# SpecStream

> The live web data engine for teams and agents. Ask for any dataset in plain words, and SpecStream builds it from the entire live web: every match found, extracted and enriched into structured records, then monitored for change. Each request is priced before any work runs. People and agents share the same workspace, permissions and balance.

## When to use SpecStream

Use SpecStream when a person or an agent needs data about real things that meet stated conditions: companies, local businesses, grants, tenders, clinical trials, papers, datasets, regulations, vulnerabilities, apps, repositories, packages, podcasts or books. A search API returns pages to read, and a lead database returns what it collected months ago. SpecStream reads the live web for the request and returns each match with a source URL and read time on every value, and charges 5¢, 15¢ or $1 for each match, at a price shown before any work runs. Results go to Google Sheets, Excel, Notion, Airtable, monday.com, HubSpot, Salesforce, Pipedrive, Attio, Outlook drafts or a signed webhook, and a saved request can run on a schedule and report what changed.

Jobs it does well, each with a request to run:

- Grants & tenders: open grants and tenders that fit a client, with the deadline, the value and the official notice: https://specstream.ai/for/grants-and-tenders
- Nonprofits by cause: US nonprofits by cause, state and revenue, from their Form 990 returns, with assets and officers: https://specstream.ai/for/nonprofits
- Companies that use a tool: companies that use a given tool, seen on their own website and DNS records: https://specstream.ai/for/companies-that-use-a-tool
- Investors & deal flow: funding filings (US Form D, UK share allotments) and new executives (SEC 8-K), delivered weekly: https://specstream.ai/for/funding-and-new-executives
- Sales & agencies: companies that fit what you sell to, with the contacts they published, sent to your CRM: https://specstream.ai/for/sales-leads
- Local businesses: every shop, clinic and trade that fits, with address, phone and website, on a map: https://specstream.ai/for/local-businesses
- Research: papers, clinical trials, regulations and datasets: https://specstream.ai/for/research
- Creators: creators by niche and follower range, with the contact they published: https://specstream.ai/for/creators
- AI agents: data your agent can act on, at a price it agreed to: https://specstream.ai/for/ai-agents
- New matches every week: the same request on a schedule, with the new matches and what changed: https://specstream.ai/for/new-matches-weekly
- Map your spreadsheet: up to 20,000 rows from CSV, Excel, Google Sheets or Airtable on a map: https://specstream.ai/for/map-your-spreadsheet
- Signals & feeds: kept-current feeds of funding rounds (SEC Form D), new executives (8-K filings), layoffs (WARN notices), UK share allotments, failed banks, tenders and grants, by API, webhook, RSS or CSV: https://specstream.ai/signals
- Ready requests to open, edit and price before running: https://specstream.ai/templates

Where it fits, next to what people use today:

- Next to a database: the data is read when you ask, so the small, local and new businesses are in it, and you pay per match, never per seat.
- Next to stitching tools together (a list, an enrichment table, an email finder): you get the finished data, only the rows that fit, ready for your sheet or CRM.
- Next to asking a chat assistant: up to 10,000 rows in one request, and it stays up to date on a schedule.
- Next to a search API, for agents: rows your agent can act on, with their sources, at a price it agreed to first.
- What's new, dated: https://specstream.ai/changelog

## Connect an agent

- MCP endpoint: https://api.specstream.ai/mcp (Streamable HTTP)
- Authentication: a SpecStream API key as a Bearer token, or a registered OAuth client with read, prepare, or execute access
- SpecStream for agents: https://specstream.ai/agents
- Connection details: https://specstream.ai/docs/mcp
- AI assistants (setup for every client below): https://specstream.ai/docs/mcp/clients
- Claude: https://specstream.ai/docs/mcp/claude
- ChatGPT: https://specstream.ai/docs/mcp/chatgpt
- Cursor: https://specstream.ai/docs/mcp/cursor
- VS Code: https://specstream.ai/docs/mcp/vscode
- Perplexity: https://specstream.ai/docs/mcp/perplexity
- Gemini: https://specstream.ai/docs/mcp/gemini
- Grok: https://specstream.ai/docs/mcp/grok
- Le Chat (Mistral): https://specstream.ai/docs/mcp/le-chat
- Manus: https://specstream.ai/docs/mcp/manus
- LM Studio: https://specstream.ai/docs/mcp/lm-studio
- Cherry Studio: https://specstream.ai/docs/mcp/cherry-studio
- AnythingLLM: https://specstream.ai/docs/mcp/anythingllm
- Agents pay per result (no account; x402 or MPP): https://specstream.ai/docs/agent-payments
- Agent frameworks (OpenAI API, LangChain, LlamaIndex, CrewAI, Vercel AI SDK): https://specstream.ai/docs/frameworks
- Every integration, by kind: https://specstream.ai/integrations
- A2A agent: https://api.specstream.ai/a2a (A2A 1.0 and 0.3 JSON-RPC). Agent Card: https://api.specstream.ai/.well-known/agent-card.json. Without credentials it reads the request back and prices it; with an API key or OAuth token it runs the maximum you approve and returns the rows with their sources.
- REST API: https://specstream.ai/docs/api
- SDKs & CLI, Python and TypeScript clients and a command line: https://specstream.ai/docs/sdk
- Webhooks, signed calls when a find finishes, a monitor spots something new or records are added to a collection: https://specstream.ai/docs/webhooks
- Desktop & phone apps on the same account and balance (the Windows app is out; Mac, iPhone and Android are coming soon): https://specstream.ai/apps
- Browser extension (Chrome and Firefox are out; Edge and Opera are coming soon): https://specstream.ai/apps#browser

## What it can find

Local businesses, companies and organizations, people, creators, storefront products, app listings, job postings, scholarly papers, tenders, grants, software repositories, software packages, datasets, clinical trials, regulatory documents, vulnerabilities, books, podcasts, web pages and documents (including PDF, Word, Excel, PowerPoint, EPUB, and OpenDocument files, and scanned pages read with OCR), and encyclopedia entries from Wikidata and Wikipedia. Available fields and sources vary by record type and request. Each record type has a page with example requests, the details you can ask for, and the sources (What you can ask for: https://specstream.ai/things-worth-finding):

- Local businesses: https://specstream.ai/things-worth-finding/local-businesses
- Companies: https://specstream.ai/things-worth-finding/companies
- People: https://specstream.ai/things-worth-finding/people
- Creators: https://specstream.ai/things-worth-finding/creators
- App store listings: https://specstream.ai/things-worth-finding/apps
- Job postings: https://specstream.ai/things-worth-finding/job-postings
- Papers: https://specstream.ai/things-worth-finding/papers
- Tenders: https://specstream.ai/things-worth-finding/tenders
- Grants: https://specstream.ai/things-worth-finding/grants
- Repositories: https://specstream.ai/things-worth-finding/repositories
- Packages: https://specstream.ai/things-worth-finding/packages
- Datasets: https://specstream.ai/things-worth-finding/datasets
- Clinical trials: https://specstream.ai/things-worth-finding/clinical-trials
- Regulations: https://specstream.ai/things-worth-finding/regulations
- Vulnerabilities: https://specstream.ai/things-worth-finding/vulnerabilities
- Books: https://specstream.ai/things-worth-finding/books
- Podcasts: https://specstream.ai/things-worth-finding/podcasts
- Pages and documents: https://specstream.ai/things-worth-finding/pages-and-documents
- Encyclopedia entries: https://specstream.ai/things-worth-finding/encyclopedia-entries

Where SpecStream looks, by kind of source and record type: https://specstream.ai/docs/sources

## How a request works

1. Describe what to find in everyday language: place, quantity, conditions and exclusions, order, and the details to return. Conditions support thresholds and AND, OR, and NOT. Returned details can be required or optional. A request asks for up to 10,000 results (25 when it names no number) and can be up to 16,000 characters; a run returns the matches it finds, which can be fewer, and the largest runs take up to four hours. Requests can be written in any language; names, quoted literals and URLs are kept as data, and sources in other languages are read.
2. The request is read into editable parts (the `compile` tool). Ambiguous or unsupported details stay visible instead of being dropped. `spec_options` lists the supported fields.
3. Get a quote for the exact request revision and operation (`quote_capacity_work`). Quotes are valid for 30 minutes.
4. Authorize with the quote ID and hash, an approved maximum in cents (`confirm_max_cents`) and a stable request key (`authorize_capacity_work`). The maximum is reserved before work starts.
5. A result counts as a match when its source page supports every required detail. Each value keeps its source and the time it was read. Missing, conflicting, and blocked information stays marked as such.

## Where the data comes from

- Reading your request: a model reads the request in any wording or language and maps its exact words to conditions. Negation, alternatives and exclusions carry through. Uncertain readings stay visible and can't authorize work.
- Sources: each candidate is read on its own pages first, then on other sources that name it, such as filings, listings, profiles and news. Archived copies, such as Common Crawl, are labeled historical with their capture date.
- Values: every value keeps its source URL and the time it was read. Numbers, dates, places and follower counts are the values the pages publish, and every quote is confirmed on the page it cites.
- Contacts: emails and phone numbers come only from pages their owners published.
- One result per thing: records are matched on stable identifiers, so one record found through several routes is one result, charged once.
- Source and read time through the API: `get_find_decisions` and `GET /v1/finds/{id}/decisions` return each value with its reason, source and time. `replay_find_decisions` and `GET /v1/finds/{id}/decisions/replay` re-read the saved record without new searches or model calls, for $0.
- Amounts keep the currency they were published in; nothing is converted.
- No model is fine-tuned on customer requests, lists or results.
- Purchases are idempotent: authorization carries a stable request key, and an uncertain response is recovered instead of bought again. Running work can be stopped, keeping the matches already found. Progress and results stream while work runs.

## Money rules

- Saving a request or adding balance never starts work.
- Matching results are charged at their rate, with the additions below, never more than the approved maximum. What a run doesn't use goes back to the balance automatically. A find with no match costs nothing; checking your own imported list bills the rows it checked, matched or not.
- Rates per matching result, set by the job each finished row replaces: Standard $0.05 (up to 5 facts), Expanded $0.15 (up to 15 facts and one related list), Deep $1.00 (anything beyond). Each worded condition or detail counts as a fact. A person’s email you asked for adds 5¢ to the row and a phone 15¢, on any list; a likely email built from the company’s pattern counts and is labeled likely. A company left out costs nothing. Ten related records a result are included, then 10¢ for each further ten. Checking your own imported list is 5¢ for each row checked; a row that matches pays its rate instead; importing is free. A refresh or monitor recheck charges half the rate, rounded up, for rows that changed and new rows: 3¢, 8¢ and 50¢ (a contact adds half: 3¢ email, 8¢ phone). Each row that comes back unchanged or no longer matches is 1¢. The quote says which applies.
- Imported addresses placed on the map: free, each at its postcode.
- Retries and backup sources don't change the price.
- Reading, organizing, comparing, exporting, and sharing owned work costs nothing, and so does replaying a saved find.
- Free accounts get $10 of work once, for the first 30 days, with 1 member and 1 list kept up to date. Every plan turns its full price into work, and unused monthly balance rolls over for one more month. Pro, the recommended plan, is $19 a month with $19 of work: 1 member and 10 lists kept up to date. Scale is $99 a month with $99 of work: 10 members and 50 lists kept up to date. Team is $499 a month with $499 of work: 50 members and 250 lists kept up to date. No plan limits how many finds run at once. Enterprise starts at $2,500 a month with unlimited members, SAML single sign-on, SCIM and audit log export. Every other feature is on every plan, Free included: schedules and their alerts, sheets and CRMs kept in sync, webhooks, Zapier, Make and n8n, Slack and Teams, maps and routes, exports, the API, MCP and every app. Top-ups start at $10 and don't expire. Done for you, where SpecStream writes and runs your recurring lists into your tools, is priced by quote.
- Keys and delegated access can carry spending ceilings, per-run and monthly limits, and operation scopes. Stay inside them.

## After the results arrive

- Collections: table, list, map, compare, and history views; saved views; snapshots; exports to CSV, Excel, JSON, and HTML, plus KML and GeoJSON for records with coordinates; share links created and copied in one step, with expiry and revocation. Maps color pins by match status or by a column, with a legend. Several collections combine into one, keeping each place or company once. Selected places become a route of up to 11 stops, opened as Google Maps directions. The Columns menu shows how many records have each detail and fills the missing ones through Add details, after a quote. People add their own fields (Status, Checkbox, Text) to track a list; SpecStream never changes them, exports include them, and share links leave them out. A share link can show the request that found the list with "Run this yourself", and its map can be embedded on another site.
- Import: CSV or Excel files, or a link to a Google Sheet, an Excel workbook in OneDrive or SharePoint, or an Airtable table or view. Up to 20,000 rows and 20 MB, extra columns kept. Linked lists can be checked daily for new and changed rows. Imported values stay labeled as yours, and importing doesn't start paid work.
- Follow-up work: find more results (excluding ones already owned), add details to results you already have (the dialog suggests details for places and companies and shows how the request reads), or recheck them.
- Activity: runs with their outcome and charge, imports, snapshots, monitors, and records sent to tools. Each collection links the request that found it and the monitors that recheck it; the Monitoring list shows each monitor's last run. When a recheck changes a detail, the collection lists each change with the value before, the value now, and where it was seen.
- Schedules repeat a saved operation with a cadence, timezone, per-run limit, and monthly limit. Missing funds or permission blocks an occurrence instead of creating debt. New and changed records from a run can notify the owner, post in Slack, go to a connected tool, or call a webhook (`monitor.found`). A found list without a schedule offers "Watch for new ones", a weekly schedule that looks for more results; only new matches are charged.
- Projects: owner, editor, and viewer roles; member spending limits; notes with email mentions.
- Destinations: Google Sheets, Microsoft Excel, Airtable, Notion and monday.com; HubSpot, Salesforce, Pipedrive and Attio as companies, contacts, accounts, leads, organizations or people, matched on domain or website to update existing records; Outlook drafts; signed webhooks. Field mapping, optional continuous sending of new matches, and reconciliation of uncertain writes. Each tool has its own page and setup guide: https://specstream.ai/integrations
- Automation: Zapier, Make, and n8n with triggers (find completed, monitor found something new, records added to a collection) and actions (describe a request, quote, start a find within a maximum, get matching records, rerun a saved request). Slack and Teams. Signed webhooks for `find.completed`, `monitor.found`, and `collection.records_added`.
- Templates: ready-made requests for each record type, opened in the request box to edit before anything runs.
- Security: passkeys; SAML single sign-on with Okta, Microsoft Entra ID, Google Workspace, OneLogin, JumpCloud, or any SAML 2.0 provider, with domain verification and SCIM provisioning; an audit log of sign-ins, access, spending, exports, and automation, kept 12 months and exportable as CSV.

## Docs

- Docs: https://specstream.ai/docs
- Quickstart: https://specstream.ai/docs/getting-started
- Sources on every value: https://specstream.ai/docs/evidence
- Where SpecStream looks: https://specstream.ai/docs/sources
- How a request works: https://specstream.ai/research#how-it-works
- The workspace: https://specstream.ai/docs/workspace
- Desktop, phone & browser: https://specstream.ai/docs/apps
- Connect your tools: https://specstream.ai/docs/integrations
- Keeping data current: https://specstream.ai/docs/schedules
- Webhooks: https://specstream.ai/docs/webhooks
- Agents pay per result: https://specstream.ai/docs/agent-payments
- Agent frameworks: https://specstream.ai/docs/frameworks
- Billing: https://specstream.ai/docs/billing
- Account & access: https://specstream.ai/docs/account
- Pricing: https://specstream.ai/pricing
- Signals & feeds: https://specstream.ai/signals
- Talk to a human: https://specstream.ai/talk-to-a-human