Web search that returns READY-TO-USE grounding text, not links: for one query you get the passages an independent search index extracted from the pages it ranked, flattened into chunks with their source URL and title and an approximate token count, under a token budget you choose (1024-32768, default 4096). Use it to ground a model or a RAG pipeline in one call, where search + extract would take several.

LLM inference proxy (Premium tier) - o3 or o3-mini reasoning models. Supports vision (up to 2 image URLs) and structured output (response_format: json_object or json_schema). No API key needed; pay per call via x402. Input capped at 32k chars, output at 2048 tokens.

LLM inference proxy (Pro tier) - GPT-4o or GPT-4.1. Supports vision (up to 2 image URLs) and structured output (response_format: json_object or json_schema). No API key needed; pay per call via x402. Input capped at 16k chars, output at 2048 tokens.

Compute the monthly (or per-period) payment on a fully-amortizing loan: mortgage, auto, student loan, business loan. Returns the periodic payment, total paid over the term, and total interest. Matches Excel's PMT(rate, nper, -principal). Use this when you just need the payment number, not the full per-period schedule (see amortization).

Placeholder text.

Generate placeholder lorem ipsum text. Supports paragraphs or sentences mode with configurable counts. Uses a fixed vocabulary and deterministic assembly.

Render CommonMark + GFM markdown to HTML.

Standard matrix multiplication of two 2D arrays: A[m×k] × B[k×n] → C[m×n]. Validates that inner dimensions match. Each dimension is capped at 100.

Inspect any audio/video URL with ffprobe: container, duration, bitrate, and per-stream codec details (sample rate/channels for audio, resolution/fps for video) as JSON. Body: {"url":"https://…/file.mp4"}.

Read from a wallet-scoped namespace.

Atomically write (or release) a key only if its current value equals `expected` - the coordination primitive for distributed locks and optimistic concurrency across agents. Acquire a lock: expected=null + a value + ttlSeconds. Release it: expected=<your token> with no value (deletes on match). Update safely: expected=<old>, value=<new>. Returns whether it swapped and the current value.

Delete a remembered document by id from the recall store.

Share your namespace with another wallet so different agents can coordinate through it. Grant read or readwrite access to a grantee wallet, optionally with a TTL. This is the cross-agent sharing a single agent cannot provide for itself.

List the wallets you've granted access to your namespace, with their mode and expiry.

Atomically increment (or decrement) a numeric key and return the new value - a coordination primitive for counters, locks, and rate budgets shared across agents. Creates the key at 0 if absent.

Tamper-evident history of every change to a namespace - an append-only, hash-chained audit log the server attests to (provenance an agent can't forge for itself).

Recall remembered text by similarity to a query (ranked by cosine similarity), not by exact key. Returns the top-k matches with scores. The retrieval half of the wallet-scoped semantic memory.

Store a piece of text for later similarity recall - a per-wallet semantic index an agent cannot host in-session. Returns an id. Prefer memory-write for structured state you will look up by exact key; this pair is for fuzzy, meaning-based retrieval. Pair with /api/memory/recall to retrieve by meaning, not exact key.

Revoke a previously granted wallet's access to your namespace.

Fetch page metadata for a URL: title, description, OpenGraph, Twitter cards, canonical URL, favicon.

Look up the MEV-Boost payment for a specific slot or block number via the Flashbots relay. Returns the builder, proposer, value paid in ETH, gas usage, and tx count. Returns found=false if the block was built locally or via a different MEV-Boost relay (Flashbots is one of several).

Builder market share aggregated over the last N MEV-Boost blocks via the Flashbots relay. Returns per-builder block count, share percent, total ETH paid to proposers, and average per block. Use to track block-builder concentration or detect a new builder gaining share.

Last N MEV-Boost-built blocks via the Flashbots relay. For each block: slot, block number/hash, builder pubkey, proposer pubkey, gas used/limit, ETH paid to the proposer, and tx count. Use to monitor block-builder market share, proposer revenue, or to correlate an arbitrage opportunity with the block it landed in.

Look up a MIME type by file extension.

Modular arithmetic operations: mod (a mod m), modpow (a^b mod m via fast exponentiation), and modinverse (a^-1 mod m via extended Euclidean algorithm). Uses BigInt for arbitrary precision. Useful for cryptography and number theory.

Check text for harmful content using OpenAI moderation (omni-moderation-latest). Returns flagged status, category breakdown (harassment, hate, self-harm, sexual, violence, etc.), and confidence scores. No API key needed; pay per call via x402. Text capped at 10k chars.

Encode text to Morse code or decode it back. mode: encode (default) or decode. Words separated by ' / '.

Compute simple (SMA) and exponential (EMA) moving averages over a numeric series. Returns one value per input position - the first (window-1) SMA values are null since there isn't enough history. EMA uses the standard alpha = 2/(window+1) smoothing factor used in technical analysis.

Run 2-5 web searches in one call with a 20% volume discount vs. individual searches. Each query returns ranked results (title, URL, snippet). Ideal for multi-faceted research or comparing sources on different aspects of a topic. If some queries fail upstream, the call returns the ones that succeeded with an error on each failed query and is charged; a call where every query fails is not charged.

Your own purchase history, keyed to the wallet that pays for the call - no wallet parameter, no signup: the x402 payment IS the identity, so nobody can read another wallet's profile. Returns totals, per-tool counts, per-chain breakdown, recent receipts with settle tx hashes (independently verifiable on-chain), and this wallet's refunds: any payment that settled for a call that failed to deliver, with its status and our refund transaction once sent.

Convert text to the NATO phonetic alphabet (Alpha, Bravo, Charlie...) or decode phonetic words back to text. Letters and digits supported. Word boundaries marked with " | ". Deterministic, pure CPU.

Fetch NFT collection (ERC-721 / ERC-1155) metadata by contract address on Ethereum, Base, Polygon, Arbitrum, or Optimism: name, symbol, token type, total supply, deployer, OpenSea slug + image, floor (ETH), Twitter/Discord, and verification (safelist) status. Use to label a contract address with its collection identity before pulling tokens or sales.

Get the current floor price for an NFT collection across OpenSea + LooksRare (Ethereum). Returns per-marketplace floor (ETH), collection URL, retrieval timestamp, and the cheapest cross-marketplace floor with the marketplace tag. Use to price a collection without scraping marketplace UIs, or to detect cross-venue arbitrage gaps.

Return the NFTs owned by a wallet address on a given network. Each row carries the collection name, contract address, token ID, image URL (where available), and ERC-721 vs ERC-1155 standard. Up to 100 per call - paginate with `pageKey` from the previous response.

Resolve the metadata for a single NFT: title, description, image URLs (original + cached CDN), attributes/traits, and the standard (ERC-721 vs ERC-1155). Useful when you have a contract+tokenId and need the display data without re-fetching the whole collection.

Recent sales for an NFT collection (or specific token) on any of 5 chains, ordered descending by block. Returns buyer, seller, marketplace (OpenSea/LooksRare/X2Y2/Blur), price (units + wei), protocol/royalty fees, tx hash, block, and quantity for ...

Compute the net present value of a stream of cashflows at a given discount rate. Index 0 is treated as t=0 (today, not discounted); indices 1..n are discounted by (1+rate)^t. Matches Excel's NPV but with the conventional t=0 treatment most finance textbooks use (Excel itself starts discounting at t=1 - see notes). Use for capital-budgeting decisions: positive NPV = creates value at the discount rate; negative = destroys value.

Format a number: thousands separators, fixed decimals, optional currency, or compact (1.2K / 3.4M).

Run read-only SQL against Coinbase's indexed, DECODED blockchain data - base.events (decoded logs with parameters), base.transactions, base.blocks, base.decoded_user_operations, base.transaction_attributions (builder codes), plus solana.instructions and hyperevm.events. ClickHouse-dialect SELECTs, server-side grammar validation, up to 50k rows / 30s / 12 joins. Ask Base anything - token flows, event analytics, gas studies - in one call, no indexer to run.

The table + column schema for the onchain-sql tool - every queryable table (base.events, base.transactions, base.blocks, base.decoded_user_operations, base.transaction_attributions, solana.instructions, …) with its columns and types. Fetch once, then write SQL with confidence.

Generate a single-use Coinbase Onramp URL that lets a human fund any wallet with a card or Apple Pay - the fastest way to put real USDC into an agent's wallet. Returns the ready-to-open URL, and a fee-inclusive quote when Coinbase provides one. Networks: base, ethereum, polygon, arbitrum, optimism, solana.

Compare two OpenAPI 3.x or Swagger 2.x documents and return a structured diff: added / removed / changed endpoints, with a conservative "is any change breaking?" flag. Breaking = an endpoint or required-2xx status was removed, a required parameter was added, an optional param became required, a param type changed, or a JSON body field became required. Pure CPU - deterministic, no network, no $ref dereferencing (resolve refs upstream if needed).

Flatten an OpenAPI 3.x or Swagger 2.x spec into a structured list of callable endpoints - one row per operation with method, path, operationId, summary, tags, parameters (name / in / required / type), JSON-body flag, and documented response codes. Includes per-method and per-tag counts so an agent can pick what to call next without parsing the full spec. Output is sorted by path then method for stable, agent-friendly grouping. Pure CPU - deterministic, no network, no $ref dereferencing.

Score an OpenAPI 3.x or Swagger 2.x spec on agent-readiness - i.e. does an LLM-driven caller have what it needs to call the API correctly without guessing. Returns a 0..100 score, severity counts, and a structured list of violations with stable rule codes. Checks: documented title/servers/paths, per-operation summary/description/operationId/tags, documented 2xx + error responses, param descriptions/schemas/examples, response descriptions, JSON response schemas.

Synthesize a JSON response body for one operation + status code in an OpenAPI 3.x or Swagger 2.x spec. Locate the operation by operationId or method+path. Pick the response by explicit `status`, else the first 2xx, else the first documented response. Generation precedence: operation-level `example`/`examples` > schema `example` > recursive type-inferred walk (objects walk properties, arrays return [mock(items)], enums return the first value, $ref is surfaced literally).

Shrink an OpenAPI 3.x or Swagger 2.x document for LLM context by stripping verbose meta-fields (examples, descriptions, summaries, tags, externalDocs, deprecated). Categories are explicit; default is `["examples", "descriptions"]`. User-defined property names under `properties: { ... }` are protected - only the meta-field `example` keyword inside a property schema is removed, not a user property literally named "example".

For one operation, return the minimum set of inputs an agent must provide to make a successful call. Locate the op by `operationId` (preferred) or by `method`+`path`. Output is a single flat array tagged by `in`: `path` (always required), `query` / `header` / `cookie` (only when `required: true`), and `body.field` entries naming the top-level required fields of a required JSON request body.

Inline every local `$ref` in an OpenAPI 3.x or Swagger 2.x document so downstream tools see a self-contained spec. Resolves `#/components/...` (OpenAPI) and `#/definitions/...` (Swagger) JSON-pointer refs anywhere in the document, including inside components themselves. Per-branch cycle detection: if A→B→A, the second A is left as a `$ref` and recorded under `circular`. External refs (http://, file://, ./other.yaml) are never fetched - they're reported under `external` and left as-is.

Search operations in an OpenAPI 3.x or Swagger 2.x spec against a free-text query. Tokenizes the query (lowercase, alphanumeric runs), scores each operation by which fields the tokens match - operationId +3, path +3, tags +2, summary +2, description +1 per matched token - and returns ranked results with a `matches` array naming the contributing fields. Sort: score descending, then path ascending for stability. Limit defaults to 10 (max 100).

Resolve authentication requirements across an OpenAPI 3.x or Swagger 2.x document. Returns the catalog of security schemes (`components.securitySchemes` in OpenAPI 3, `securityDefinitions` in Swagger 2) verbatim, the document-level default, and the *effective* security for each operation after layering.