충족하는 가장 비용 효율적인 모델로 요청을 자동으로 라우팅합니다. 평균적으로 고객은 API 비용을 40% 절감합니다.
내 API 키를 사용할 수 있나요?
네! 모든 제공업체의 키를 가져오거나 간편한 청구를 위해 Swfte의 풀링된 액세스를 사용할 수 있습니다.
Keep exploring
[ 06 ] One platform
Capability and control, in one platform.
Start with one layer and prove it. Add the next when you are ready. The controls come with you.