Skip to main content
Dated release history. For current parameters, availability and rates, use the API reference, data coverage and pricing.
​
  • ChemRxiv is now full text for commercial use: valyu/valyu-chemrxiv holds 14,500+ full-text ChemRxiv preprints from 2017 to the present, limited to papers licensed CC BY 4.0 so every result is cleared for commercial use with attribution. Papers under non-commercial licences (CC BY-NC, CC BY-NC-ND) are excluded. Results include the title, authors, abstract, publication date, DOI, full-text passages and figures, with links in the form https://chemrxiv.org/doi/full/<doi>. The dataset updates monthly. Pass included_sources=["valyu/valyu-chemrxiv"] to the Search API or DeepResearch.
​
  • Rate limits now scale with your plan. Search, Contents, Answer, and creating or polling DeepResearch tasks each have their own limit, lower on cheaper plans and higher on paid plans. There is no daily quota. A request over the limit returns 429 Too Many Requests with a Retry-After header. The body names the limit you hit, your plan, and how to raise it. See rate limits.
  • DeepResearch concurrency limits. Your plan caps how many of your DeepResearch tasks run at once, across tasks, batches and tasks added to a batch. A task over the cap is still created and still holds its credits. It waits as queued and starts automatically when one of your running tasks finishes. See concurrency limits.
​
  • DeepResearch v2: A new research engine on the same endpoint, POST /v1/deepresearch/tasks. Add deliverables and the files come back with the cited report: xlsx, pptx, docx, pdf or csv, up to 10 files a task, in every mode except instant. A bare format works: "deliverables": ["xlsx"]. The first file is included; each extra file uses 16 credits. See the example reports.
  • Passage-level citations: Sources can carry locations, the page and the verbatim excerpt the report cited, so your app can show the line a claim rests on. See citations and passages.
  • Instant mode: mode: "instant" returns a sourced answer in under 30 seconds for 8 credits. It doesn’t return files.
  • Parallel researchers: Heavy and max split a big question into sub-questions and run up to four researchers in parallel, then one writer joins their cited notes.
  • Pin a run to a date: search.reference_date pins research to what was published by that date, for backtests. Dates from January 2026 are supported.
  • Upgrading: Older SDKs keep receiving v1 response shapes. The released SDKs don’t expose locations or instant yet; call the status endpoint directly for them.
  • Wiley HLS: Eligible startups can search the full text of more than one million peer-reviewed articles from 900+ Wiley health and life sciences journals, figures and equations included, pay-per-use. Pass included_sources=["wiley/wiley-hls"] to Search or DeepResearch. See the Wiley HLS guide.
  • New usage prices: Web and news results now use 0.1 credits each, and arXiv, PubMed, bioRxiv and medRxiv results 0.05 credits. Contents uses 0.2 credits per extracted URL, plus 0.2 for summaries or structured extraction. DeepResearch tasks use 25 credits for fast, 100 for standard, 500 for heavy and 2,500 for max. Licensed dataset prices are unchanged. See pricing.
  • Strict source scope: Set scope_policy: "strict" on POST /v1/search to search exactly the datasets in included_sources. The response adds a source_scope object listing which datasets were selected, returned and omitted. Without it, a request that lists several datasets may search only the most relevant of them.
  • Search filters: excluded_sources accepts preset names such as finance or academic, and site: and -site: in a query now filter every result, from the web and from datasets.
  • DeepResearch credit holds: Creating a task reserves its base price until the task is charged. If your available balance cannot cover it, the request returns 402 with the required and available amounts; if billing is briefly unavailable it returns 503 and creates nothing. Adding tasks to a batch reserves them all at once. See billing and credit holds.
  • New datasets: FDA drug recalls (valyu/valyu-openfda-drug-recalls) and device recalls (valyu/valyu-openfda-device-recalls), on the Serious Business plan and in the health preset.
  • Answer billing: Each search result is billed once per answer, even if several tool calls return it, and cost.total_deduction_dollars is exactly the amount charged.
​
  • Semantic datasource search: GET /v1/datasources/search?q= finds datasets from a natural-language description - Wiley medical literature, FDA drug labels, insider transactions, CPI inflation - ranked by embedding similarity blended with exact-term matching, so terse acronyms and descriptive phrases both work. Returns the dataset id values to pass as included_sources on the Search API, with a relevance_score per result. A query that matches nothing returns an empty list rather than a weak guess.
​
  • Agents That Provision Themselves: A human still creates the Valyu account, but from there an agent can provision itself end to end - request a login from the CLI, mint its own API keys with a hard spending cap, check balance, top up the account balance if a card is on file, and rotate or revoke keys, all programmatically. valyu login runs a device-style flow: you get a short code, approve it in the browser, and the key is minted and stored locally. It also runs fully headless by streaming the auth events, so a cloud agent can authenticate without a browser.
  • Account management from the CLI: New CLI commands manage your account end to end. Every command speaks JSON for scripts and prints human-friendly output by default. Errors return a machine-readable code and a recovery hint, so an agent can read the failure and self-correct instead of stalling. Management keys carry explicit scopes for least-privilege access, while data and search access is implicit on every key - so any key can run queries without an extra scope.
​
  • New SEC filing types: The valyu/valyu-sec-filings dataset now covers nine additional filing types, searchable in natural language through the Search API - pin the dataset with included_sources=["valyu/valyu-sec-filings"] or let Valyu route automatically. Also available via the Answer API, DeepResearch, and the Python and JavaScript SDKs:
    • S-1: IPO and securities registration statements - query a company’s pre-IPO business model, financials, risk factors, and use of proceeds.
    • S-4: M&A and exchange-offer registration statements - pull deal terms, exchange ratios, fairness opinions, and pro forma financials for combined companies.
    • Form 10 (10-12B / 10-12G): Spin-off and Section 12 new-listing registration statements - often the first detailed disclosure from a newly independent business.
    • DEF 14A (+ DEFA14A / DEFM14A / DEFR14A family): Definitive proxy statements - search executive compensation, say-on-pay votes, board nominees, governance matters, and shareholder proposals.
    • SC TO-T: Third-party tender offer statements - retrieve offer prices, conditions, financing, and expiration terms.
    • SC 14D9: Target-board responses to tender offers - recommendations to shareholders, fairness opinions, golden-parachute compensation tables, and deal background.
    • 6-K: Foreign private issuer current reports - interim earnings, dividends, and material events reported in home markets.
    • 20-F: Foreign private issuer annual reports (the FPI equivalent of a 10-K) - audited financials, risk factors, and operating reviews for non-US companies on US exchanges.
    • N-PORT (NPORT-P): Monthly mutual fund / ETF portfolio holdings reports - fund-level positions, asset allocations, and risk metrics.
  • New dataset - ChemRxiv: Added valyu/valyu-chemrxiv, 30K+ chemistry preprints from ChemRxiv spanning organic, inorganic, physical, and materials chemistry. Query it through the Search API with included_sources=["valyu/valyu-chemrxiv"], or via the Python and JavaScript SDKs.
  • Workflows (Beta): You can author your own org-scoped templates as workflows: save a prompt, sources, deliverables, and mode once with typed {variables} via POST /v1/workflows, publish new versions with PATCH, and share them across your organization. Browse 44 Valyu-curated templates that are now live for templated DeepResearch. Pass workflow_id + workflow_params to POST /v1/deepresearch/tasks to run one, or use them via the deepresearch create methods in the SDKs. Get started with the Workflows quickstart.
​
  • Workflows (Beta): Templated, versioned DeepResearch you can run from the API. Pass workflow_id + workflow_params to POST /v1/deepresearch/tasks, or manage workflows via the new Workflows endpoints. See the Workflows guide.
  • Subscription plans in docs: Documented the subscription plans of the time (Pay-as-you-go, $29, $89, $449, and Enterprise) and which data sources each includes. See Pricing & plans.
  • Expanded data source coverage: Documented additional specialised sources across genomics, cybersecurity, compliance & sanctions, economics, transportation, automotive, environment, and pulse signals. See Data coverage and the per-vertical use cases.
  • AI-native onboarding: A curated llms.txt and AGENTS.md, expanded contextual menu, hidden agent hints, and copy-paste <Prompt> blocks across the API, SDK, and integration docs.
​
  • Valyu CLI upgrade: The Valyu CLI now includes access to the most powerful Deep Research features: deliverables, human-in-the-loop, code execution, and screenshots.
  • Source biases: New sourceBiases parameter in the Search & DeepResearch API lets you fine-tune how search results are prioritized across different data sources, giving you more control over result ranking.
​
  • Deep Research screenshots: Deep Research can now capture website screenshots and include them directly in generated reports for richer visual context.
  • Deep Research code execution: The Deep Research API now supports code execution, enabling reports with computed results, data analysis, and programmatic outputs.
  • Deep Research HITL (human-in-the-loop): Added human-in-the-loop support for Deep Research, allowing you to review, guide, and refine research tasks during execution.
  • Valyu CLI: Shipped the official Valyu CLI for running searches, deepresearch workflows, and interacting with Valyu directly from your terminal.
​
  • 13F/G/D SEC filing search: You can now search across SEC 13F, 13G, and 13D filings through the Search API, adding the institutional ownership layer of public markets.
  • Async Contents API: Added async support for the Contents API, enabling non-blocking content extraction for large-scale workflows. Available in both Python and JavaScript SDKs.
  • Source biases: New sourceBiases parameter in the Search API lets you fine-tune how search results are prioritized across different data sources, giving you more control over result ranking.
​
  • Bulk batch results: New batch.getResults() method lets you fetch all results from a batch in one paginated call instead of individual requests per task. Ideal for teams running hundreds of DeepResearch tasks in parallel.
  • DeepResearch email alerts: Added alert_email support for custom email notifications on task completion. A new custom_url field lets you pass a custom report URL instead of the default platform.valyu.ai URL.
All changes above are live in the Python and JavaScript SDKs.
  • Postman-ready OpenAPI spec: You can now easily import Valyu OpenAPI spec directly into Postman to test all endpoints interactively.
​
  • Deep Research document uploads: You can now attach and upload internal documents as sources for Deep Research reports via the Platform or API. Uploaded documents are properly cited in the final report.
  • New “max” Deep Research mode: Exhaustive research with maximum quality, fact verification, and comprehensive analysis. Runs for up to ~180 minutes. Now available in TypeScript, Python SDKs and n8n integration.
  • Breaking change: Removed total_deduction_pcm from the Search API response. Users relying on this field should update to total_deduction_dollars.
  • Deep Research strategy controls: The strategy field in the Deep Research API now allows fine-grained control over the writing style of generated reports.
​
  • Valyu n8n integration: Updated Valyu’s Base URL from api.valyu.network to api.valyu.ai/v1 in the native Valyu integration for n8n.
  • Contents API improvements: Improved support for additional file type fetching and sitemap extraction.
  • Datasources API: Added fulltext search capabilities to our datasources search list API.
  • Economics search quality: Improved search accuracy for economics data across FRED and BLS sources.
  • Faster Deep Research: Fast mode for Deep Research now runs with significantly reduced latency and improved output quality.
  • Faster web search: Enhanced web search performance with optional fast mode for quicker results with shorter content snippets.
  • AI SDK fast mode: Added fastMode configuration option to the webSearch tool in the Vercel AI SDK. When enabled, returns faster responses with shorter content snippets, ideal for general-purpose queries where speed is preferred over detail.
  • Agno integration: Added Agno framework integration documentation with 8 specialised search tools.
​
  • Country code filtering: Added country_code support to the DeepResearch API for filtering search results by geography. Both JavaScript and Python SDKs updated.
  • Search relevance improvements: Made some improvements in the Search API for more accurate and contextually relevant results.
​
  • New Healthcare datasources: Added ChEMBL (2.5M+ compounds), PubChem (110M+ compounds), Open Targets (60K+ targets), NPI Registry (8M+ healthcare providers), and WHO ICD-10/11 codes (70K+ codes) to the search API.
  • Performance improvements: Improved DeepResearch quality, Batch API reliability, Polymarket data search accuracy, and faster content extraction.
  • Datasources SDK methods: Added datasources() and datasources_categories() methods with category filtering to both JavaScript and Python SDKs.
  • AI SDK Tools: Extended the bioSearch tool to include the new healthcare/life sciences datasources (ChEMBL, PubChem, Open Targets, NPI Registry, WHO ICD).
  • MCP: Extended valyu_bio_search MCP tool with the new healthcare datasources.
  • AI SDK size presets: Control content size per search result with short, medium, large, or max presets across all 7 AI SDK search tools.
​
  • New data sources: Added UK rail schedules, legislation, case law, ship tracking, and parliamentary data to expand coverage across transportation, legal, and government domains.
  • Batch API for Deep Research: Process multiple Deep Research tasks efficiently with the new Batch API, including status tracking and progress monitoring for large-scale research operations.
  • Page screenshot capture: The Content API now supports capturing page screenshots for visual content analysis tasks.
  • Enhanced search filtering: New filtering options including date range filtering, source exclusion lists, and category filters give you more control over search results.
  • Model renamed: The “lite” model has been deprecated in favor of “fast” for clearer naming consistency.
  • Datasources API: New tool manifest endpoint, /datasources, that enables AI agents to discover available data sources with metadata, schemas, and pricing information for the Search API.
  • Ship tracking enhancements: Added vessel type filtering (tanker, cargo, cruise, yacht, military, etc.) and natural language search via AIS for global ship tracking.
  • Polymarket historical data: Added historical pricing data from Polymarket prediction markets.
  • Integration guide updates: The AI SDK and MCP integration guides now include datasources tools for discovering available data sources.
  • MCP documentation: Added missing tools documentation for patents, biomedical research, and economics data to the MCP server docs.
  • Deep Research batch quickstart: New quickstart guide for Deep Research batch operations with clearer documentation.
​
  • Deep Research deliverables (beta): Deep Research can now generate additional files alongside your final report across CSV, Excel, PowerPoint, Word, and PDF.
  • TOON output support: Deep Research now supports TOON-formatted output for structured, machine-readable deliverables.
  • Mode renamed: The “Lite” Deep Research mode has been renamed to “Standard” to better reflect its balanced speed and depth.
  • Clearer reasoning summaries: The reasoning trace now includes more concise, readable summaries so you can quickly understand how Deep Research arrived at its conclusions.
  • Claude Agent SDK integration: New guide for building AI agents with Valyu’s search capabilities using the Claude Agent SDK.
  • SDK updates: All SDKs (Python, JavaScript, Rust) have been updated to reflect the latest API changes, including new export formats and TOON output support.
​
  • Breaking change: The /v1/deepresearch/tasks/{id}/update endpoint now enforces phase-aware validation. This means that follow up instructions are rejected once the writing phase begins. Update your integrations to submit all follow-up instructions during the active research phase, before report generation starts.
​
  • We launched Valyu AgentCore to bring Valyu’s search capabilities to AWS Bedrock AgentCore and Strands Agents. Check out our co-founders Harvey and Hirsh on the AWS Let’s Build A Startup Podcast w/ the legend Giuseppe Battista showing what it can do 👉 Podcast link.
  • We shipped a new Vercel AI SDK guide showing how to use Valyu’s AI SDK tools for academic research.
  • We’ve added biomedical and economic search tools: valyu_bio_search, valyu_economics_search to the Valyu MCP server.
  • We shipped a new “Powered by Valyu” badge. Users can now show off their Valyu-powered apps with this badge.
  • We added deepresearch API support to the Rust SDK.
  • We’ve migrated the v1/deepsearch API endpoint to v1/search. We recommend updating your API calls accordingly. The v1/deepsearch endpoint will remain supported until April 30, 2026.
  • The deepresearch API now support search configs (included_sources, search_type).
  • Deep Research can now generate detailed charts (line, bar, horizontal bar, area, pie, doughnut, radar, scatter).
  • Deep Research PDF generation has been improved, displaying charts/tables as well.
  • The new deepresearch modes - Fast, Standard and Heavy are now supported in the JS & Python SDKs.
​
  • We shipped a better numbered citation system for both Search and Deep Research results.
  • Our Content API now automatically extracts content from PDF URLs.
  • We’ve now improved token/context handling for the Answers API endpoint for more predictable performance.
  • We enhanced Deep Research system prompts for better research quality.
  • We’ve added Kalshi and Polymarket prediction markets as a data source so you can search over them in natural language.
  • We made some improvements on API key propagation in our infrastructure.
  • We made two big changes to our N8N Integration. 1. We added the Answers and DeepResearch API to the N8N node. 2. “Search” is now “Search”, “Contents” is now “Extraction”.
  • We’ve now added complete documentation for the Rust SDK with examples for Search, Contents, and Answers endpoints.
  • We shipped complete documentation for LangChain-JS integration with code examples.
  • We’ve also added a Rust SDK documentation.
  • We now have an examples page showcasing real world apps built on Valyu.
​
  • We’ve localized deepresearch summaries to the user’s language preference.
  • We made improvements to the deepresearch report format.
  • We improved the German labor statistics search quality.
  • We enhanced Deep Research prompts for research and writing phases with better context passing.
  • We updated the /listtasks endpoint to list for Deep Research.
  • We added the enterprise API usage plan.
  • We improved the contents API to have more concise AI summaries.
  • We fixed structured output generation for Deep Research reports.
  • We updated the answers endpoint response to use tx_id instead of ai_tx_id and also added a extraction_metadata type.
  • We added source filtering support to the AI SDK web Search tool.
  • We updated transport type handling and dependencies for Valyu remote mcp.
  • We added a complete implementation guide for Deep Research with SDK examples.
​
  • We implemented streaming support for real-time content delivery for the answer endpoint.
  • We improved response generation for the answer endpoint.
  • We improved citation structure in answer endpoint responses.
  • We fixed missing titles in contents API with high extract effort setting.
  • We added new financial datasets - Commodities, ETFs, and funds datasets.
  • We improved content extraction with image URLs support.
  • We added streaming support for the answer endpoint in the JS & Python SDKs.
  • We added structured output format for Deep Research.
  • We added full DeepResearch API support with task management to the JS & Python SDK.
  • We updated API reference across Python and JavaScript SDK docs for answer endpoint.
  • We updated pricing documentation to reflect new datasets.
​
  • Support for up to 100 search results (upon request) in the JS & Python SDKs.
  • Added news search type for dedicated news searches in the JS & Python SDKs.
  • Added urlOnly search type to return shortened snippets only in the JS SDK.
  • Added url_only parameter to return shortened snippets only in Python SDK.
  • publication_date field now available in search results in the Python SDK.
  • relevance_score field now optional in fast_mode or url_only searches.
  • Added docs for news search, url_only, and expanded result limits.
  • Added docs for the updated API reference with the new endpoints and parameters.
  • max_num_results now supports 1 - 100 results. Standard API keys support 1-20 results while higher limits require requesting a special API key through the platform dashboard.
​
  • We enhanced response formatting for the answer endpoint.
  • We fixed handling of concurrent citation contexts in research results.
  • We updated the JS SDK name from valyu to valyu-js.
​
  • Bug fixes.