> ## Documentation Index > Fetch the complete documentation index at: https://docs.parallel.ai/llms.txt > Use this file to discover all available pages before exploring further. # Decisions > Answer typed questions using current web evidence. ## OpenAPI ````yaml /public-openapi.json post /v1/decisions openapi: 3.1.0 info: title: Parallel API description: Parallel API contact: name: Parallel Support url: https://parallel.ai email: support@parallel.ai version: 0.1.2 servers: - url: https://api.parallel.ai description: Parallel API security: - ApiKeyAuth: [] tags: - name: Search description: >- Search returns ranked URLs with extended excerpts suitable for LLM consumption. Inputs are a natural-language objective and optional keyword queries. Source policies allow including or excluding specific domains and have configurable output sizes. The returned extended snippets contain dense, relevant information from relevant pages. - Result: ranked list with URL, title, and long text excerpts - name: Image Search description: >- Image Search returns ranked images from the web. Inputs are keyword queries and an optional natural-language objective. `fast`: lowest latency and cost. `advanced` (default): slower, with higher-quality results. - Result: title, direct image URL, source page URL, and dimensions - name: Extract description: >- Extract returns excerpts or full content from one or more URLs. Inputs are a list of URLs and an optional search objective and keyword queries. The returned excerpts or full content is formatted as markdown and suitable for LLM consumption. - Result: excerpts or full content from the URL formatted as markdown - name: Decisions description: >- Answer typed choice, score, and probability questions using current web evidence, with shared citations for the request. - name: Tasks description: >- The Task API executes web research and extraction tasks. Clients submit a natural-language objective with an optional input schema; the service plans retrieval, fetches relevant URLs, and returns outputs that conform to a provided or inferred JSON schema. Supports deep research style queries and can return rich structured JSON outputs. Processors trade-off between cost, latency, and quality. Each processor supports calibrated confidences. - Output metadata: citations, excerpts, reasoning, and confidence per field Task Groups enable batch execution of many independent Task runs with group-level monitoring and failure handling. - Submit hundreds or thousands of Tasks as a single group - Observe group progress and receive results as they complete - Real-time updates via Server-Sent Events (SSE) - Add tasks to an existing group while it is running - Group-level retry and error aggregation - name: FindAll description: >- The FindAll API discovers and evaluates entities that match complex criteria from natural language objectives. Submit a high-level goal and the service automatically generates structured match conditions, discovers relevant candidates, and evaluates each against the criteria. Returns comprehensive results with detailed reasoning, citations, and confidence scores for each match decision. Streaming events and webhooks are supported. - name: Monitor description: >- The Monitor API watches the web for material changes on a fixed frequency. Each monitor runs once on creation and then on its configured schedule, emitting events when meaningful changes are detected. - `event_stream` monitors track a search query and emit an event for each new material change. - `snapshot` monitors track a specific task run's output and emit an event when the output changes. Results can be polled via the events endpoint or delivered via webhooks. - name: Memory description: >- The Memory API retrieves and manages memories created by Tasks, Monitors, and FindAll runs. Memories can be personal or isolated with a `memory_scope_key`. - name: Chat API (Beta) description: >- The Chat API provides a programmatic chat-style text generation interface. It accepts a sequence of messages and returns model responses. Intended for assistant-like interactions and evaluation. Streaming responses are supported. - name: Responses API description: >- An OpenAI-Responses-compatible interface for answers grounded in live web research, with URL citations. Point any Responses-API client — the OpenAI Python SDK, OpenAI TypeScript SDK, the Agents SDK, or raw HTTP — at `https://api.parallel.ai` with your Parallel API key, set `model` to `parallel`, and call `/v1/responses`. - `input` accepts a plain string or an array of role/content messages (canonical OpenAI shape; text content only). - `reasoning.effort` (`low`/`medium`/`high`/`auto`) controls how much research is performed, trading response time for answer quality. - Multi-turn via `previous_response_id`. - Structured outputs via `text.format = {"type": "json_schema", "name": ..., "schema": {...}}`. - Streaming (`stream=true`) emits the standard OpenAI Responses SSE lifecycle: `response.created` and `response.in_progress`, then output item / content part / text delta events with URL-citation annotations, the matching `*.done` events, and a terminal `response.completed` — or `response.failed` if the request fails mid-stream. paths: /v1/decisions: post: tags: - Decisions summary: Decisions description: Answer typed questions using current web evidence. operationId: grounded_decisions_v1_decisions_post requestBody: content: application/json: schema: $ref: '#/components/schemas/GroundedDecisionRequest' example: state: company: Clay website: https://www.clay.com questions: has_cli: type: noul instructions: Does Clay offer an official command-line interface (CLI)? required: true responses: '200': description: Successful Response content: application/json: schema: $ref: '#/components/schemas/GroundedDecisionResponse' examples: live: summary: Captured response from a live request. value: mode: fast answers: has_cli: type: noul noul: 0.9992643798509062 usage: input_tokens: 4212 output_tokens: 0 citations: - url: https://developers.clay.com/concepts/cli-basics title: CLI basics excerpts: - >- Understand the Clay CLI output contract, exit codes, environment variables, and authentication. The Clay CLI (clay) is JSON-first: every command writes machine-readable JSON to stdout on success and a structured error envelope to stderr on failure. There are no spinners, colors, or progress bars, so agents and scripts can branch on the output and exit code directly. clay login opens a browser to authorize the CLI with OAuth — this works from a human terminal or an agent's shell tool. Use clay whoami as the canonical check that authentication is working, and clay logout to clear a stored credential. Each sign-in covers one workspace — the one you pick on the consent screen. To work across several, run clay login once per workspace: signing into a second keeps the first. clay workspaces list shows what you are signed into, clay workspaces current shows which one commands run against, and clay workspaces switch switches. - url: https://university.clay.com/docs/cli-security-privacy title: CLI security & privacy - Clay Docs excerpts: - >- Understand CLI authentication, role requirements, and how Clay handles data sent from your terminal. Overview Clay's CLI lets your team drive a Clay workspace from a terminal or coding agent instead of the UI. Because that means credentials live on a local machine and commands run outside the app, admins need to know who can sign in and what those sessions can reach. ‍ Note: The CLI is part of Clay's Agent Plugin, which is in open beta and currently supported on Mac and Linux. Windows is not supported during the beta. ‍ How CLI sessions are authenticated The recommended path is browser-based OAuth: Treat that file like any other credential on disk — it is not stored in your operating system's keychain. * Running clay logout clears the stored credentials for that machine. API keys are for calling the Clay public API directly — they don't sign in to the CLI. The supporting documentation — the SOC 2 Type II report, the current AI subprocessor list, and security questionnaire responses — is available through the Clay Trust Center , or by contacting security@clay.com . Your responsibilities: * Decide who gets CLI access — CLI access implies edit rights on the workspace. Grant Editor deliberately. How Clay handles data sent through the CLI When your team runs a CLI command: FAQs If we drive the CLI from a coding agent, what does that agent's provider see? The CLI itself communicates only with Clay. But when you run it through a coding agent, whatever you type into that agent — and whatever command output it reads back — is processed by that agent's provider under your agreement with them, not Clay's. * Do not sell my data © Clay 2025 – Born in Brooklyn. Claymation illustrations by the wonderful Hudson Christie . - url: https://docs.clayplatform.io/claycli/docs/cli.html title: Claycli · Clay CLI excerpts: - >- Clay CLI ### 3.9.1 * GitHub Claycli This is the Clay command line utility. Here we will provide you with documentation and examples that will help you to run all the functionalities that you need. ## Installation To install claycli you just need to type the following command: ``` $ npm install -g claycli Clay CLI ### 3.9.1 > [](#usage) Usage If installed globally, call `clay` from the command line with the [command](#commands) and `options` that you need. ``` Clay CLI ### 3.9.1 $ clay < command > [options] `Much like `git` ; `claycli` is configured using a [dotfile](https://medium.com/@webprolific/getting-started-with-dotfiles-43c3602fd789) ( `.clayconfig` ) in your home folder. In it, you may specify references to api keys and `urls / site prefixes` that you use frequently. Clay CLI ### 3.9.1 For `urls` and `site prefixes` , it will assume `http://` and port `80` unless you specify otherwise. Note that a `_site prefix_` is everything before the api route, e.g. `http://domain.com/site1` in `http://domain.com/site1/_components/article` .` [keys] local = ha8yds9a8shdf98asdf qa = 8 quwqwer09ewr0w9uer Clay CLI ### 3.9.1 prod = bj34b6345k634jnk63n4 [urls] local-site1 = https://localhost.site1.com: 3001 local-site2 = site2.com/site- 2 # http and port 80 Clay CLI ### 3.9.1 ``` For smaller `Clay` installations (or, ironically, for very large teams where devs spend most of their time on individual sites), you may specify a default api key and `url / site prefix` by using the `CLAYCLI_DEFAULT_KEY` and `CLAYCLI_DEFAULT_URL` environment variables. ## [](#commands) Commands * [`config`](#config) Clay CLI ### 3.9.1 * [`lint`](#lint) * [`import`](#import) * [`export`](#export) * [`compile`](#compile) - url: >- https://university.clay.com/fr/docs/audiences-for-agents title: Audiences for agents and the CLI - Clay Docs excerpts: - >- Audiences for agents and the CLI Read, segment, and manage your workspace's people, companies, and deals from the Clay CLI — and from the Clay agents that run it for you. Overview Audiences is the home for your workspace's own people, companies, and deals. The clay audiences command group brings those records to your terminal: you can create and edit segments, manage field definitions, count or read records, and read the activities and signal events attached to them — all without opening the Clay app. ‍ Note: These commands need Audiences enabled for your workspace, plus the Clay Agent Plugin installed and signed in. Install and sign-in steps live in the plugin doc, linked below. ‍ * The Clay CLI — you or your coding agent run clay audiences commands directly in a terminal. * Sculptor — Clay's in-app go-to-market co-pilot runs the same commands for you when you ask it about your records. * The Agent surface in Clay — runs the same commands while it reasons through a play and builds the workflow. FAQs Does reading Audiences from the CLI cost credits? Not today — Audiences is CLI-only, and no MCP tool serves it. Reps who connect a chat app to Clay over MCP get people and company search plus any Functions your team has enabled, and admins can additionally let them query owned accounts and opportunity data from the CRM. See MCP in Clay for how that access is configured. Do Clay's in-app agents really run the same commands? Yes. All three surfaces load the same Audiences instructions from the Clay Agent Plugin, so a capability that lands in the CLI reaches Sculptor and the Agent surface with it. That's also why the limitations above apply everywhere rather than to the terminal alone. How do I copy a segment? - url: https://developers.clay.com/quickstart title: Quickstart - Clay docs excerpts: - >- Build with Clay from your coding agent, or call the Clay API from your own applications. There are two ways to use Clay as a developer: install the agent plugin to build with Clay inside your coding agent, or call the Public API directly from your own applications and services. Agent plugin The agent plugin gives your coding agent Clay's skills and the clay CLI — the fastest way to build with Clay from a coding agent. It requires a coding-agent host with a shell — Claude Code, Codex, or Cursor — and doesn't work in general-purpose chat apps like Claude Desktop, claude.ai, or ChatGPT that can't run the CLI. Copy this into your agent of choice and let it install and set up the plugin for you: Set up the Clay plugin by following the steps in https://github.com/clay-run/agent-plugins This runs the plugin's setup skill, which puts clay on your PATH, signs you in, and verifies the CLI. Sign-in opens a browser to authorize the CLI. Agent plugin repository API Use the Public API to call Clay from your own applications and services — backend jobs, internal tools, queues, and product features. Requests authenticate with a Clay API key in the clay-api-key header. API reference Endpoint paths, request bodies, response schemas, and an interactive try-it console. Get an API key * Ask your agent to generate one for you after installing the agent plugin. * Run clay api-keys create --name "" with the clay CLI. The key is returned once, in the apiKey field. * Create one yourself in Clay under API and CLI → API keys. Keep keys server-side. - url: https://www.clay.com/agent-plugin title: >- Clay CLI: Build on Clay with any coding agent | clay.com excerpts: - >- description: Use Clay's CLI to build workflows and governance directly into your GTM system. Build in Clay directly via a coding agent with CLI. title: "Clay CLI: Build on Clay with any coding agent | clay.com" image: https://cdn.prod.website-files.com/61477f2c24a826836f969afe/6a4fbb2bf6b13e7ff04d15c7_open%20graph%20cli%20api.jpg Agent plugin Build on Clay with any coding agent Use Clay's API to build Clay data, workflows, and governance directly into your GTM system. Build in Clay directly via a coding agent with CLI. Install the Clay Plugin to start building Copy and paste into your agent of choice Set up the Clay plugin by following the steps in https://github.com/clay-run/agent-plugins Copied Call any functions, workflows, and Claygents from Clay's API so your coding agent references centrally governed GTM logic — including lead scoring and outbound copy — instead of rebuilding one-offs. Get started Use Case 03 — WORKFLOW AUTOMATION Vibecode GTM workflows Build workflows in natural language via Clay CLI, then manage observability in-product, using Clay Workflows. Get started Build on Clay with any coding agent Install the Clay Plugin and start building data, workflows, and governance into your GTM system. Install the Clay Plugin to start building Copy and paste into your agent of choice Set up the Clay plugin by following the steps in https://github.com/clay-run/agent-plugins Copied Developer docs press Enter to run — press R to reset { "@context" : "https://schema.org", "@type" : "WebSite", "name" : "Clay", "alternateName" : "Clay", "url" : "https://clay.com/" } - url: >- https://university.clay.com/docs/clay-api-cli?dub_id=DHFbP3TR3TaCG4pm title: Installing Clay Plugin (API & CLI) - Clay Docs excerpts: - >- Head to Guide: Using Clay Plugin for go-to-market plays you can run once the Plugin is connected — source your market, score it for fit, and route who gets worked first. ‍ Note: The Plugin installs Clay’s skills and the Clay CLI into your coding agent. For the full CLI quickstart, Public API setup, code examples, and guides for Searches, Routines, and Workflows, see developers.clay.com . They're separate surfaces, not two halves of one product. The Plugin installs Clay's skills and the Clay CLI into your coding agent, and the CLI is what you or your agent run in a shell for one-off runs, scripting, and inspecting data. It authenticates with the session created by clay login , so there's no separate key to configure. Can I build or write to Clay tables via CLI or API? There are no current plans to support table building from the API or CLI. Can I build Workflows with the API or CLI? Yes, Workflows are available from the CLI and Plugin in Beta . For production use, Clay-managed functions and custom functions are the more stable option. Are credit budgets available via the developer platform? Use Company URL Finder in Clay to turn a company name into that company's website domain. View article View article View article Settings & admin IP access restrictions Read, segment, and manage your workspace's people, companies, and deals from the Clay CLI — and from the Clay agents that run it for you. View article View article View article Other popular resources Experts Find a Clay Expert Explore our network of Clay experts and agencies. View experts View experts View experts Community * Do not sell my data © Clay 2025 – Born in Brooklyn. Claymation illustrations by the wonderful Hudson Christie . - url: https://developers.clay.com/ title: Overview - Clay docs excerpts: - >- Use Clay from agents, terminals, backend services, internal tools, and custom apps. Clay used to live primarily in the Clay UI. Now Clay also works where your agents, developers, and business systems already are. Use the developer platform to search Clay's GTM data, run reusable Clay logic, read table data, and work with Audiences programmatically. Start here Install the Clay plugin Best for Claude Code, Codex, Cursor, and agent-first work. The plugin ships skills and the Clay CLI. Call the API directly Best for backend jobs, internal tools, custom apps, queues, and product workflows. Core primitives * Primitive | Use it for * Searches | Find companies and people using Clay's GTM database * Routines | Run Clay-managed functions, custom functions, and Workflows * Tables (Enterprise) | Query existing Clay table data from known tables * Audiences | Work with records, saved segments, activities, and signal events Run a search, then enrich with a function Pull a target list from Clay data, then run Clay-managed functions on the results. Give an agent access to a Clay function Let an agent call Clay logic as part of its normal work. Build a Workflow (Alpha) Create, edit, validate, run, and inspect a Workflow from the plugin or CLI. - url: https://university.clay.com/fr/docs/clay-api-cli title: Installing Clay Plugin (API & CLI) - Clay Docs excerpts: - >- Installing Clay Plugin (API & CLI) Head to Guide: Using Clay Plugin for go-to-market plays you can run once the Plugin is connected — source your market, score it for fit, and route who gets worked first. ‍ Note: The Plugin installs Clay’s skills and the Clay CLI into your coding agent. * A paid coding agent plan: Claude (Pro, Max, Team, or Enterprise) for Claude Code, or ChatGPT (Plus, Pro, Business, Edu, or Enterprise) for Codex. If your company manages Claude, check that Claude Code is enabled for you. * Editor or Admin access in the Clay workspace you'll connect. * On Windows, admin rights (or help from IT) for one step. ‍ * Plan | Search results per request | Total search results * Free | 50 | 100/mo * Trial | 50 | 10k per 14 days * Paid self-serve plans | 500 | 1M/yr * Enterprise | 500 | 10M/yr FAQs Does using the API or CLI cost extra credits? The Plugin gives agent environments (Claude Code, Codex, Cursor) Clay's skills and the Clay CLI — the fastest path for agent-first and interactive coding workflows. The Public API is better suited for background jobs and your own apps, where you want to call Clay directly instead of installing anything. The Public API uses its own key rather than your CLI sign-in. Create one with the clay api-keys create command — the key is shown only once, so store it somewhere safe. They're separate surfaces, not two halves of one product. The Plugin installs Clay's skills and the Clay CLI into your coding agent, and the CLI is what you or your agent run in a shell for one-off runs, scripting, and inspecting data. It authenticates with the session created by clay login , so there's no separate key to configure. Can I build or write to Clay tables via CLI or API? There are no current plans to support table building from the API or CLI. - url: https://university.clay.com/docs/clay-api-cli title: Installing Clay Plugin (API & CLI) - Clay Docs excerpts: - >- Head to Guide: Using Clay Plugin for go-to-market plays you can run once the Plugin is connected — source your market, score it for fit, and route who gets worked first. ‍ Note: The Plugin installs Clay’s skills and the Clay CLI into your coding agent. * A paid coding agent plan: Claude (Pro, Max, Team, or Enterprise) for Claude Code, or ChatGPT (Plus, Pro, Business, Edu, or Enterprise) for Codex. If your company manages Claude, check that Claude Code is enabled for you. * Editor or Admin access in the Clay workspace you'll connect. * On Windows, admin rights (or help from IT) for one step. ‍ There is no additional cost to use the Plugin. The Plugin also gives you access to: * Audiences — Read and segment the workspace's own people, companies, and deals. See Audiences for agents and the CLI . * Searches — Find companies and people using structured filters over Clay's GTM dataset. Learn more about Searches in this video walkthrough . FAQs Does using the API or CLI cost extra credits? No. API and CLI calls consume the same credits and actions as the equivalent work done in-product. There is no additional cost because work is triggered via the developer platform instead of the UI. What's the difference between the Plugin and the Public API? The Plugin gives agent environments (Claude Code, Codex, Cursor) Clay's skills and the Clay CLI — the fastest path for agent-first and interactive coding workflows. The Public API is better suited for background jobs and your own apps, where you want to call Clay directly instead of installing anything. The Public API uses its own key rather than your CLI sign-in. Create one with the clay api-keys create command — the key is shown only once, so store it somewhere safe. Can I build or write to Clay tables via CLI or API? decision_id: decisions_34b4be1b1405e19ba5516d9bb76c56ad '422': description: Request validation error content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' example: type: error error: ref_id: message: Request validation error. x-code-samples: - lang: Python source: |- import os import requests response = requests.post( "https://api.parallel.ai/v1/decisions", headers={"x-api-key": os.environ["PARALLEL_API_KEY"]}, json={ "state": { "company": "Clay", "website": "https://www.clay.com", }, "questions": { "has_cli": { "type": "noul", "instructions": "Does Clay offer an official command-line interface (CLI)?", } }, }, timeout=30, ) response.raise_for_status() decision = response.json() answer = decision["answers"]["has_cli"] if answer["type"] == "refusal": print("No decision returned; review this question.") else: print("Estimated probability:", answer["noul"]) for citation in decision["citations"]: print(citation["url"]) - lang: TypeScript source: |- async function main() { const response = await fetch("https://api.parallel.ai/v1/decisions", { method: "POST", headers: { "Content-Type": "application/json", "x-api-key": process.env.PARALLEL_API_KEY!, }, body: JSON.stringify({ state: { company: "Clay", website: "https://www.clay.com", }, questions: { has_cli: { type: "noul", instructions: "Does Clay offer an official command-line interface (CLI)?", }, }, }), }); if (!response.ok) { throw new Error(`Decision failed: ${response.status} ${await response.text()}`); } const decision = await response.json(); const answer = decision.answers.has_cli; if (answer.type === "refusal") { console.log("No decision returned; review this question."); } else { console.log("Estimated probability:", answer.noul); } for (const citation of decision.citations) { console.log(citation.url); } } main().catch(console.error); - lang: cURL source: |- curl https://api.parallel.ai/v1/decisions \ -H "Content-Type: application/json" \ -H "x-api-key: $PARALLEL_API_KEY" \ -d '{ "state": { "company": "Clay", "website": "https://www.clay.com" }, "questions": { "has_cli": { "type": "noul", "instructions": "Does Clay offer an official command-line interface (CLI)?" } } }' - lang: Java source: |- import java.net.URI; import java.net.http.HttpClient; import java.net.http.HttpRequest; import java.net.http.HttpResponse; import java.time.Duration; public class DecisionsExample { public static void main(String[] args) throws Exception { String body = "{" + "\"state\":{\"company\":\"Clay\"," + "\"website\":\"https://www.clay.com\"}," + "\"questions\":{\"has_cli\":{\"type\":\"noul\"," + "\"instructions\":\"Does Clay offer an official command-line interface (CLI)?\"}}}"; HttpRequest request = HttpRequest.newBuilder() .uri(URI.create("https://api.parallel.ai/v1/decisions")) .timeout(Duration.ofSeconds(30)) .header("Content-Type", "application/json") .header("x-api-key", System.getenv("PARALLEL_API_KEY")) .POST(HttpRequest.BodyPublishers.ofString(body)) .build(); HttpResponse response = HttpClient.newHttpClient() .send(request, HttpResponse.BodyHandlers.ofString()); if (response.statusCode() >= 400) { throw new RuntimeException("Decision failed: " + response.statusCode() + " " + response.body()); } System.out.println(response.body()); } } components: schemas: GroundedDecisionRequest: properties: mode: type: string const: fast title: Mode default: fast description: Decision execution mode. state: $ref: '#/components/schemas/JsonValue' description: >- Context shared by all questions. Accepts any JSON value. Together with the largest question, this may use at most 32,000 estimated input tokens. questions: type: object maxProperties: 100 title: Questions description: >- 1–100 questions keyed by names containing only ASCII letters, numbers, underscores, and hyphens. State plus the largest question may use at most 32,000 estimated input tokens. Each value is a predicate (`noul`), choice, or score question. patternProperties: ^[A-Za-z0-9_-]+$: oneOf: - $ref: '#/components/schemas/GroundedChoiceQuestion' - $ref: '#/components/schemas/GroundedScoreQuestion' - $ref: '#/components/schemas/GroundedNoulQuestion' discriminator: propertyName: type mapping: choice: '#/components/schemas/GroundedChoiceQuestion' noul: '#/components/schemas/GroundedNoulQuestion' score: '#/components/schemas/GroundedScoreQuestion' minProperties: 1 include_excerpts: type: boolean title: Include Excerpts description: >- When false, every citation has an empty excerpts array. Use for token efficiency if LLM consuming downstream results. default: true type: object required: - state - questions title: Decision request additionalProperties: false GroundedDecisionResponse: properties: mode: type: string const: fast title: Mode answers: additionalProperties: oneOf: - $ref: '#/components/schemas/GroundedChoiceAnswer' - $ref: '#/components/schemas/GroundedScoreAnswer' - $ref: '#/components/schemas/GroundedNoulAnswer' - $ref: '#/components/schemas/GroundedRefusalAnswer' discriminator: propertyName: type mapping: choice: '#/components/schemas/GroundedChoiceAnswer' noul: '#/components/schemas/GroundedNoulAnswer' refusal: '#/components/schemas/GroundedRefusalAnswer' score: '#/components/schemas/GroundedScoreAnswer' type: object title: Answers description: >- Answers keyed by the question names from the request. Any question can return a refusal. usage: $ref: '#/components/schemas/GroundedDecisionUsage' citations: items: $ref: '#/components/schemas/GroundedDecisionCitation' type: array title: Citations description: >- Web sources included in the shared decision context. Applies to the whole request, without a per-answer source mapping. Can be empty. decision_id: description: >- Decision identifier: `decisions_` followed by a UUID without hyphens. examples: - decisions_cad0a6d2dec046bd95ae900527d880e7 title: Decision Id type: string type: object required: - mode - answers - usage - citations - decision_id title: Decision response ErrorResponse: properties: type: type: string const: error title: Type description: Always 'error'. error: $ref: '#/components/schemas/Error' description: Error. type: object required: - type - error title: ErrorResponse description: Response object used for non-200 status codes. JsonValue: {} GroundedChoiceAnswer: properties: type: type: string const: choice title: Type default: choice choice: type: string title: Choice description: Selected option name from the question criteria. probabilities: additionalProperties: type: number type: object title: Probabilities description: Estimated probability for each option, keyed by option name. confidence: anyOf: - type: number - type: 'null' title: Confidence description: Model-reported confidence, when available. type: object required: - choice - probabilities title: Choice GroundedScoreAnswer: properties: type: type: string const: score title: Type default: score score: type: number title: Score description: >- Probability-weighted mean of the zero-based level indices. Can be fractional. legend: additionalProperties: type: string type: object title: Legend description: >- Maps string indices to the level descriptions from the question criteria. probabilities: additionalProperties: type: number type: object title: Probabilities description: Estimated probability for each level, keyed by its string index. confidence: anyOf: - type: number - type: 'null' title: Confidence description: Model-reported confidence, when available. type: object required: - score - legend - probabilities title: Score GroundedNoulAnswer: properties: type: type: string const: noul title: Type default: noul noul: type: number title: Noul description: Estimated probability that the condition is true, from 0 to 1. type: object required: - noul title: Predicate GroundedRefusalAnswer: properties: type: type: string const: refusal title: Type default: refusal description: >- The downstream decision provider refused this question; no value or probabilities are returned. type: object title: Refusal GroundedDecisionUsage: properties: input_tokens: title: Input Tokens type: integer output_tokens: type: integer title: Output Tokens description: Always 0 for compatibility with the upstream contract. const: 0 type: object required: - input_tokens - output_tokens title: Usage GroundedDecisionCitation: properties: url: type: string title: Url description: Source page URL. title: anyOf: - type: string - type: 'null' title: Title description: Page title. Omitted when unavailable. excerpts: items: type: string type: array title: Excerpts description: >- Passages used to ground the decision. Empty when include_excerpts is false. type: object required: - url - excerpts title: Citation Error: properties: ref_id: type: string title: Reference ID description: Reference ID for the error. message: type: string title: Message description: Human-readable message. detail: anyOf: - additionalProperties: true type: object - type: 'null' title: Detail description: Optional detail supporting the error. type: object required: - ref_id - message title: Error description: An error message. securitySchemes: ApiKeyAuth: type: apiKey in: header name: x-api-key ````