AI Gateway: OpenAI's format, open models, and your own providers · 9603c76 · flagon-io/g1t

359	359		&[
360	360		Op::ListIntegrations,
361	361		Op::ConnectIntegration,
	362	+	Op::UpdateIntegration,
362	363		Op::DisconnectIntegration,
363	364		Op::TestIntegration,
364	365		Op::GetModelRoutes,
⋯
461	462		Op::ListEvents => "List repository events",
462	463		Op::ListIntegrations => "List integrations",
463	464		Op::ConnectIntegration => "Connect an integration",
	465	+	Op::UpdateIntegration => "Update an integration",
464	466		Op::DisconnectIntegration => "Disconnect an integration",
465	467		Op::TestIntegration => "Test an integration",
466	468		Op::GetContext => "Look up a ticket",

169	169		ListEvents,
170	170		ListIntegrations,
171	171		ConnectIntegration,
	172	+	UpdateIntegration,
172	173		DisconnectIntegration,
173	174		TestIntegration,
174	175		GetContext,
⋯
629	630		}
630	631
631	632		impl Op {
632		−	pub const ALL: [Op; 219] = [
	633	+	pub const ALL: [Op; 220] = [
633	634		Op::Whoami,
634	635		Op::GetWorkspace,
635	636		Op::CreateWorkspace,
⋯
711	712		Op::ListEvents,
712	713		Op::ListIntegrations,
713	714		Op::ConnectIntegration,
	715	+	Op::UpdateIntegration,
714	716		Op::DisconnectIntegration,
715	717		Op::TestIntegration,
716	718		Op::GetContext,
⋯
939	941		Op::ListEvents => "list_events",
940	942		Op::ListIntegrations => "list_integrations",
941	943		Op::ConnectIntegration => "connect_integration",
	944	+	Op::UpdateIntegration => "update_integration",
942	945		Op::DisconnectIntegration => "disconnect_integration",
943	946		Op::TestIntegration => "test_integration",
944	947		Op::GetContext => "get_context",
⋯
1267	1270		"A workspace's integrations: its own model provider, the alert sources that open issues (Sentry, Datadog, webhooks), and the trackers whose tickets agents can read (Jira, Linear). Secrets are never returned. Members only."
1268	1271		}
1269	1272		Op::ConnectIntegration => {
1270		−	"Connect a workspace to an outside system. provider is a model provider (anthropic, openai, gemini, xai, mistral, deepseek, azure_openai, openrouter, groq, together, fireworks, cerebras, anthropic_endpoint or openai_endpoint: your own key, billed by that provider, and free on g1t while it is being built out; a workspace can connect several and route each kind of work with set_model_routes), or sentry, datadog, webhook, jira or linear. config holds the settings each needs; secret is the API key or token. For datadog and webhook, g1t makes the signing secret and returns it once. Owners only."
	1273	+	"Connect a workspace to an outside system. provider is a model provider (anthropic, openai, gemini, xai, mistral, deepseek, azure_openai, openrouter, groq, together, fireworks, cerebras, anthropic_endpoint or openai_endpoint: your own key, billed by that provider, and free on g1t while it is being built out; a workspace can connect several and route each kind of work with set_model_routes), or sentry, datadog, webhook, jira or linear. config holds the settings each needs; secret is the API key or token, kept encrypted and never returned (secret_hint shows its last four characters). For a model provider, config.gateway_models chooses which AI Gateway requests go to it by the model they name: ids such as gpt-5.5, or prefixes ending in * such as gpt-* or ollama/* (a /* prefix is taken off before sending); absent, an Anthropic key or Anthropic-compatible endpoint takes claude-* and the others take nothing. Requests on the workspace's own provider are counted and never charged. For datadog and webhook, g1t makes the signing secret and returns it once. Owners only."
	1274	+	}
	1275	+	Op::UpdateIntegration => {
	1276	+	"Change an integration: its name, its config (replaced whole when given) or its secret (a new key replaces the old one, write-only). Use it to rotate a model provider's key or to choose its config.gateway_models, the AI Gateway models it takes. Fields left out are kept. Secrets are never returned. Owners only."
1271	1277		}
1272	1278		Op::DisconnectIntegration => {
1273	1279		"Remove an integration and its secrets. Agents already running on a model provider being removed stop reaching it. Owners only."
⋯
1517	1523		"Who a workspace's invoices are made out to: the billing `email`, `name`, `address`, tax ID (`tax_id_type`, `tax_id`), `po_number` and the invoices' `language`, with the default `payment_method` as far as it is safe to show (its kind, brand, last four digits and expiry). `customer` is false until the workspace has been set up to pay. Tax is worked out from the address: `tax_location` says whether it is enough for that (a country, and in the US a ZIP code), `tax_address_needed_at` is set while g1t is holding a charge for want of one, `tax_id_status` is Stripe's check of the tax ID (`pending`, `verified`, `unverified` or `unavailable`), and `tax_exempt` is `none`, `exempt` or `reverse`. Members of the workspace only."
1518	1524		}
1519	1525		Op::ListGatewayRequests => {
1520		−	"A workspace's recent AI Gateway requests, newest first: each with its `id`, `created_at`, `model`, the access token that sent it (`token_id`, `token_name`), its tokens by kind (`input`, `output`, `cache_read`, `cache_write`), what they cost at the model's price (`cost_micros`) and what the workspace was charged (`charged_micros`, before included usage and AI credit paid for it; 0 on the workspace's own provider key, `own_key`), the HTTP `status` it was answered with, whether it was `streamed`, `duration_ms`, and `error` for one that was refused or failed. Prompts and answers are never kept. `limit` is how many, 50 unless given and 200 at most; pass `next` from one page as `before` for the next. Requests are kept `retention_days` (30). Members of the workspace only."
	1526	+	"A workspace's recent AI Gateway requests, newest first: each with its `id`, `created_at`, `model`, the access token that sent it (`token_id`, `token_name`), its tokens by kind (`input`, `output`, `cache_read`, `cache_write`, and of those writes `cache_write_hour` to the hour-long cache), the `format` it was sent in (`anthropic` or `openai`), who served it (`provider`: `anthropic` or `workers-ai` on g1t's account, the connection's provider on the workspace's own, and `connection`, that connection's name), what they cost at the model's price (`cost_micros`) and what the workspace was charged (`charged_micros`, before included usage and AI credit paid for it; 0 on the workspace's own provider key, `own_key`), the HTTP `status` it was answered with, whether it was `streamed`, `duration_ms`, and `error` for one that was refused or failed. Prompts and answers are never kept. `limit` is how many, 50 unless given and 200 at most; pass `next` from one page as `before` for the next. Requests are kept `retention_days` (30). Members of the workspace only."
1521	1527		}
1522	1528		Op::ListUserTeams => {
1523	1529		"The teams someone is in within a workspace, as list_teams describes them, leaving out secret teams you cannot see. Members of the workspace only."
⋯
2260	2266		"name": { "type": "string", "description": "What to call it. The provider's name if left out." },
2261	2267		"config": {
2262	2268		"type": "object",
2263		−	"description": "Settings. repo (owner/name) is where alerts open issues; assign puts an agent on each; label names the label (bug). organization is the Sentry org's slug. site is Jira's address; email the account its token belongs to; keys the project or team keys it answers for. base_url and auth_header (x-api-key or authorization) are for your own endpoint; model overrides the model for every kind of work. write_back (default true) tells the outside system when the work lands.",
	2269	+	"description": "Settings. repo (owner/name) is where alerts open issues; assign puts an agent on each; label names the label (bug). organization is the Sentry org's slug. site is Jira's address; email the account its token belongs to; keys the project or team keys it answers for. base_url and auth_header (x-api-key or authorization) are for your own endpoint; model overrides the model for every kind of work; gateway_models (model ids, or prefixes ending in * such as gpt-* or ollama/*) chooses which AI Gateway requests go to a model provider. write_back (default true) tells the outside system when the work lands.",
2264	2270		},
2265		−	"secret": { "type": "string", "description": "The API key or token g1t uses to call it." },
	2271	+	"secret": { "type": "string", "description": "The API key or token g1t uses to call it. Write-only: kept encrypted, never returned." },
2266	2272		"signing_secret": { "type": "string", "description": "For sentry: the integration's client secret." },
2267	2273		}),
2268	2274		&["workspace", "provider"],
2269	2275		),
	2276	+	Op::UpdateIntegration => object(
	2277	+	json!({
	2278	+	"workspace": workspace_schema(),
	2279	+	"id": { "type": "string", "description": "The integration's id." },
	2280	+	"name": { "type": "string", "description": "A new name." },
	2281	+	"config": {
	2282	+	"type": "object",
	2283	+	"description": "Its settings, replaced whole: the same fields as connect_integration's config. For a model provider, gateway_models chooses the AI Gateway models it takes.",
	2284	+	},
	2285	+	"secret": { "type": "string", "description": "A new API key or token, replacing the old one. Write-only: kept encrypted, never returned." },
	2286	+	"signing_secret": { "type": "string", "description": "For sentry: a new client secret." },
	2287	+	}),
	2288	+	&["workspace", "id"],
	2289	+	),
2270	2290		Op::GetModelRoutes => object(json!({ "workspace": workspace_schema() }), &["workspace"]),
2271	2291		Op::ListWebhooks => object(hook_owner(json!({})), &[]),
2272	2292		Op::ListWorkflows => repo_only(),
⋯
2910	2930		\| Op::CreateRepo
2911	2931		\| Op::ListIntegrations
2912	2932		\| Op::ConnectIntegration
	2933	+	\| Op::UpdateIntegration
2913	2934		\| Op::DisconnectIntegration
2914	2935		\| Op::TestIntegration
2915	2936		\| Op::GetModelRoutes
⋯
4094	4115		)
4095	4116		.await
4096	4117		}
	4118	+	Op::UpdateIntegration => {
	4119	+	let config = match &input["config"] {
	4120	+	Value::Null => Value::Null,
	4121	+	config => camel_keys(config),
	4122	+	};
	4123	+	pass(
	4124	+	integrations,
	4125	+	"update",
	4126	+	&json!({
	4127	+	"actor": actor(),
	4128	+	"workspace": workspace(),
	4129	+	"id": text(input, "id"),
	4130	+	"name": optional_text(input, "name"),
	4131	+	"config": config,
	4132	+	"secret": optional_text(input, "secret"),
	4133	+	"signingSecret": optional_text(input, "signing_secret"),
	4134	+	}),
	4135	+	)
	4136	+	.await
	4137	+	}
4097	4138		Op::DisconnectIntegration \| Op::TestIntegration => {
4098	4139		pass(
4099	4140		integrations,

3713	3713		},
3714	3714		"notes": "`signing_secret` is set only when g1t made it, for `datadog` and `webhook`, and is shown only this once. A model provider is tested as it is connected. See [integrations](/guides/integrations/) and [model providers](/guides/models/)."
3715	3715		},
	3716	+	"update_integration": {
	3717	+	"params": {
	3718	+	"workspace": "flagon-io",
	3719	+	"id": "con_01kpx5c2d8e4f6g0h2j4k6m8n0"
	3720	+	},
	3721	+	"request": {
	3722	+	"config": {
	3723	+	"base_url": "https://gpu.flagon.dev/v1",
	3724	+	"gateway_models": ["ollama/*"]
	3725	+	},
	3726	+	"secret": "sk-office-gpu-key"
	3727	+	},
	3728	+	"response": {
	3729	+	"id": "con_01kpx5c2d8e4f6g0h2j4k6m8n0",
	3730	+	"workspace": "flagon-io",
	3731	+	"provider": "openai_endpoint",
	3732	+	"kind": "models",
	3733	+	"name": "Office GPU",
	3734	+	"config": {
	3735	+	"write_back": true,
	3736	+	"base_url": "https://gpu.flagon.dev/v1",
	3737	+	"gateway_models": ["ollama/*"]
	3738	+	},
	3739	+	"secret_hint": "…-key",
	3740	+	"webhook_url": null,
	3741	+	"created_by": "syntaqx",
	3742	+	"created_at": "2026-10-07T12:10:44.512Z",
	3743	+	"last_used_at": "2026-10-07T14:00:02.000Z",
	3744	+	"last_error": null,
	3745	+	"models": ["llama3.3:70b", "qwen3-coder:30b"]
	3746	+	},
	3747	+	"notes": "The secret is write-only: it is kept encrypted and only its last four characters come back, as `secret_hint`. With `gateway_models` set to `ollama/*`, an AI Gateway request for `ollama/qwen3-coder:30b` reaches this endpoint as `qwen3-coder:30b`, on the workspace's own account. See [the AI Gateway guide](/guides/ai-gateway/#your-own-providers)."
	3748	+	},
3716	3749		"disconnect_integration": {
3717	3750		"params": {
3718	3751		"workspace": "flagon-io",
⋯
5966	5999		"workspace": "flagon-io"
5967	6000		},
5968	6001		"query": {
5969		−	"limit": 2
	6002	+	"limit": 3
5970	6003		},
5971	6004		"response": {
5972	6005		"requests": [
⋯
5980	6013		"output": 512,
5981	6014		"cache_read": 12000,
5982	6015		"cache_write": 0,
5983		−	"cost_micros": 11200,
5984		−	"charged_micros": 11200,
	6016	+	"cache_write_hour": 0,
	6017	+	"cost_micros": 10000,
	6018	+	"charged_micros": 10000,
5985	6019		"status": 200,
5986	6020		"own_key": false,
	6021	+	"format": "anthropic",
	6022	+	"provider": "anthropic",
	6023	+	"connection": null,
5987	6024		"streamed": true,
5988	6025		"duration_ms": 4210,
5989	6026		"error": null
5990	6027		},
5991	6028		{
	6029	+	"id": "gw_9b3d1f7a5c2e4b6d8f0a1c33",
	6030	+	"created_at": "2026-10-07T14:00:02.000Z",
	6031	+	"model": "qwen3-coder:30b",
	6032	+	"token_id": "tok_01kkr5a2b8c4d6e0f3g5h7j9k1",
	6033	+	"token_name": "triage-bot",
	6034	+	"input": 2210,
	6035	+	"output": 96,
	6036	+	"cache_read": 0,
	6037	+	"cache_write": 0,
	6038	+	"cache_write_hour": 0,
	6039	+	"cost_micros": 0,
	6040	+	"charged_micros": 0,
	6041	+	"status": 200,
	6042	+	"own_key": true,
	6043	+	"format": "openai",
	6044	+	"provider": "openai_endpoint",
	6045	+	"connection": "Office GPU",
	6046	+	"streamed": false,
	6047	+	"duration_ms": 1830,
	6048	+	"error": null
	6049	+	},
	6050	+	{
5992	6051		"id": "gw_1a7e3c5b9d2f4e6a8c0b2d41",
5993	6052		"created_at": "2026-10-07T13:58:40.000Z",
5994	6053		"model": "claude-opus-5-5",
⋯
5998	6057		"output": 0,
5999	6058		"cache_read": 0,
6000	6059		"cache_write": 0,
	6060	+	"cache_write_hour": 0,
6001	6061		"cost_micros": 0,
6002	6062		"charged_micros": 0,
6003	6063		"status": 402,
6004	6064		"own_key": false,
	6065	+	"format": "openai",
	6066	+	"provider": "",
	6067	+	"connection": null,
6005	6068		"streamed": false,
6006	6069		"duration_ms": 38,
6007	6070		"error": "The flagon-io workspace is out of AI credit and has used this month's included usage, so the AI Gateway refuses requests to g1t's models. An owner can buy AI credit or turn on auto-reload at /flagon-io/-/billing#ai-credit."
⋯
6010	6073		"next": "gw_1a7e3c5b9d2f4e6a8c0b2d41",
6011	6074		"retention_days": 30
6012	6075		},
6013		−	"notes": "To send requests, see the AI Gateway guide: POST https://models.g1t.sh/anthropic/v1/messages with a workspace access token that has `models:write`. `charged_micros` is what the request was charged before included usage and AI credit paid for it; the payment itself is on the statement as AI Gateway. Pass `next` as `before` for the next page; it is null on the last."
	6076	+	"notes": "To send requests, see the AI Gateway guide: POST https://models.g1t.sh/anthropic/v1/messages (Anthropic's format) or https://models.g1t.sh/openai/v1/chat/completions (OpenAI's) with a workspace access token that has `models:write`. `format` is the format a request was sent in; `provider` who served it (`anthropic` or `workers-ai` on g1t's account, the connection's provider on the workspace's own, empty when it was refused before reaching one), and `connection` the workspace's own connection by name. `cache_write_hour` is the part of `cache_write` written to the hour-long cache. `charged_micros` is what the request was charged before included usage and AI credit paid for it; the payment itself is on the statement as AI Gateway. Pass `next` as `before` for the next page; it is null on the last."
6014	6077		},
6015	6078		"list_invoices": {
6016	6079		"params": {

485	485		&[],
486	486		),
487	487		route(
	488	+	"PATCH",
	489	+	"/workspaces/:workspace/integrations/:id",
	490	+	Op::UpdateIntegration,
	491	+	&[],
	492	+	),
	493	+	route(
488	494		"DELETE",
489	495		"/workspaces/:workspace/integrations/:id",
490	496		Op::DisconnectIntegration,

287	287		a("revoke_invite", Op::RevokeWorkspaceInvite, "Revoke a pending invite"),
288	288		a("list_integrations", Op::ListIntegrations, "Model providers, alert sources, trackers"),
289	289		a("connect_integration", Op::ConnectIntegration, "Connect one"),
	290	+	a("update_integration", Op::UpdateIntegration, "Change one: rotate its key, choose its AI Gateway models"),
290	291		a("disconnect_integration", Op::DisconnectIntegration, "Remove one"),
291	292		a("test_integration", Op::TestIntegration, "Check its credentials"),
292	293		a("get_model_routes", Op::GetModelRoutes, "Where each kind of work's model requests go"),
⋯
417	418		\| Op::RemoveEmail
418	419		\| Op::RemoveCollaborator
419	420		\| Op::DisconnectIntegration
	421	+	\| Op::UpdateIntegration
420	422		\| Op::DeleteWebhook
421	423		\| Op::DeleteActionsSecret
422	424		\| Op::DeleteActionsVariable

1	1		---
2	2		title: AI Gateway
3		−	description: Send your own code's model requests through g1t with a workspace access token, paid from AI credit at the model's price, with a log of every request.
	3	+	description: Send your own code's model requests through g1t in Anthropic's or OpenAI's format, with a workspace access token, to Claude, open models or your own providers, with a log of every request.
4	4		---
5	5
6		−	The AI Gateway takes model requests from your own code, in Anthropic's
7		−	Messages format, and sends them to the model. You point an Anthropic SDK,
8		−	Claude Code or anything else that speaks that format at one base URL and
9		−	give it a workspace access token as its API key:
	6	+	The AI Gateway takes model requests from your own code and sends them to
	7	+	the model. It speaks two formats, and any model works in either:
	8	+
	9	+	\| Format \| Base URL \| For \|
	10	+	\| --- \| --- \| --- \|
	11	+	\| Anthropic's Messages API \| `https://models.g1t.sh/anthropic` \| Anthropic's SDKs, Claude Code, and anything else that speaks that format \|
	12	+	\| OpenAI's Chat Completions API \| `https://models.g1t.sh/openai/v1` \| OpenAI's SDKs, and any tool that lets you set an OpenAI-compatible base URL \|
10	13
11		−	\| \| \|
12		−	\| --- \| --- \|
13		−	\| Base URL \| `https://models.g1t.sh/anthropic` \|
14		−	\| API key \| A workspace access token (`g1t_…`) with the `models:write` scope \|
	14	+	In both, the API key is a workspace access token (`g1t_…`) with the
	15	+	`models:write` scope.
15	16
16		−	Each request is charged to the workspace at the model's price and paid from
17		−	the plan's included usage and [AI credit](/guides/usage-and-billing/#ai-credit).
18		−	While the gateway is in beta there is no markup. If the workspace has
19		−	connected its own Anthropic key, requests go there instead and cost
20		−	nothing on g1t. Every request is logged with its model, tokens, cost and
21		−	status. Prompts and answers are never kept.
	17	+	The model a request names decides where it goes:
	18	+
	19	+	- g1t's models: Claude on Anthropic, and open models on Workers AI.
	20	+	Each request is charged to the workspace at the model's price and paid
	21	+	from the plan's included usage and
	22	+	[AI credit](/guides/usage-and-billing/#ai-credit). While the gateway is in
	23	+	beta there is no markup.
	24	+	- Your own providers: an Anthropic key, an OpenAI key, or any endpoint
	25	+	that speaks either API, connected under
	26	+	[Integrations](/guides/models/#connect-a-provider). You choose which models
	27	+	go to each. Those requests are counted and never charged on g1t.
	28	+
	29	+	Every request is logged with its format, who served it, its model, tokens,
	30	+	cost and status. Prompts and answers are never kept.
22	31
23	32		## Before you start
24	33
⋯
26	35
27	36		- The workspace on the g1t plan, with AI credit or this month's
28	37		included usage left. See [AI credit](/guides/usage-and-billing/#ai-credit).
29		−	- The workspace's own Anthropic key, connected under
30		−	[Integrations](/guides/models/#connect-a-provider). Then the plan is not
31		−	needed and nothing is charged.
	38	+	- One of the workspace's own model providers, connected under
	39	+	[Integrations](#your-own-providers). Requests for the models it takes need
	40	+	no plan and cost nothing on g1t.
32	41
33	42		## Make a token
34	43
⋯
51	60
52	61		## Send a request
53	62
54		−	The gateway answers the same routes as Anthropic's API, below the base URL:
	63	+	The token goes in `x-api-key` or in `Authorization: Bearer`, in either
	64	+	format. The gateway answers these routes:
55	65
56	66		\| Route \| What it does \|
57	67		\| --- \| --- \|
58	68		\| `POST /anthropic/v1/messages` \| A message, streamed (`"stream": true`) or whole. Logged and charged. \|
59		−	\| `POST /anthropic/v1/messages/count_tokens` \| Counts a request's input tokens. Not logged, and costs nothing. \|
	69	+	\| `POST /anthropic/v1/messages/count_tokens` \| Counts a request's input tokens. Not logged, and costs nothing. For a model that does not speak Anthropic's API, an estimate. \|
	70	+	\| `POST /openai/v1/chat/completions` \| A chat completion, streamed or whole. Logged and charged. \|
	71	+	\| `POST /openai/v1/embeddings` \| Embeddings, from an embeddings model. Logged and charged by their input tokens. \|
	72	+	\| `GET /openai/v1/models` \| The models this workspace can use, with g1t's prices. \|
60	73
61		−	The token goes in `x-api-key`, or in `Authorization: Bearer`. Request and
62		−	answer bodies are Anthropic's, unchanged, and so are streamed events.
	74	+	### Anthropic's format
63	75
64	76		With curl:
65	77
⋯
69	81		-H "anthropic-version: 2023-06-01" \
70	82		-H "content-type: application/json" \
71	83		-d '{
72		−	"model": "claude-sonnet-5-5",
	84	+	"model": "claude-haiku-5-5",
73	85		"max_tokens": 1024,
74	86		"messages": [{ "role": "user", "content": "Write a commit message for: fix the login redirect" }]
75	87		}'
⋯
86	98		});
87	99
88	100		const message = await client.messages.create({
89		−	model: "claude-sonnet-5-5",
	101	+	model: "claude-haiku-5-5",
90	102		max_tokens: 1024,
91	103		messages: [{ role: "user", content: "Write a commit message for: fix the login redirect" }],
92	104		});
⋯
105	117		)
106	118
107	119		message = client.messages.create(
108		−	model="claude-sonnet-5-5",
	120	+	model="claude-haiku-5-5",
109	121		max_tokens=1024,
110	122		messages=[{"role": "user", "content": "Write a commit message for: fix the login redirect"}],
111	123		)
112	124		```
113	125
114		−	### Claude Code
	126	+	Request and answer bodies are Anthropic's, and so are streamed events. To
	127	+	an open model, such as `workers-ai/@cf/openai/gpt-oss-120b`, the request is
	128	+	translated: messages, system prompt, images, tools and tool results,
	129	+	`tool_choice`, stop sequences, `output_config.effort` (as
	130	+	`reasoning_effort`, `xhigh` and `max` as `high`) and `output_config.format`
	131	+	(as a JSON schema). The answer comes back as an Anthropic message, tool
	132	+	calls included. Server tools have no counterpart there and are refused on
	133	+	g1t's models.
	134	+
	135	+	### OpenAI's format
	136	+
	137	+	With curl:
	138	+
	139	+	```sh
	140	+	curl https://models.g1t.sh/openai/v1/chat/completions \
	141	+	-H "Authorization: Bearer $G1T_TOKEN" \
	142	+	-H "content-type: application/json" \
	143	+	-d '{
	144	+	"model": "anthropic/claude-haiku-5-5",
	145	+	"messages": [{ "role": "user", "content": "Write a commit message for: fix the login redirect" }]
	146	+	}'
	147	+	```
	148	+
	149	+	With OpenAI's TypeScript SDK:
	150	+
	151	+	```ts
	152	+	import OpenAI from "openai";
	153	+
	154	+	const client = new OpenAI({
	155	+	baseURL: "https://models.g1t.sh/openai/v1",
	156	+	apiKey: process.env.G1T_TOKEN,
	157	+	});
	158	+
	159	+	const completion = await client.chat.completions.create({
	160	+	model: "workers-ai/@cf/openai/gpt-oss-120b",
	161	+	messages: [{ role: "user", content: "Label this issue: the login page is blank on Safari" }],
	162	+	});
	163	+	```
	164	+
	165	+	With OpenAI's Python SDK:
	166	+
	167	+	```python
	168	+	import os
	169	+
	170	+	from openai import OpenAI
	171	+
	172	+	client = OpenAI(
	173	+	base_url="https://models.g1t.sh/openai/v1",
	174	+	api_key=os.environ["G1T_TOKEN"],
	175	+	)
	176	+
	177	+	completion = client.chat.completions.create(
	178	+	model="anthropic/claude-sonnet-5-5",
	179	+	messages=[{"role": "user", "content": "Summarize this diff in one sentence."}],
	180	+	stream=True,
	181	+	stream_options={"include_usage": True},
	182	+	)
	183	+	for chunk in completion:
	184	+	print(chunk.choices[0].delta.content or "" if chunk.choices else "", end="")
	185	+	```
	186	+
	187	+	Embeddings:
	188	+
	189	+	```sh
	190	+	curl https://models.g1t.sh/openai/v1/embeddings \
	191	+	-H "Authorization: Bearer $G1T_TOKEN" \
	192	+	-H "content-type: application/json" \
	193	+	-d '{ "model": "workers-ai/@cf/baai/bge-m3", "input": ["fix the login redirect"] }'
	194	+	```
	195	+
	196	+	What OpenAI's format supports, to any model:
	197	+
	198	+	\| In the request \| \|
	199	+	\| --- \| --- \|
	200	+	\| `messages` \| `system`, `developer`, `user`, `assistant` and `tool` messages. User content can be text, `image_url` (a URL or a `data:` URL) and, to Claude, `file` with `file_data` (a PDF as a `data:` URL). \|
	201	+	\| `tools`, `tool_choice`, `parallel_tool_calls` \| Function tools. To Claude, `required` is `any` and a named function is that tool. \|
	202	+	\| `stream`, `stream_options.include_usage` \| Server-sent chunks, ending with `data: [DONE]`. With `include_usage`, a last chunk carries the usage. \|
	203	+	\| `max_tokens`, `max_completion_tokens` \| To Claude, 8,192 when neither is given. \|
	204	+	\| `temperature`, `top_p`, `stop`, `user` \| As given. Some models refuse sampling settings. \|
	205	+	\| `reasoning_effort` \| To Claude, `output_config.effort`; `minimal` and `none` are `low`. \|
	206	+	\| `response_format` \| `json_schema` is a JSON schema the answer follows. `json_object` asks Claude for one JSON object. \|
	207	+	\| `thinking` \| To Claude, passed as it is, for a caller that sets Anthropic's thinking. \|
	208	+
	209	+	To Claude, the answer's `usage` counts cached tokens in `prompt_tokens`,
	210	+	with `prompt_tokens_details.cached_tokens`; Claude's thinking comes back as
	211	+	`reasoning_content`. Claude's thinking blocks must go back with the tool
	212	+	calls they led to, so the gateway carries them in the first tool call's
	213	+	`id`: send the `id` back unchanged in the assistant message and the `tool`
	214	+	message, as OpenAI's SDKs do. `n` above 1 is refused for Claude.
	215	+
	216	+	### Claude Code and other tools
115	217
116		−	Set two environment variables before you start it:
	218	+	Claude Code speaks Anthropic's format. Set two environment variables
	219	+	before you start it:
117	220
118	221		```sh
119	222		export ANTHROPIC_BASE_URL=https://models.g1t.sh/anthropic
⋯
123	226
124	227		Claude Code's own small requests go to Claude Haiku 4.5, which the gateway
125	228		offers. To choose the main model, also set `ANTHROPIC_MODEL`, such as
126		−	`claude-opus-5-5`.
	229	+	`claude-opus-5-5`, or `ANTHROPIC_SMALL_FAST_MODEL`, such as
	230	+	`claude-haiku-5-5`.
127	231
	232	+	Any other tool that lets you set an OpenAI-compatible base URL and key
	233	+	works the same way: give it `https://models.g1t.sh/openai/v1` and the
	234	+	token, and a model id from [the models list](#models).
	235	+
128	236		## Models
129	237
130		−	On g1t's models the gateway offers these. Prices are per million tokens,
131		−	the provider's list price; cache writes are five-minute ones.
	238	+	### Model ids
132	239
133		−	\| Model \| `model` \| Input \| Output \| Cache reads \| Cache writes \|
134		−	\| --- \| --- \| --- \| --- \| --- \| --- \|
135		−	\| Claude Opus 5.5 \| `claude-opus-5-5` \| $4.00 \| $20.00 \| $0.20 \| $5.00 \|
136		−	\| Claude Sonnet 5.5 \| `claude-sonnet-5-5` \| $2.00 \| $10.00 \| $0.20 \| $2.50 \|
137		−	\| Claude Haiku 4.5 \| `claude-haiku-4-5`, `claude-haiku-4-5-20251001` \| $1.00 \| $5.00 \| $0.10 \| $1.25 \|
	240	+	A request names a model:
138	241
139		−	A request for any other model is refused with `400` before it reaches the
140		−	provider, and the error names the models offered. On the workspace's own
141		−	key, a request can name any model that key can use.
	242	+	\| Id \| Goes to \|
	243	+	\| --- \| --- \|
	244	+	\| `anthropic/claude-sonnet-5-5` \| Claude on g1t's account, in either format \|
	245	+	\| `claude-sonnet-5-5` \| The same, as Anthropic's API names it \|
	246	+	\| `workers-ai/@cf/openai/gpt-oss-120b` \| An open model on g1t's account, in either format \|
	247	+	\| `@cf/openai/gpt-oss-120b` \| The same \|
	248	+	\| Any id one of your own providers takes, such as `gpt-5.5` or `ollama/llama3.3` \| That provider, with its key. See [your own providers](#your-own-providers). \|
142	249
143		−	On g1t's models a request is charged only by its tokens, so what the
144		−	provider bills some other way is refused with `400` for now:
	250	+	Your own providers come first: when one of them takes a model, the request
	251	+	goes there, even one that names a model g1t offers. An Anthropic key takes
	252	+	`claude-*` unless you choose otherwise, so with one connected, Claude goes to
	253	+	your key.
145	254
	255	+	`GET /openai/v1/models` lists what the workspace can use: its own
	256	+	providers' models first, then g1t's, cheapest Claude first. Each has
	257	+	`billed_to` (`workspace` or `g1t`), `connection` (your provider's name) and,
	258	+	on g1t's models, `pricing` in dollars per million tokens:
	259	+
	260	+	```json
	261	+	{
	262	+	"object": "list",
	263	+	"data": [
	264	+	{
	265	+	"id": "anthropic/claude-haiku-5-5",
	266	+	"object": "model",
	267	+	"created": 0,
	268	+	"owned_by": "anthropic",
	269	+	"name": "Claude Haiku 5.5",
	270	+	"kind": "chat",
	271	+	"billed_to": "g1t",
	272	+	"connection": null,
	273	+	"pricing": {
	274	+	"currency": "usd",
	275	+	"input": 0.1,
	276	+	"output": 0.5,
	277	+	"cache_read": 0.01,
	278	+	"cache_write": 0.125,
	279	+	"cache_write_1h": 0.2,
	280	+	"long_prompt": { "above_tokens": 100000, "input": 0.5, "output": 2.5, "cache_read": 0.05, "cache_write": 0.625, "cache_write_1h": 1 }
	281	+	}
	282	+	}
	283	+	]
	284	+	}
	285	+	```
	286	+
	287	+	### Claude, on Anthropic
	288	+
	289	+	Prices are per million tokens, Anthropic's list price. Cache writes are
	290	+	five-minute ones; one-hour cache writes (`"ttl": "1h"`) cost twice the
	291	+	input price.
	292	+
	293	+	\| Model \| `model` \| Input \| Output \| Cache reads \| Cache writes \| One-hour cache writes \|
	294	+	\| --- \| --- \| --- \| --- \| --- \| --- \| --- \|
	295	+	\| Claude Haiku 5.5 \| `claude-haiku-5-5` \| $0.10 \| $0.50 \| $0.01 \| $0.125 \| $0.20 \|
	296	+	\| Claude Haiku 5.5, prompts over 100,000 tokens \| `claude-haiku-5-5` \| $0.50 \| $2.50 \| $0.05 \| $0.625 \| $1.00 \|
	297	+	\| Claude Sonnet 5.5 \| `claude-sonnet-5-5` \| $2.00 \| $10.00 \| $0.10 \| $2.50 \| $4.00 \|
	298	+	\| Claude Opus 5.5 \| `claude-opus-5-5` \| $4.00 \| $20.00 \| $0.20 \| $5.00 \| $8.00 \|
	299	+	\| Claude Haiku 4.5 \| `claude-haiku-4-5`, `claude-haiku-4-5-20251001` \| $1.00 \| $5.00 \| $0.10 \| $1.25 \| $2.00 \|
	300	+
	301	+	Claude Haiku 5.5 is the cheapest Claude and the one to start with. It is
	302	+	priced by the prompt's length: a request whose prompt (its input, cache
	303	+	read and cache write tokens) is longer than 100,000 tokens is charged
	304	+	entirely at the higher prices. It takes effort, like Opus: set
	305	+	`output_config.effort` in Anthropic's format, or `reasoning_effort` in
	306	+	OpenAI's.
	307	+
	308	+	### Open models, on Workers AI
	309	+
	310	+	Prices are per million tokens, Cloudflare's list price. Workers AI has no
	311	+	prompt-cache price: cached tokens, where a model reports them, cost what
	312	+	input does.
	313	+
	314	+	\| Model \| `model` \| Input \| Output \|
	315	+	\| --- \| --- \| --- \| --- \|
	316	+	\| GLM-5.3 Flash \| `workers-ai/@cf/zai-org/glm-5.3-flash` \| $0.15 \| $0.50 \|
	317	+	\| gpt-oss-20b \| `workers-ai/@cf/openai/gpt-oss-20b` \| $0.20 \| $0.30 \|
	318	+	\| Llama 4 Scout \| `workers-ai/@cf/meta/llama-4-scout-17b-16e-instruct` \| $0.27 \| $0.85 \|
	319	+	\| gpt-oss-120b \| `workers-ai/@cf/openai/gpt-oss-120b` \| $0.35 \| $0.75 \|
	320	+	\| Mistral Small 3.1 \| `workers-ai/@cf/mistralai/mistral-small-3.1-24b-instruct` \| $0.351 \| $0.555 \|
	321	+	\| DeepSeek V4 Flash \| `workers-ai/@cf/deepseek-ai/deepseek-v4-flash-0731` \| $0.44 \| $1.32 \|
	322	+	\| Nemotron 3 120B \| `workers-ai/@cf/nvidia/nemotron-3-120b-a12b` \| $0.50 \| $1.50 \|
	323	+	\| Kimi K2.6 \| `workers-ai/@cf/moonshotai/kimi-k2.6` \| $0.95 \| $4.00 \|
	324	+	\| DeepSeek V4 Pro \| `workers-ai/@cf/deepseek-ai/deepseek-v4-pro-0813` \| $1.32 \| $3.96 \|
	325	+	\| GLM-5.3 \| `workers-ai/@cf/zai-org/glm-5.3` \| $1.40 \| $4.40 \|
	326	+
	327	+	Embeddings, through `POST /openai/v1/embeddings` only:
	328	+
	329	+	\| Model \| `model` \| Input \|
	330	+	\| --- \| --- \| --- \|
	331	+	\| BGE M3 \| `workers-ai/@cf/baai/bge-m3` \| $0.012 \|
	332	+	\| BGE Base (English) \| `workers-ai/@cf/baai/bge-base-en-v1.5` \| $0.067 \|
	333	+
	334	+	Open models cost much less per call than Claude, and suit one-shot work:
	335	+	titles, summaries, labels, triage, embeddings. In a long loop that sends
	336	+	the same context every turn, Claude's cache reads close most of that gap.
	337	+
	338	+	### What is refused on g1t's models
	339	+
	340	+	A request for a model nobody offers is refused with `404` before it
	341	+	reaches a provider, and the error names the models offered. On g1t's
	342	+	models a request is charged only by its tokens, so what a provider bills
	343	+	some other way is refused with `400` for now:
	344	+
146	345		\| Not offered on g1t's models yet \| In the request \|
147	346		\| --- \| --- \|
148	347		\| Fast mode \| `speed` other than `standard` \|
149	348		\| Inference in one region \| `inference_geo` other than `global` \|
150	349		\| Server-side fallbacks \| `fallbacks` \|
151		−	\| Server tools, such as web search, web fetch and code execution \| A tool whose `type` is not your own (`custom` or none) or a client tool (`bash_…`, `text_editor_…`, `computer_…`, `memory_…`) \|
	350	+	\| Server tools, such as web search, web fetch and code execution \| In Anthropic's format, a tool whose `type` is not your own (`custom` or none) or a client tool (`bash_…`, `text_editor_…`, `computer_…`, `memory_…`). In OpenAI's, a tool that is not a `function`, or `web_search_options`. \|
152	351		\| Containers and skills \| `container` \|
153	352
154		−	All of them work on the workspace's own key, which the provider bills. In
155		−	Claude Code on g1t's models, its web search fails for this reason; the
156		−	rest of Claude Code works.
	353	+	All of them work on your own provider, which bills them. In Claude Code on
	354	+	g1t's models, its web search fails for this reason; the rest of Claude Code
	355	+	works.
157	356
158		−	Anthropic's format is the one served today. OpenAI's format and open
159		−	models are coming later.
	357	+	## Your own providers
	358	+
	359	+	Connect a model provider under Integrations, and choose which models
	360	+	your own code's gateway requests send to it. Requests there use its key,
	361	+	are counted in the log, and are never charged on g1t. Any provider works:
160	362
	363	+	\| Provider \| Takes, unless you choose \|
	364	+	\| --- \| --- \|
	365	+	\| An Anthropic key, or an Anthropic-compatible endpoint \| `claude-*` \|
	366	+	\| An OpenAI key, or any other provider \| Nothing until you choose \|
	367	+	\| An OpenAI-compatible endpoint: a self-hosted vLLM or Ollama, LiteLLM, another provider \| Nothing until you choose \|
	368	+
	369	+	1. Open the workspace's Integrations and choose a provider under
	370	+	Model providers. For your own server, choose **OpenAI-compatible
	371	+	endpoint or Anthropic-compatible endpoint** and give its base URL.
	372	+	2. Paste its key. It is sealed when saved and never shown again: the page,
	373	+	the API and MCP show only its last four characters.
	374	+	3. Under AI Gateway models, list the models to send there, separated by
	375	+	spaces:
	376	+
	377	+	\| Write \| Takes \|
	378	+	\| --- \| --- \|
	379	+	\| `gpt-5.5` \| That model only \|
	380	+	\| `gpt-*` \| Every model whose id starts with `gpt-` \|
	381	+	\| `ollama/*` \| Every model named `ollama/…`, sent without the prefix: `ollama/llama3.3` arrives as `llama3.3` \|
	382	+	\| `*` \| Every model \|
	383	+	\| Nothing \| No gateway requests \|
	384	+
	385	+	4. Select Connect. To change the list or replace the key later, open
	386	+	Change its AI Gateway models or key under the provider.
	387	+
	388	+	The first provider, in the order they were connected, that takes a model
	389	+	gets its requests. Either format reaches either kind of provider: a Claude
	390	+	key answers OpenAI-format requests, and an OpenAI-compatible endpoint
	391	+	answers Claude Code. Agent runs choose their models under
	392	+	[routing](/guides/models/), apart from this list.
	393	+
	394	+	From code, connect one with
	395	+	[`POST /workspaces/{workspace}/integrations`](/reference/api/integrations/connect-integration/)
	396	+	and change it with
	397	+	[`PATCH /workspaces/{workspace}/integrations/{id}`](/reference/api/integrations/update-integration/),
	398	+	with `config.gateway_models`, or the `workspace` MCP tool's
	399	+	`connect_integration` and `update_integration` actions. Both need
	400	+	`workspace:admin` and act for an owner. The key is write-only: neither
	401	+	returns it.
	402	+
	403	+	```sh
	404	+	curl -X PATCH https://api.g1t.sh/workspaces/acme/integrations/con_01kpx5c2d8e4f6g0h2j4k6m8n0 \
	405	+	-H "Authorization: Bearer $G1T_ADMIN_TOKEN" \
	406	+	-H "content-type: application/json" \
	407	+	-d '{ "config": { "base_url": "https://gpu.acme.dev/v1", "gateway_models": ["ollama/*"] }, "secret": "…" }'
	408	+	```
	409	+
	410	+	`config` replaces the provider's settings whole, so send the ones it has
	411	+	with the change.
	412	+
161	413		## What it costs
162	414
163	415		\| Where it goes \| You pay \|
164	416		\| --- \| --- \|
165	417		\| g1t's models \| Its tokens at the model's price above, with no markup while the gateway is in beta \|
166		−	\| The workspace's own Anthropic key \| Nothing on g1t. The provider bills you for the model. \|
	418	+	\| Your own providers \| Nothing on g1t. The provider bills you for the model. \|
167	419
168	420		On g1t's models:
169	421
170	422		- Each request that used tokens is one line on the statement, under
171	423		AI Gateway, such as *AI Gateway: Claude Sonnet 5.5, 14,352 tokens,
172		−	token release-notes*.
	424	+	token release-notes*. A Claude Haiku 5.5 request over 100,000 prompt
	425	+	tokens says long-prompt price.
173	426		- The plan's included usage pays first, then AI credit. Trial credit and
174	427		g1t's open-source pool never pay for gateway requests.
175	428		- It is not an agent run, so the [agent rate](/guides/usage-and-billing/#the-agent-rate)
⋯
178	431		like any other usage, and shows on Usage under the AI Gateway product.
179	432		- A workspace with a 100% discount gets it free through the discount; an
180	433		enterprise is invoiced for it after use.
181		−
182		−	### Your own key
183		−
184		−	When the workspace has an Anthropic or Anthropic-compatible model provider
185		−	under [Integrations](/guides/models/), the gateway sends every request to
186		−	the first one connected, with its key. Those requests are logged with their
187		−	tokens and marked Own key, and g1t charges nothing for them. Remove the
188		−	provider and requests go to g1t's models again within a few seconds.
189	434
190	435		## Limits and errors
191	436
⋯
196	441		usage left. If auto-reload is on, g1t tries it first.
197	442		- It is not on the g1t plan.
198	443
199		−	Errors are Anthropic's shape, so SDKs raise their usual errors:
	444	+	Errors are in the format of the route, so SDKs raise their usual errors.
	445	+	Anthropic's:
200	446
201	447		```json
202	448		{ "type": "error", "error": { "type": "billing_error", "message": "The acme workspace is out of AI credit …" } }
203	449		```
204	450
205		−	\| Status \| `error.type` \| Why \|
206		−	\| --- \| --- \| --- \|
207		−	\| `400` \| `invalid_request_error` \| The body is not JSON, the model is not offered, or the request asks for something [not offered on g1t's models yet](#models). \|
208		−	\| `401` \| `authentication_error` \| The token is unknown, expired or deleted. \|
209		−	\| `402` \| `billing_error` \| Out of AI credit, over the spend limit, or not on the plan. The message says what an owner can do. \|
210		−	\| `403` \| `permission_error` \| Not a workspace's token, or it lacks `models:write`. \|
211		−	\| `404` \| `not_found_error` \| A route the gateway does not answer. \|
	451	+	OpenAI's:
212	452
213		−	An error from the model provider, such as `429` or `529`, comes back as
214		−	the provider sent it. Refused and failed requests are logged with their
215		−	status and why, and cost nothing.
	453	+	```json
	454	+	{ "error": { "message": "The acme workspace is out of AI credit …", "type": "insufficient_quota", "param": null, "code": "insufficient_quota" } }
	455	+	```
	456	+
	457	+	\| Status \| Anthropic's `error.type` \| OpenAI's `error.type` (`code`) \| Why \|
	458	+	\| --- \| --- \| --- \| --- \|
	459	+	\| `400` \| `invalid_request_error` \| `invalid_request_error` \| The body is not JSON, the model is of the wrong kind, the request asks for something [not offered on g1t's models yet](#what-is-refused-on-g1ts-models), or it cannot be said to the model (such as `n` above 1 to Claude). \|
	460	+	\| `401` \| `authentication_error` \| `authentication_error` (`invalid_api_key`) \| The token is unknown, expired or deleted, or your provider refused its key. \|
	461	+	\| `402` \| `billing_error` \| `insufficient_quota` (`insufficient_quota`) \| Out of AI credit, over the spend limit, or not on the plan. The message says what an owner can do. \|
	462	+	\| `403` \| `permission_error` \| `permission_error` \| Not a workspace's token, or it lacks `models:write`. \|
	463	+	\| `404` \| `not_found_error` \| `invalid_request_error` (`model_not_found`) \| No provider offers the model, or a route the gateway does not answer. \|
	464	+
	465	+	An error from the model provider, such as `429` or `529`, comes back with
	466	+	its status and message, in the route's format. A provider's key never
	467	+	appears in an error or the log, even when the provider quotes it. Refused
	468	+	and failed requests are logged with their status and why, and cost
	469	+	nothing. Every answer carries `x-g1t-request-id`, the request's id in the
	470	+	log.
216	471
217		−	A deleted token, a provider added under Integrations, or AI credit just
218		−	bought takes effect within about ten seconds.
	472	+	A deleted token, a provider added or changed under Integrations, or AI
	473	+	credit just bought takes effect within about ten seconds.
219	474
220	475		## See every request
221	476
⋯
226	481		\| Column \| \|
227	482		\| --- \| --- \|
228	483		\| Time \| When it was sent. Hover for the exact time, how long it took and whether it streamed. \|
229		−	\| Model \| The model it named. On the workspace's own key, the one that answered. \|
230		−	\| Input, Output, Cache read, Cache write \| Its tokens by kind. \|
231		−	\| Cost \| What it was charged, before included usage and AI credit paid for it, or Own key. \|
	484	+	\| Model \| The model it named on g1t's models, or the one that answered on your own provider; below it, the format it was sent in. \|
	485	+	\| Served by \| g1t's account and the provider (g1t · Anthropic, g1t · Workers AI), or your provider by name. None when it was refused first. \|
	486	+	\| Input, Output \| Its tokens by kind. Hover Input for all of them. \|
	487	+	\| Cache \| Cache reads, then cache writes. Hover for how many writes were to the one-hour cache. \|
	488	+	\| Cost \| What it was charged, before included usage and AI credit paid for it, or Not charged on your own provider. \|
232	489		\| Status \| The status it was answered with. Hover a refusal or failure for why. \|
233	490		\| Token \| The name of the token that sent it. \|
234	491
⋯
237	494		From code, list them with
238	495		[`GET /workspaces/{workspace}/gateway/requests`](/reference/api/billing/list-gateway-requests/),
239	496		or the `billing` MCP tool's
240		−	[`gateway_requests`](/reference/mcp/#billing) action. Both need
241		−	`models:read`, which the Read only and Agent presets include.
	497	+	[`gateway_requests`](/reference/mcp/#billing) action. Each request has
	498	+	`format`, `provider`, `connection`, `model` and its tokens, with
	499	+	`cache_write_hour` for one-hour cache writes. Both need `models:read`,
	500	+	which the Read only and Agent presets include.

208	208		\| --- \| --- \|
209	209		\| `workspace` `list_integrations` \| `GET /workspaces/{workspace}/integrations` \|
210	210		\| `workspace` `connect_integration` \| `POST /workspaces/{workspace}/integrations` \|
	211	+	\| `workspace` `update_integration` \| `PATCH /workspaces/{workspace}/integrations/{id}` \|
211	212		\| `workspace` `test_integration` \| `POST /workspaces/{workspace}/integrations/{id}/test` \|
212	213		\| `workspace` `disconnect_integration` \| `DELETE /workspaces/{workspace}/integrations/{id}` \|
213	214		\| `search` `ticket` \| `GET /repos/{owner}/{name}/context?reference=` \|

18	18		kind of work, which provider and model it runs on. Each provider bills
19	19		you for the model directly. Open to every workspace now.
20	20
21		−	To call models from your own code with a workspace token, paid from the
22		−	same AI credit, use the [AI Gateway](/guides/ai-gateway/).
	21	+	To call models from your own code with a workspace token, in Anthropic's
	22	+	or OpenAI's format, paid from the same AI credit or sent to these same
	23	+	providers, use the [AI Gateway](/guides/ai-gateway/). Each provider's
	24	+	AI Gateway models say which of those requests go to it; see
	25	+	[your own providers](/guides/ai-gateway/#your-own-providers).
23	26
24	27		## Auto
25	28

548	548		\| [`revoke_invite`](/reference/api/invites/revoke-workspace-invite/) \| Revoke a workspace's pending invite. Owners only. \| `workspace`, `id` \| `workspace:admin` \|
549	549		\| [`list_integrations`](/reference/api/integrations/list-integrations/) \| The workspace's connections. Secrets are never returned. Members only. \| `workspace` \| `workspace:read` \|
550	550		\| [`connect_integration`](/reference/api/integrations/connect-integration/) \| Connect a model provider (Anthropic, OpenAI, Gemini, or a compatible endpoint), Sentry, Datadog, a webhook, Jira or Linear, with `config` and `secret`. Owners only. \| `workspace`, `provider` \| `workspace:admin` \|
	551	+	\| [`update_integration`](/reference/api/integrations/update-integration/) \| Change one: its `name`, its `config` (replaced whole) or its `secret` (write-only, never returned). Rotates a model provider's key, or sets `config.gateway_models`, the [AI Gateway](/guides/ai-gateway/#your-own-providers) models it takes. Owners only. \| `workspace`, `id` \| `workspace:admin` \|
551	552		\| [`disconnect_integration`](/reference/api/integrations/disconnect-integration/) \| Remove it and its secrets. Owners only. \| `workspace`, `id` \| `workspace:admin` \|
552	553		\| [`test_integration`](/reference/api/integrations/test-integration/) \| Check its credentials against the system it connects to. Owners only. \| `workspace`, `id` \| `workspace:admin` \|
553	554		\| [`get_model_routes`](/reference/api/integrations/get-model-routes/) \| Which provider and model each kind of work goes to. Members only. \| `workspace` \| `workspace:read` \|
⋯
583	584		\| [`buy_ai_credit`](/reference/api/billing/buy-ai-credit/) \| A payment page (`url`) to buy `amount_cents` of credit, in whole dollars from $10 to $1,000, for a person to open and pay; it returns to the workspace's billing page. Owners, as people. \| `workspace`, `amount_cents` \| `billing:write` \|
584	585		\| [`invoices`](/reference/api/billing/list-invoices/) \| Every invoice (`invoices`, in cents), g1t's itemised usage invoices (`usage_invoices`), and what the next one comes to so far (`upcoming`). \| `workspace` \| `billing:read` \|
585	586		\| [`billing_details`](/reference/api/billing/get-billing-details/) \| Who invoices are made out to, and the payment method on file as far as it is safe to show. \| `workspace` \| `billing:read` \|
586		−	\| [`gateway_requests`](/reference/api/billing/list-gateway-requests/) \| The workspace's recent [AI Gateway](/guides/ai-gateway/) requests, newest first: model, tokens by kind, `cost_micros`, `charged_micros`, `status`, `own_key` and the token that sent each. `limit` (50, at most 200) and `before` (the last page's `next`) page through them. Kept 30 days. \| `workspace` \| `models:read` \|
	587	+	\| [`gateway_requests`](/reference/api/billing/list-gateway-requests/) \| The workspace's recent [AI Gateway](/guides/ai-gateway/) requests, newest first: model, `format` (`anthropic` or `openai`), `provider` and `connection` (who served it), tokens by kind with `cache_write_hour`, `cost_micros`, `charged_micros`, `status`, `own_key` and the token that sent each. `limit` (50, at most 200) and `before` (the last page's `next`) page through them. Kept 30 days. \| `workspace` \| `models:read` \|
587	588
588	589		## `notifications`
589	590

Binary or large file; its contents are not shown.

7	7
8	8		import type { GatewayRequest } from "@g1t/contracts";
9	9
10		−	import { duration, shortCount, statusTone, tokenKinds } from "../lib/gateway";
	10	+	import { cacheKinds, duration, formatLabel, servedBy, shortCount, statusTone, tokenKinds } from "../lib/gateway";
11	11		import { money } from "../lib/usage";
12	12		import { TimeAgo } from "./ui";
13	13		import { Badge } from "./ui/badge";
⋯
18	18		const NUM = "px-3 py-2 text-right tabular-nums whitespace-nowrap";
19	19
20	20		/** The columns, so the skeleton has as many as the table. */
21		−	const COLUMNS = ["Time", "Model", "Input", "Output", "Cache read", "Cache write", "Cost", "Status", "Token"] as const;
	21	+	const COLUMNS = ["Time", "Model", "Served by", "Input", "Output", "Cache", "Cost", "Status", "Token"] as const;
22	22
23	23		/** One request's row. */
24	24		function Row({ request }: { request: GatewayRequest }) {
25	25		const status = statusTone(request.status);
	26	+	const served = servedBy(request);
26	27		return (
27	28		<tr className="text-sm">
28	29		<td className="px-3 py-2 whitespace-nowrap text-muted">
⋯
32	33		</span>
33	34		</Hint>
34	35		</td>
35		−	<td className="px-3 py-2 font-mono text-xs whitespace-nowrap">{request.model}</td>
	36	+	<td className="max-w-[15rem] px-3 py-2 whitespace-nowrap">
	37	+	<Hint label={request.model}>
	38	+	<span tabIndex={0} className="block truncate font-mono text-xs">
	39	+	{request.model}
	40	+	</span>
	41	+	</Hint>
	42	+	<span className="block text-[0.6875rem] text-faint">{formatLabel(request.format)} format</span>
	43	+	</td>
	44	+	<td className="max-w-[12rem] truncate px-3 py-2 whitespace-nowrap">
	45	+	<Hint label={served.hint}>
	46	+	<span tabIndex={0} className={request.ownKey ? "" : "text-muted"}>
	47	+	{served.label}
	48	+	</span>
	49	+	</Hint>
	50	+	</td>
36	51		<td className={NUM}>
37	52		<Hint label={tokenKinds(request)}>
38	53		<span tabIndex={0}>{shortCount(request.input)}</span>
39	54		</Hint>
40	55		</td>
41	56		<td className={NUM}>{shortCount(request.output)}</td>
42		−	<td className={NUM}>{shortCount(request.cacheRead)}</td>
43		−	<td className={NUM}>{shortCount(request.cacheWrite)}</td>
44	57		<td className={NUM}>
	58	+	<Hint label={cacheKinds(request)}>
	59	+	<span tabIndex={0}>
	60	+	{shortCount(request.cacheRead)}
	61	+	<span className="text-faint"> / </span>
	62	+	{shortCount(request.cacheWrite)}
	63	+	</span>
	64	+	</Hint>
	65	+	</td>
	66	+	<td className={NUM}>
45	67		{request.ownKey ? (
46		−	<Hint label="Sent with the workspace's own provider key: counted, not charged.">
	68	+	<Hint label="Sent to the workspace's own provider: counted, not charged.">
47	69		<span tabIndex={0}>
48		−	<Badge>Own key</Badge>
	70	+	<Badge>Not charged</Badge>
49	71		</span>
50	72		</Hint>
51	73		) : (
⋯
69	91		}
70	92
71	93		/** Whether a column holds a number, read right-aligned. */
72		−	const numeric = (index: number) => index >= 2 && index <= 6;
	94	+	const numeric = (index: number) => index >= 3 && index <= 6;
73	95
74	96		/** The table's frame: its heading row, around its body. */
75	97		function Frame({ children }: { children: ReactNode }) {
⋯
114	136		<tr key={row} className="text-sm">
115	137		{COLUMNS.map((column, index) => (
116	138		<td key={column} className={numeric(index) ? NUM : "px-3 py-2"}>
117		−	<SkeletonLine className={numeric(index) ? "ml-auto w-10" : index === 1 ? "w-36" : "w-16"} />
	139	+	<SkeletonLine className={numeric(index) ? "ml-auto w-10" : index === 1 ? "w-36" : index === 2 ? "w-24" : "w-16"} />
118	140		</td>
119	141		))}
120	142		</tr>

7	7		import { type ReactNode, useState } from "react";
8	8		import { Form, Link } from "react-router";
9	9
10		−	import { type Connection, MODEL_TASKS, MODEL_TIERS, type ModelRoute, type ModelTask, type ModelTier, type Provider, PROVIDERS } from "@g1t/contracts";
	10	+	import { type Connection, MODEL_TASKS, MODEL_TIERS, type ModelRoute, type ModelTask, type ModelTier, type Provider, PROVIDERS, gatewayPatterns } from "@g1t/contracts";
11	11
12	12		import { Field, Input, SubmitButton } from "./ui";
13	13		import { Hint } from "./ui/hint";
⋯
259	259		</Field>
260	260		);
261	261
	262	+	/** What to suggest a provider's AI Gateway models be, as typed. */
	263	+	const GATEWAY_EXAMPLES: Partial<Record<Provider, string>> = {
	264	+	openai: "gpt-*",
	265	+	gemini: "gemini-*",
	266	+	xai: "grok-*",
	267	+	mistral: "mistral-* codestral-*",
	268	+	deepseek: "deepseek-*",
	269	+	openrouter: "openrouter/*",
	270	+	groq: "groq/*",
	271	+	together: "together/*",
	272	+	fireworks: "fireworks/*",
	273	+	cerebras: "cerebras/*",
	274	+	openai_endpoint: "ollama/*",
	275	+	anthropic_endpoint: "claude-*",
	276	+	azure_openai: "azure/*",
	277	+	};
	278	+
	279	+	/**
	280	+	* Which of the AI Gateway's requests go to a model provider, by the model
	281	+	* they name: for connecting one, and for changing it after.
	282	+	*/
	283	+	export function GatewayModelsField({ provider, value }: { provider: Provider; value?: string[] }) {
	284	+	const defaults = gatewayPatterns(provider, {});
	285	+	return (
	286	+	<Field
	287	+	label="AI Gateway models"
	288	+	hint="Your own code's requests to the AI Gateway that name one of these models come here, counted and never charged by g1t. Model ids, or prefixes ending in ; a prefix ending in / is taken off before sending, so ollama/llama3.3 arrives as llama3.3. Separate them with spaces. Empty: none."
	289	+	>
	290	+	<Input
	291	+	name="gatewayModels"
	292	+	defaultValue={(value ?? defaults).join(" ")}
	293	+	placeholder={GATEWAY_EXAMPLES[provider] ?? "model-id or prefix-*"}
	294	+	className="font-mono text-[0.8125rem]"
	295	+	autoComplete="off"
	296	+	spellCheck={false}
	297	+	/>
	298	+	</Field>
	299	+	);
	300	+	}
	301	+
262	302		/** What connecting a model provider asks for. */
263	303		export function ModelProviderFields({ provider }: { provider: Provider }): ReactNode {
	304	+	return (
	305	+	<>
	306	+	<ModelProviderKeyFields provider={provider} />
	307	+	<GatewayModelsField provider={provider} />
	308	+	</>
	309	+	);
	310	+	}
	311	+
	312	+	function ModelProviderKeyFields({ provider }: { provider: Provider }): ReactNode {
264	313		const entry = MODEL_CATALOG[provider];
265	314		if (provider === "azure_openai") {
266	315		return (

1	1		import assert from "node:assert/strict";
2	2		import { test } from "node:test";
3	3
4		−	import { duration, shortCount, statusTone, tokenKinds, totalTokens } from "./gateway.ts";
	4	+	import { cacheKinds, duration, formatLabel, parseGatewayModels, servedBy, shortCount, statusTone, tokenKinds, totalTokens } from "./gateway.ts";
5	5
6	6		test("token counts read short", () => {
7	7		assert.equal(shortCount(812), "812");
⋯
29	29		assert.equal(duration(4_210), "4.2 s");
30	30		assert.equal(duration(125_000), "2 min 5 s");
31	31		});
	32	+
	33	+	test("a request says which format it came in and who served it", () => {
	34	+	assert.equal(formatLabel("openai"), "OpenAI");
	35	+	assert.equal(formatLabel("anthropic"), "Anthropic");
	36	+	assert.deepEqual(servedBy({ ownKey: false, provider: "workers-ai", connection: null }).label, "g1t · Workers AI");
	37	+	assert.deepEqual(servedBy({ ownKey: false, provider: "anthropic", connection: null }).label, "g1t · Anthropic");
	38	+	const own = servedBy({ ownKey: true, provider: "openai_endpoint", connection: "Office GPU" });
	39	+	assert.equal(own.label, "Office GPU");
	40	+	assert.match(own.hint, /OpenAI-compatible endpoint: counted, not charged/);
	41	+	assert.equal(servedBy({ ownKey: false, provider: "", connection: null }).label, "None");
	42	+	});
	43	+
	44	+	test("a connection's AI Gateway models are typed as a list", () => {
	45	+	assert.deepEqual(parseGatewayModels(" gpt-, ollama/ claude-haiku-5-5 "), ["gpt-", "ollama/", "claude-haiku-5-5"]);
	46	+	assert.deepEqual(parseGatewayModels(""), []);
	47	+	});
	48	+
	49	+	test("cache tokens read in words, hour-long writes named", () => {
	50	+	assert.equal(cacheKinds({ cacheRead: 12_000, cacheWrite: 2_400, cacheWriteHour: 0 }), "12,000 read from the cache, 2,400 written");
	51	+	assert.equal(cacheKinds({ cacheRead: 0, cacheWrite: 2_400, cacheWriteHour: 2_400 }), "0 read from the cache, 2,400 written, 2,400 of them to the hour-long cache");
	52	+	});

10	10		/** The base URL an Anthropic SDK or Claude Code is pointed at. */
11	11		export const GATEWAY_BASE_URL = "https://models.g1t.sh/anthropic";
12	12
	13	+	/** The base URL an OpenAI SDK, or any tool that speaks OpenAI's API, is pointed at. */
	14	+	export const GATEWAY_OPENAI_BASE_URL = "https://models.g1t.sh/openai/v1";
	15	+
	16	+	/** How a request's format reads. */
	17	+	export function formatLabel(format: GatewayRequest["format"] \| undefined): string {
	18	+	return format === "openai" ? "OpenAI" : "Anthropic";
	19	+	}
	20	+
	21	+	/** Who served a request, as people read it, and a longer line for its hint. */
	22	+	export function servedBy(request: Pick<GatewayRequest, "ownKey" \| "provider" \| "connection">): { label: string; hint: string } {
	23	+	if (request.ownKey) {
	24	+	const name = request.connection \|\| "Your provider";
	25	+	return { label: name, hint: `${name}, the workspace's own ${providerName(request.provider)}: counted, not charged.` };
	26	+	}
	27	+	if (!request.provider) return { label: "None", hint: "Refused before it reached a model." };
	28	+	return { label: `g1t · ${providerName(request.provider)}`, hint: `g1t's account at ${providerName(request.provider)}, charged at the model's price.` };
	29	+	}
	30	+
	31	+	const PROVIDER_NAMES: Record<string, string> = {
	32	+	anthropic: "Anthropic",
	33	+	"workers-ai": "Workers AI",
	34	+	openai: "OpenAI",
	35	+	openai_endpoint: "OpenAI-compatible endpoint",
	36	+	anthropic_endpoint: "Anthropic-compatible endpoint",
	37	+	azure_openai: "Azure OpenAI",
	38	+	gemini: "Google Gemini",
	39	+	openrouter: "OpenRouter",
	40	+	};
	41	+
	42	+	/** A provider's name, as people know it. */
	43	+	export function providerName(provider: string): string {
	44	+	return PROVIDER_NAMES[provider] ?? (provider \|\| "provider");
	45	+	}
	46	+
	47	+	/** A connection's AI Gateway models, as a field shows them, from what was typed. */
	48	+	export function parseGatewayModels(text: string): string[] {
	49	+	return text
	50	+	.split(/[\s,]+/)
	51	+	.map((model) => model.trim())
	52	+	.filter(Boolean);
	53	+	}
	54	+
13	55		/** A count of tokens, short: `812`, `12.4K`, `3.1M`. */
14	56		export function shortCount(n: number): string {
15	57		if (n < 1_000) return String(n);
⋯
32	74		return `${n(request.input)} input, ${n(request.output)} output, ${n(request.cacheRead)} cache read, ${n(request.cacheWrite)} cache write`;
33	75		}
34	76
	77	+	/** A request's cache tokens in words, for a hint: read, written, and how many of the writes last an hour. */
	78	+	export function cacheKinds(request: Pick<GatewayRequest, "cacheRead" \| "cacheWrite" \| "cacheWriteHour">): string {
	79	+	const n = (v: number) => v.toLocaleString("en-US");
	80	+	const hour = request.cacheWriteHour ? `, ${n(request.cacheWriteHour)} of them to the hour-long cache` : "";
	81	+	return `${n(request.cacheRead)} read from the cache, ${n(request.cacheWrite)} written${hour}`;
	82	+	}
	83	+
35	84		/** How a status reads, and how much it matters. */
36	85		export function statusTone(status: number): { label: string; tone: "success" \| "warn" \| "danger" \| "neutral" } {
37	86		if (status >= 200 && status < 300) return { label: String(status), tone: "success" };

536	536		<p className="font-medium">
537	537		AI Gateway <span className="ml-1 rounded bg-accent/15 px-1.5 py-0.5 text-xs text-accent">Free during beta</span>
538	538		</p>
539		−	<p className="text-xs text-faint">Your own code calling Claude models through g1t with a workspace token, paid from AI credit; with your own Anthropic key, free</p>
	539	+	<p className="text-xs text-faint">Your own code calling Claude and open models through g1t, in Anthropic's or OpenAI's format, paid from AI credit; on your own providers, free</p>
540	540		</td>
541	541		<td className="px-4 py-3 text-muted">What the provider charges</td>
542	542		<td className="hidden px-4 py-3 tabular-nums sm:table-cell">{gateway?.markupPercent ?? 0}%</td>

1		−	import { BookOpen, KeyRound } from "lucide-react";
	1	+	import { BookOpen, KeyRound, Plug } from "lucide-react";
2	2		import { Suspense } from "react";
3	3		import { Await, Link, data } from "react-router";
4	4
5	5		import type { Route } from "./+types/gateway";
6	6		import { GatewaySkeleton, GatewayTable } from "../../components/gateway";
7	7		import { ButtonLink, CopyLine, EmptyState } from "../../components/ui";
8		−	import { GATEWAY_BASE_URL, GATEWAY_DOCS } from "../../lib/gateway";
	8	+	import { GATEWAY_BASE_URL, GATEWAY_DOCS, GATEWAY_OPENAI_BASE_URL } from "../../lib/gateway";
9	9		import { page } from "../../lib/meta";
10	10		import { billing } from "../../lib/services.server";
11	11		import { getViewer, roleIn } from "../../lib/session.server";
⋯
37	37		<div className="space-y-6">
38	38		<section className="space-y-3">
39	39		<p className="max-w-3xl text-sm text-muted">
40		−	Send your own code's model requests in Anthropic's Messages format to this base URL, with one of the workspace's
41		−	access tokens that has the <code className="font-mono text-xs">models:write</code> scope as the API key. On g1t's
42		−	models each request is charged at the model's price and paid from AI credit; with the workspace's own Anthropic key
43		−	under Integrations it is only counted. Prompts and answers are never kept.
	40	+	Send your own code's model requests to one of these base URLs, in Anthropic's or OpenAI's format, with one of the
	41	+	workspace's access tokens that has the <code className="font-mono text-xs">models:write</code> scope as the API key.
	42	+	Any model works in either format: Claude and open models on g1t's account are charged at the model's price and paid
	43	+	from AI credit; models on the workspace's own providers under Integrations are only counted. Prompts and answers are
	44	+	never kept.
44	45		</p>
45		−	<div className="max-w-xl">
46		−	<CopyLine text={GATEWAY_BASE_URL} />
47		−	</div>
	46	+	<dl className="grid max-w-3xl gap-3 md:grid-cols-2">
	47	+	<div className="min-w-0">
	48	+	<dt className="mb-1.5 text-xs text-faint">Anthropic format</dt>
	49	+	<dd>
	50	+	<CopyLine text={GATEWAY_BASE_URL} />
	51	+	</dd>
	52	+	</div>
	53	+	<div className="min-w-0">
	54	+	<dt className="mb-1.5 text-xs text-faint">OpenAI format</dt>
	55	+	<dd>
	56	+	<CopyLine text={GATEWAY_OPENAI_BASE_URL} />
	57	+	</dd>
	58	+	</div>
	59	+	</dl>
48	60		<div className="flex flex-wrap gap-2">
49	61		{owner && (
50	62		<ButtonLink variant="quiet" to={`/${slug}/-/tokens`}>
⋯
52	64		Access tokens
53	65		</ButtonLink>
54	66		)}
	67	+	<ButtonLink variant="quiet" to={`/${slug}/-/integrations`}>
	68	+	<Plug size={14} />
	69	+	Your own providers
	70	+	</ButtonLink>
55	71		<ButtonLink variant="quiet" to={GATEWAY_DOCS} reloadDocument>
56	72		<BookOpen size={14} />
57	73		How to use it

Binary or large file; its contents are not shown.