Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsUse Liquid AI’s d1 as a decision component inside your agent—not as the agent harness itself. Send it the current state and a bounded question, such as whether a condition is met, which available action to choose, or how to score the state. Your application then interprets the returned probabilities, applies its own policy, executes an allowed action, and gathers the next state.
What d1 does in an agent
Liquid AI describes d1 as a decision model that accepts text, images, or both, along with one or more questions, and returns probabilities rather than generated tokens. Its documented question types are noul for a yes-or-no decision, choice for probabilities across labels, and score for weighted probabilities across levels of a scale. Multiple questions about the same state can be sent in one request, but Liquid says each question is billed as its own prompt. See Liquid AI’s d1 launch announcement and its model overview.
As an Amazon Associate I earn from qualifying purchases.
This makes d1 suitable for agent steps whose possible outcomes can be named in advance—for example, routing a support ticket or selecting the next action from a web page. The surrounding application still needs to manage state, tool execution, action validation, retries, and guardrails. Liquid’s own agentic AI overview frames an agent as a model working together with a harness.
Build the decision loop
- Gather the current state. Collect the relevant text or visual input from your application, such as the current page and available actions.
- Ask a bounded question. Use
noul,choice, orscoreto express a decision that your application can act on. - Apply your own policy. Read the returned probabilities and decide whether to accept a choice, require a threshold, ask again, or escalate. The launch example exposes a
noulprobability atresponse.json()["answers"]["defect"]["noul"]; do not assume that example establishes every production response detail. - Validate and execute. Check that the proposed outcome maps to an allowed tool action, then execute it in your harness. Do not let an unconstrained model output directly authorize arbitrary tools.
- Observe the result and repeat. Gather the new state after the action and submit the next decision request as needed.
Liquid’s launch post illustrates this pattern with a web agent choosing the next action from the options on a flight-search page. That example demonstrates a decision call within an agent workflow, not a complete tool-executing framework.
#1 Best Overall
Call the Liquid AI API
Liquid’s October 5, 2026 launch post documents the hosted API endpoint https://api.liquid.ai/decisions/v1/systemone, model name d1, and bearer-token authorization. It instructs developers to create an API key in the Liquid AI Console under Dashboard → API Keys. The following Python pattern is adapted from the launch example; consult the official announcement and live API documentation for the current production schema.
import base64
import requests
LIQUID_API_KEY = "..." # Load securely from your environment or secret store.
with open("board.jpg", "rb") as image_file:
image_data = base64.b64encode(image_file.read()).decode("ascii")
response = requests.post(
"https://api.liquid.ai/decisions/v1/systemone",
headers={"Authorization": f"Bearer {LIQUID_API_KEY}"},
json={
"model": "d1",
"images": [f"data:image/jpeg;base64,{image_data}"],
"state": "Camera image of a circuit board on the production line.",
"questions": {
"defect": {
"type": "noul",
"instructions": "Does this circuit board have a defect?"
}
}
},
)
response.raise_for_status()
result = response.json()
defect_probability = result["answers"]["defect"]["noul"]The example includes an image, but a text-only decision can omit visual input. Treat the response access shown above as the launch post’s example rather than a complete guarantee of response validation or error behavior.
Choose a question type that matches the action
noul: Ask whether a defined condition holds, such as whether a page shows an error or a part appears defective. Your application decides what probability is sufficient to proceed.choice: Provide the available labels when the agent must select among known options—for example, one of the actions currently offered by a page. Validate the selected label against the application’s actual allowed actions.score: Use a defined scale when the downstream policy depends on a rating rather than a yes/no answer or a single label.
Prefer a general language-model call for a step that must generate a free-form explanation, compose content, or produce outcomes that cannot sensibly be enumerated. Liquid positions d1 for decisions across fixed outcomes; the cited material does not establish an independent apples-to-apples benchmark for a particular application.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Send screenshots or other visual state
The launch example sends an image as a base64 data URL in the images array, alongside a textual state and named question. Liquid says images count as input tokens at 1.5 tokens per 32×32-pixel patch; its example estimates a 1024×1024 image at 1,536 tokens. It also says each question is billed as a separate prompt, including the text and all images, so repeating the same screenshot across several questions can affect input usage.
Liquid’s launch post reports visual-inspection demonstrations with 85–97% accuracy across four production lines using the public VisA dataset, and says Wordle could be solved from screenshots without developers constructing a textual board representation. These are vendor-reported demonstrations, not independent benchmarks or guarantees for another workload. The post describes image support in the Liquid API; it said Vercel and OpenRouter supported text-only d1 at launch, with vision forthcoming. Provider capabilities can change, so check current provider documentation.
Account for cost, latency, and changing availability
Liquid’s October 5, 2026 launch post says d1 is billed only on input tokens, with no output-token charge, and lists a price of $0.04 per million input tokens. It says each question is billed as its own prompt and images use the same input-token rate as text. These are launch-post terms, not a current quote: verify pricing, plan conditions, request limits, and provider availability in the live official documentation before estimating production cost.
Liquid reports 200–300 ms for text decisions. Its launch post also reports that, across six applications, d1 matched or beat GPT-6.1 Sol on four and cost 19× to 200× less in that comparison. Liquid says each application was run once on October 5, 2026, with listed prices and up to eight requests in flight. Treat these figures as dated vendor comparisons, not performance or savings guarantees for your workload.
Free tools Windows power users keep installed
One-click scans. No signup required.
Check the production details before shipping
The launch announcement is an integration example, not a full production API reference. Before deploying, consult the current official decision-model documentation for request validation, supported image formats and limits, rate limits, error and timeout behavior, retry guidance, and service terms. The cited launch material does not establish those details, so build your harness to handle failures safely rather than assuming a response will always arrive in the example’s shape.
Best Value
Do not confuse d1 with Liquid’s on-device model
Liquid’s August 4, 2026 release describes LFM2.5-2.6B as a separate on-device model trained for agentic workloads such as planning, tool use, and multi-step tasks, with weights available on Hugging Face. That is a distinct model and deployment path. The d1 launch describes a hosted decision API, not a local version of d1.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
