Migration guide
Already on Jev? Point your existing client at Decision Models
Checked 15 November 2026, from static source inspection, not run against production. Jev is a TypeSafe AI product. Decision Models is independent and offers a Jev-compatible API.
Create a separate Decision Models API key. Select an available model from GET https://api.decisionmodels.io/v1/models; do not reuse a TypeSafe model ID. Start with one text question. Save your old configuration for rollback. Follow the migration checklist before switching traffic.
Choose the right URL
| Client | URL setting | Value |
|---|---|---|
| TypeSafe Python SDK | base_url | https://api.decisionmodels.io |
| TypeSafe TypeScript SDK | baseURL | https://api.decisionmodels.io |
| Vercel direct TypeSafe provider | baseURL | https://api.decisionmodels.io/v1 |
| LangChain Python / JS | base_url / baseUrl | https://api.decisionmodels.io |
Effect @effect/ai-typesafe | apiUrl | https://api.decisionmodels.io/v1 |
| Effect UAI | baseUrl | https://api.decisionmodels.io |
| TanStack AI | baseURL | https://api.decisionmodels.io |
Community n8n-nodes-jev | Credential Base URL | https://api.decisionmodels.io |
Scoped @n3ndor/n8n-nodes-typesafe-jev | Credential Base URL | https://api.decisionmodels.io |
| LiteLLM native System One | api_base | https://api.decisionmodels.io |
| Ollama Python | host | https://api.decisionmodels.io |
The Vercel and Effect providers in this table append /systemone; the other listed clients append /v1/systemone. This distinction prevents duplicate or missing /v1 paths. Pinned upstream sources are linked below.
TypeSafe SDKs
For typesafe-sdk 0.7.5, replace the client configuration with:
import os
from typesafe_sdk import TypeSafeClient
client = TypeSafeClient(
api_key=os.environ["DECISIONMODELS_API_KEY"],
base_url="https://api.decisionmodels.io",
model="s1-fast",
headers={"S1-Region": "eu"},
compression=False,
)Keep your existing text client.system_one(state=..., questions=...) call, with one question for this model. An explicit per-call model overrides the client default, so replace those IDs too.
For @typesafe-ai/sdk 0.7.0:
import { TypeSafeClient } from "@typesafe-ai/sdk";
const client = new TypeSafeClient({
apiKey: process.env.DECISIONMODELS_API_KEY,
baseURL: "https://api.decisionmodels.io",
defaultModel: "s1-fast",
defaultHeaders: { "S1-Region": "eu" },
compression: false,
});Continue with client.systemOne({state, questions}). Since 16 November 2026 the API also accepts gzip and zstd request bodies, so compression off is no longer required; the example keeps it off because the SDKs’ default compression was not tested end to end against the API. Both SDKs also read TYPESAFE_BASE_URL, TYPESAFE_API_KEY and TYPESAFE_DEFAULT_MODEL; prefer a separate process/environment for the new provider. Their optional model-list methods expect the TypeSafe discovery envelope. Read the Decision Models model list directly instead.
Sources: inspected Python constructor, TypeScript options, and model-list decoder, read 2026-11-15.
Community n8n nodes
For n8n-nodes-jev 0.2.3:
- Create its Jev (TypeSafe) API credential. Put your Decision Models key in API Key and
https://api.decisionmodels.ioin Base URL. - In the node, switch Model from From List to ID, then enter
s1-fastors1-pro. - Start with Route by Choice, which sends one question. For Ask Questions, keep the request within the selected model’s limit.
The model dropdown reads {models:[{name:...}]}. Since 16 November 2026 GET /v1/models returns that list next to data, so the dropdown should show the models; if it stays empty, use ID. The credential’s GET model-list check does not prove the key has inference permission or credit.
For @n3ndor/n8n-nodes-typesafe-jev, set the same credential Base URL and your new key. Set Model through an expression such as ={{ "s1-fast" }} because its dropdown also expects TypeSafe discovery. Keep one question for s1-fast. This scoped package is distinct from the similarly named unscoped package.
These credentials expose no region header setting. Configure the desired default region on the Decision Models key before using them.
Sources: community credential, manual model ID, scoped node properties, read 2026-11-15.
The official TypeSafe n8n node hides its Base URL setting. n8n-nodes-jev-classification exposes a URL but tests credentials using a hard-coded TypeSafe model. Neither currently has a clean setup path covered by this guide.
Vercel AI SDK, LangChain, Effect and TanStack
For the direct @ai-sdk/typesafe-ai provider, configure:
const provider = createTypeSafeAi({
apiKey: process.env.DECISIONMODELS_API_KEY,
baseURL: "https://api.decisionmodels.io/v1",
headers: { "S1-Region": "eu" },
});
const model = provider.decisionModel("s1-fast");Import createTypeSafeAi from @ai-sdk/typesafe-ai and use your existing decision/evaluation call with one text question. This configures the direct provider; it does not add a model listing to Vercel AI Gateway. Provider source, read 2026-11-15.
For LangChain, use TypeSafeClassifier(base_url="https://api.decisionmodels.io", api_key=..., model="s1-fast") in Python, or new TypeSafeClassifier({baseUrl:"https://api.decisionmodels.io", apiKey:..., model:"s1-fast"}) in JavaScript. Both read TYPESAFE_BASE_URL when the URL argument is omitted. Neither has a simple region-header constructor option; use the key’s region default or your existing custom transport.
For @effect/ai-typesafe, pass {apiUrl:"https://api.decisionmodels.io/v1", apiKey:Redacted.make(key)} to TypeSafeClient.layer and select TypeSafeDecisionModel.model("s1-fast"). Retain your existing Effect HTTP layer. transformClient can add a region header; otherwise set the key default. Skip its TypeSafe-shaped listModels() method.
For @effect-uai/typesafe-ai, pass {baseUrl:"https://api.decisionmodels.io", apiKey:Redacted.make(key)} to its Jev layer or make, and model:"s1-fast" in the decision request. Use one decision in the definition. Region comes from your injected HTTP client or the key default.
For TanStack @tanstack/ai-typesafe, configure createTypesafeDecider("s1-fast", key, {baseURL:"https://api.decisionmodels.io", headers:{"S1-Region":"eu"}}) and retain your bounded text evaluation call.
LiteLLM and Ollama Python
Use LiteLLM’s native systemone function with:
litellm.systemone(
model="typesafe/s1-fast",
api_base="https://api.decisionmodels.io",
api_key=os.environ["DECISIONMODELS_API_KEY"],
extra_headers={"S1-Region": "eu"},
state=state,
questions=questions,
)LiteLLM strips the typesafe/ prefix. Use one question in this example. For Ollama Python 0.6.3, use Client(host="https://api.decisionmodels.io", headers={"Authorization":"Bearer " + key, "S1-Region":"eu"}), then client.systemone(model="s1-fast", state=state, questions=questions). The migration checklist documents local contract-mock checks of LiteLLM 1.106.4 and Ollama Python 0.6.3; these were not production quality tests and were not rerun for this research.
Before switching traffic
Use text noul, choice or score questions. s1-fast accepts one question; s1-pro accepts three. Split larger requests yourself. Criteria, tokens and option counts must also fit the chosen model. Set S1-Region: eu or global when the client allows it; otherwise the key default applies.
Image transport differs and is outside this guide. Check errors and retry/settlement handling explicitly, including credit, limits and timeouts. Some typed clients omit extra metadata. Compare outputs on your own cases before changing traffic; preserve rollback configuration. Full compatibility and error details remain in the migration checklist.
Jev is a trademark of TypeSafe AI. Decision Models is not affiliated with TypeSafe AI and does not resell Jev.
https://api.system1models.ai keeps working.