Tools reference
Tools are the assistant's hands. Each is a function the model can call with structured arguments; results stream into the chat and, for creation tools, onto the board. This page lists all of them.
Legend. read-only tools are available in Plan mode. PRO tools need a Pro or Enterprise-Pro license. Needs lists any account or key the tool depends on; everything else works with no configuration.
Tools are grouped into tiers. With Smart tool loading off (the default) every tier you are entitled to is available immediately; with it on, the assistant starts with core and loads others on demand.
Core
Always available.
| Tool | What it does |
|---|---|
search_board read-only | Full-text search of the board's nodes, actors, organizations and narratives. The assistant is instructed to call this before creating anything, to avoid duplicates. Omit the query to browse everything. Every result carries the id of its board node, so it can be used directly for connections and edits; people and organizations that are not on the board (or were merged into another node) are flagged. |
read_node read-only | Read a node's complete content when the search result was truncated. |
search_documents read-only | Search the board's attached document folders by meaning and keywords; returns passages with file and page. |
read_document read-only | Read consecutive passages of an indexed file, e.g. around a search hit, plus the actors/organizations it mentions. |
get_board_diff read-only | Everything created or modified in a time window ("1h ago", "since yesterday", ISO dates). Useful after a collaborator's sync or a feed poll. |
create_note | Create Note nodes. |
create_connection | Create typed connections (all eleven types), directed or two-way, with reasoning. Several at once in one call: each is checked on its own, so one bad reference is reported without losing the rest, and a connection that already exists is skipped rather than duplicated. |
update_node | Update a node by ID or name. Content is merged, not replaced. |
delete_node | Delete a node (duplicates, errors, merges). |
delete_connection | Delete a connection by ID or by source/target names. |
get_cached_result read-only | Retrieve a large earlier tool result that was truncated in the chat. |
review_chat_images read-only | Look again at images you sent in the conversation before describing or saving them. Formats models can't read directly (TIFF, BMP, AVIF, HEIC, SVG) are converted automatically. |
ask_user | Ask you a multiple-choice question and wait for the answer. |
get_more_tools | Load another tier (only when Smart tool loading is on). |
Entities
| Tool | What it does |
|---|---|
create_actor | Create actors (people, handles, sources, journalists, outlets). Placed on the board as nodes; Wikipedia enrichment runs afterwards. |
create_organization | Create organizations. Same behaviour. |
create_event | Create Events with dates, significance and real source URLs. |
create_location | Create Locations with coordinates. |
create_prediction | Create Assessments (judgements with a confidence) with reasoning and a timeframe. |
create_wallet | Create a Blockchain node for a wallet address (EVM chains, Solana, Bitcoin), fetching balance, tokens and activity. Accepts ENS names. |
create_transaction | Create a Blockchain node for one transaction, parsed into sender, recipient, value, fee, method and decoded transfers. |
All entity tools accept a batch (an array) so the assistant can create ten actors in one call.
Research
| Tool | What it does | Needs |
|---|---|---|
web_search read-only | Web search. Uses TinyFish (free, live) when a key is configured, then Exa, otherwise DuckDuckGo through a hidden browser. topic: news or research (academic papers), a time range down to the last hour, domains to include or exclude, and deep: true to extract full text from the top three results. Safe search on by default. | TinyFish or Exa key optional |
search_wikipedia read-only | Wikipedia article summaries. | — |
scrape_url read-only | Read one to ten pages. With a TinyFish key they are rendered in a real browser (JavaScript sites, PDFs), and an optional question returns verbatim passages that answer it. Otherwise, and as a fallback, the same extractor as News Article nodes, with archive fallback for paywalls. | TinyFish key optional |
wallet_lookup read-only | Balance, USD value, token holdings, transaction count, first/last activity and contract flag for a wallet address on any supported chain. | Etherscan key for EVM |
wallet_transactions read-only | Recent transactions with direction, amount, counterparty and explorer links; token transfers included on EVM chains. | Etherscan key for EVM |
transaction_lookup read-only | One transaction parsed: status, confirmations, from/to, value, fee, method, contract created, and every native/token/NFT/internal transfer (inputs and outputs on Bitcoin). | Etherscan key for EVM |
browse read-only | See Browsing and procedures. Drive a hidden browser step by step: open a URL, read the rendered page as text with links inline, take a screenshot (viewport, full page or one element) that the assistant can see and add to the board, list links or forms, click (by selector or visible text), fill inputs, select, press keys, scroll, wait, back. The session persists per board, so the assistant can search, paginate and follow links on JavaScript-heavy sites, registries and search forms that scrape_url cannot handle. When a site needs a login or CAPTCHA, handoff shows the window so you can do it yourself (the assistant never types passwords or card numbers); cookies are kept so you stay signed in. | — |
run_procedure read-only | Replay a saved browse procedure with parameters (e.g. { query: "Acme" }) and return what it reads. Called without a name it lists the procedures you have. | — |
store_procedure / edit_procedure read-only | Once a browse sequence works on a site, the assistant saves it as a named procedure with {{param}} placeholders, so future requests on that site replay it instead of rediscovering the flow. edit_procedure updates, renames or deletes one. | — |
image_search read-only | Find images to use with create_media. Safe search on by default. | — |
geocode_address read-only | Address → coordinates and back (OpenStreetMap Nominatim). | — |
route_directions read-only | Driving / walking / cycling route between two points with distance, duration and turns (OSRM). | — |
map_snapshot read-only | Render a map image with markers and/or a route and return it to the model so it can see the geography. | — |
search_nearby read-only | Points of interest around a location by category or name (OpenStreetMap Overpass). | — |
youtube_transcript read-only PRO | Transcript of a video (watch, shorts, live, youtu.be URLs). Long transcripts are summarized, optionally around a focus. | — |
youtube_channel_videos read-only PRO | Recent videos from a channel by @handle, URL or channel ID. | — |
youtube_comments read-only PRO | Top-level comments with authors, likes and times. | — |
person_search read-only PRO | US public-records people search (ZabaSearch): name, city, state, age. | — |
person_details read-only PRO | Full record for a person_search result. | — |
search_court_cases read-only | US court records via CourtListener: federal PACER dockets and state case law. | CourtListener token |
get_court_case_details read-only | Parties, attorneys and docket entries for a case. | CourtListener token |
search_judges read-only | US judges: biography, appointments, affiliations. | CourtListener token |
search_github read-only | GitHub users, repositories and code - for technical profiling. | GitHub token optional (raises rate limit) |
username_search read-only | Check a username across ~480 sites (a port of Sherlock). | — |
list_username_sites read-only | List the sites username_search covers. | — |
wayback_lookup read-only | The Wayback Machine capture closest to a date (or the latest). With history, every distinct version of the page over time, so you can see when it appeared, changed or disappeared. | — |
wayback_fetch read-only | Read an archived copy of a page, by original URL or web.archive.org link. | — |
wayback_save | Ask the Wayback Machine to capture a live page now and return the permanent link: independent evidence of what a page said, in case it is edited or deleted. Takes 10 to 60 seconds. | — |
whois_lookup read-only | Who registered a domain and when: registrar, registrant (often redacted), registrant country, dates, nameservers, abuse contact. Asks the registry directly over RDAP, or WHOIS for country domains without RDAP (.io, .co, .me, .eu, .ru, .jp…). | — |
dns_lookup read-only | DNS records of a domain (A, AAAA, CNAME, MX, NS, TXT, SOA, CAA) with its SPF and DMARC policies, or the reverse name of an IP. TXT verification records can tie domains to the same owner. | — |
subdomain_search read-only | Every host name a public TLS certificate was issued for under a domain (certificate transparency via crt.sh), with first and last dates. | — |
ip_lookup read-only | Who owns an IP address: network, organisation, country, range, abuse contact, ASN and reverse DNS. Tells you the host or ISP, not a person. | — |
email_lookup read-only | Is the address well-formed, does its domain take mail, is it free webmail, and is there a public Gravatar profile (name, location, job, linked accounts). Sends nothing. | — |
phone_lookup read-only | Parse a phone number offline: country, line type where the numbering plan tells, validity, standard formats and a WhatsApp link. | — |
pivot read-only | Run the board's pivots from a node, chain onto their results, and propose what to add; the proposal opens for you to review, and nothing is added without you. Always loaded. | — |
create_identifier | Put a domain, IP, email, phone number, username or URL on the board as an Identifier node (reuses one that's already there), optionally linked to a node. | — |
reverse_image_search read-only | Where else an image appears and where it first appeared: search links for Google Lens, TinEye (dates each copy was first indexed) and Bing, which the assistant opens with browse. Images that exist only on your computer get the upload pages instead, including Yandex for faces. | — |
image_metadata read-only | EXIF/XMP/IPTC inside an image: camera and serial, capture time, GPS (with a map link), editing software, author. Takes a URL, a chat image or a media node. Social platforms strip this, so it pays off on original files. | — |
search_sanctions read-only | OFAC SDN and the US Consolidated Screening List. | — |
search_sec_filings read-only | Full-text search of SEC EDGAR filings with form and date filters. | — |
get_company_filings read-only | Company profile and recent filings by ticker or CIK. | — |
get_company_financials read-only | Structured annual financials and ratios from XBRL. | — |
get_sec_filing_content read-only | Clean text of a filing document; long ones summarized. | — |
Social
| Tool | What it does | Needs |
|---|---|---|
social_search read-only | Search Reddit, X/Twitter, Bluesky and YouTube; can also fetch a post by URL. | X search needs a connected X account |
fetch_social_post read-only | Fetch one post by URL (X, Reddit, Bluesky). | — |
create_social | Create a Social Post node from a URL; everything is fetched. | — |
search_social_users read-only | Find people on X or Bluesky by name or handle. | Connected account |
get_user_timeline read-only | Recent posts by a user. | Connected account (X) |
get_user_profile read-only | Profile for an X, Bluesky or YouTube user. | — |
linkedin_profile read-only | A LinkedIn profile by URL or public id: headline, location, industry, summary, connection degree, experience, education and skills. Counts as a profile visit on your account. | LinkedIn signed in |
linkedin_company read-only | A company page: description, industry, website, headquarters, size, followers. | LinkedIn signed in |
linkedin_search_people read-only | People search by keywords, optionally restricted to current employees of a company. | LinkedIn signed in |
linkedin_posts read-only | A member's recent posts and shared articles with reaction counts and permalinks. | LinkedIn signed in |
Which platforms work anonymously is detailed in Social media.
Media
| Tool | What it does |
|---|---|
create_media | Media node from an image/video URL, or from an image you attached to the chat. |
create_embed | Embed node for a link with preview (X, YouTube, Reddit, any page). |
create_document | Create a Document node. With sourcePath (for example a file the browse tool fetched) the file itself is attached, with page count; without it a reference node is created for you to attach the file in the editor. |
create_newsfeed | News Article node from a URL (scraped in the background). |
fetch_and_create_newsfeed | Scrape now, summarize, then create the node - the assistant's preferred way to file an article. |
create_rssfeed PRO | Subscribe to an RSS feed or a social account as a feed. |
Narratives
| Tool | What it does |
|---|---|
create_narrative | Create storylines. |
add_to_narrative | Add nodes (by ID or name) to a narrative with a relevance level. |
create_narrative_link | Link two narratives (supports, contradicts, parent/child, …). |
search_narratives read-only | Find existing narratives. |
update_narrative | Replace description, summary, status, themes, or add a paragraph to the end of the description without rewriting it. |
delete_narrative | Delete a narrative; members are unlinked, not deleted. |
Newspapers PRO
| Tool | What it does |
|---|---|
search_newspapers read-only | Historic American newspapers 1777-1963 (Library of Congress Chronicling America), returning OCR excerpts. |
read_newspaper_page read-only | Full OCR text of a page from those results. |
Custom PRO
Your own tools, written in JavaScript and dropped into the custom-tools folder. They appear here under whatever names you give them. See Custom tools.
Notes on sources
- No API keys are ever sent anywhere but the service they belong to. Exa, CourtListener and GitHub keys go to those services; the assistant never sees the key values.
- Public-records and people-search tools return whatever the public source returns. Treat results as leads, not facts, and record verification on the node.
- Sources that block automated access will fail from time to time; the tool reports the failure and the assistant tries another route or asks you.
- Rate limits: Nominatim is limited to one request per second; GitHub to 60 requests/hour without a token; SEC EDGAR asks for a descriptive user agent, which DeepState sends.