Skyvern: Create Custom LLM

← Custom llms operations · Get started

Operation skyvern.create_custom_llm_v1_custom_llms_post

Provider route: POST /v1/custom-llms. Alter supplies the connection’s credentials and routing.

Call through Alter

Install @alter-ai/sdk @alter-ai/skyvern with npm, or alter-sdk-skyvern with pip. Supply your connected customer’s grant ID and the arguments described below.

import { App } from "@alter-ai/sdk";
import { SkyvernIntegration } from "@alter-ai/skyvern";

async function execute(input: Parameters<SkyvernIntegration["customLlms"]["createCustomLlm"]>[0]) {
  const app = new App({ apiKey: process.env.ALTER_API_KEY! });
  try {
    const provider = new SkyvernIntegration(app, { grantId: process.env.PROVIDER_GRANT_ID! });
    const call = await provider.customLlms.createCustomLlm(input);
    const result = await call.wait();
    return result.bodyJson();
  } finally {
    await app.close();
  }
}

import os
from alter_sdk import App
from alter_sdk_skyvern import SkyvernIntegration

async def execute(arguments):
    async with App(api_key=os.environ["ALTER_API_KEY"]) as app:
        provider = SkyvernIntegration(app, {"grant_id": os.environ["PROVIDER_GRANT_ID"]})
        call = await provider.custom_llms.create_custom_llm(arguments)
        result = await call.wait()
        return result.body_json()

Input

Field	Type	Presence	Description and constraints
$	object	Root	Additional keys: not allowed
body	Model	Required within parent	Expand Create Custom LLM — request body — body

Referenced models: Create Custom LLM — request body — body

Response

Response media type: application/json. The table describes the pinned successful response. Referenced models link to their complete fields.

Field	Type	Presence	Description and constraints
$	Model	Root	Expand Create Custom LLM — response body

Referenced models: Create Custom LLM — response body

Schemas

Create Custom LLM — request body — body

Field	Type	Presence	Description and constraints
$	object	Root
config	Model	Required within parent	Expand Create Custom LLM — request body — body config

Referenced models: Create Custom LLM — request body — body config

Create Custom LLM — response body

Field	Type	Presence	Description and constraints
$	object	Root
custom_llm	Model	Required within parent	Expand Create Custom LLM — response body — custom llm

Referenced models: Create Custom LLM — response body — custom llm

Create Custom LLM — request body — body config

Field	Type	Presence	Description and constraints
$	object	Root
add_assistant_prefix	boolean	Optional	Default: false
api_base	JSON	Optional
api_base (anyOf 1)	string	Alternative	Maximum length: 500
api_base (anyOf 2)	null	Alternative
api_key	JSON	Optional
api_key (anyOf 1)	string	Alternative	Maximum length: 1000
api_key (anyOf 2)	null	Alternative
api_version	JSON	Optional
api_version (anyOf 1)	string	Alternative	Maximum length: 100
api_version (anyOf 2)	null	Alternative
display_name	string	Required within parent	Maximum length: 120 Minimum length: 1
extra_parameters	object	Optional	Additional keys: allowed
max_completion_tokens	JSON	Optional
max_completion_tokens (anyOf 1)	integer	Alternative	Maximum: 1000000.0 Minimum: 1.0
max_completion_tokens (anyOf 2)	null	Alternative
model_name	string	Required within parent	Maximum length: 250 Minimum length: 1
provider	string	Required within parent	One of: openai_compatible, ollama, openrouter, gemini
reasoning_effort	JSON	Optional
reasoning_effort (anyOf 1)	string	Alternative	Maximum length: 50
reasoning_effort (anyOf 2)	null	Alternative
supports_vision	boolean	Optional	Default: true
temperature	JSON	Optional
temperature (anyOf 1)	number	Alternative	Maximum: 2.0 Minimum: 0.0
temperature (anyOf 2)	null	Alternative

Create Custom LLM — response body — custom llm

Field	Type	Presence	Description and constraints
$	object	Root
config	Model	Required within parent	Expand Create Custom LLM — response body — custom llm config
created_at	string	Required within parent
id	string	Required within parent
modified_at	string	Required within parent
organization_id	string	Required within parent
valid	boolean	Required within parent

Referenced models: Create Custom LLM — response body — custom llm config

Create Custom LLM — response body — custom llm config

Field	Type	Presence	Description and constraints
$	object	Root
add_assistant_prefix	boolean	Optional	Default: false
api_base	JSON	Optional
api_base (anyOf 1)	string	Alternative	Maximum length: 500
api_base (anyOf 2)	null	Alternative
api_key	JSON	Optional
api_key (anyOf 1)	string	Alternative	Maximum length: 1000
api_key (anyOf 2)	null	Alternative
api_version	JSON	Optional
api_version (anyOf 1)	string	Alternative	Maximum length: 100
api_version (anyOf 2)	null	Alternative
display_name	string	Required within parent	Maximum length: 120 Minimum length: 1
extra_parameters	object	Optional	Additional keys: allowed
max_completion_tokens	JSON	Optional
max_completion_tokens (anyOf 1)	JSON	Alternative
max_completion_tokens (anyOf 1) (anyOf 1)	integer	Alternative	Maximum: 1000000.0 Minimum: 1.0
max_completion_tokens (anyOf 1) (anyOf 2)	string	Alternative	Pattern: ^-?[0-9]+(?:\.[0-9]+)?(?:[eE][+-]?[0-9]+)?$
max_completion_tokens (anyOf 2)	null	Alternative
model_name	string	Required within parent	Maximum length: 250 Minimum length: 1
provider	string	Required within parent	Known values: openai_compatible, ollama, openrouter, gemini
reasoning_effort	JSON	Optional
reasoning_effort (anyOf 1)	string	Alternative	Maximum length: 50
reasoning_effort (anyOf 2)	null	Alternative
supports_vision	boolean	Optional	Default: true
temperature	JSON	Optional
temperature (anyOf 1)	JSON	Alternative
temperature (anyOf 1) (anyOf 1)	number	Alternative	Maximum: 2.0 Minimum: 0.0
temperature (anyOf 1) (anyOf 2)	string	Alternative	Pattern: ^-?[0-9]+(?:\.[0-9]+)?(?:[eE][+-]?[0-9]+)?$
temperature (anyOf 2)	null	Alternative

Policy, approvals and errors

SDK function policy ID: skyvern.create_custom_llm_v1_custom_llms_post. Catalog operation: create_custom_llm_v1_custom_llms_post. Rules can name either ID; function rules distinguish managed calls and conservatively cover raw calls sharing the route. Existing grant permissions and policy rules apply. wait() returns an immediate result or waits for a required approval; it never approves the request itself. Persist the approval ID to resume after a restart.

Provider errors retain their response; JSON decoding raises ProviderAPIError for provider failures. ManagedOperationError identifies contract/input/retention errors. See error handling and provider-specific behavior. Sensitive input may be refused before approval storage. A decoding error never resubmits a write.

Official provider source