What's new

Notes for people who use Weft. Newest first.

0.32.1

Purchases paid with x402 change from pending to settled again. Since 8 October, some stayed pending after the payment went through.

0.32.0

Paid fetch results now include a purchase ID. Use it to look up the final payment status of a purchase.
Search can return a compact view with fewer details per result, for agents that need less text.

0.31.0

Fixed a spending-limit save error when connecting an agent.
Improved payment-status tracking when paid requests fail.
Added workflow guides for finance, research and content.

0.30.0

Every tool says what it does. 458 API operations whose description only repeated their name now explain what they take and return, written from each provider's own documentation.
Categories match the operation. Each operation has its own categories, so a provider appears under every category its operations cover (for example, people search under Business and TikTok data under Social), and category filters return the right operations.

0.29.0

More paid APIs your agent can call. Search now finds 1,085 ready-to-call operations, up from 873. The new operations come from the MPP directory (for example KicksDB, Serper, Firecrawl and many Orthogonal APIs) and are paid with USDC on Tempo.

0.28.1

Ready calls only when the inputs are known. A search result no longer offers a ready weft_fetch call when the provider does not document its inputs, so your agent does not pay for a request that is missing a required value.
Contract caching works behind every client. A repeated contract request with a cached ETag now returns 304 Not Modified, also for clients that receive a weak ETag.

0.28.0

Search results are ready to call. Each MCP search result can include a ready weft_fetch call. The new weft_contract tool returns the full request contract for a result.
No payment for a broken request. When a request body does not match a reviewed contract, Weft refuses it before it pays and tells your agent which field to fix.
Newer MCP clients connect. Weft serves MCP protocol 2026-07-28 next to the earlier version, so new and existing clients both work.
More reviewed providers in search. Researched provider contracts and new search embeddings improve which endpoints Weft finds.
Live wallet updates. Wallet balance and agent connections update without a page reload.

0.27.1

Set spending caps when you approve an agent. Review and edit its per-purchase, daily, and weekly limits before connecting it.
Name API keys for their job. Key labels describe what the key does rather than assigning a buyer or seller role.
Clearer seller key examples. Setup examples no longer use client names that suggest the wrong key scope.

0.27.0

Read product updates in Weft. Open What's New from the dashboard or Learn. The dashboard shows unread releases and remembers when you close them.
Start selling from the dashboard. The seller entry and setup pages guide you through a seller key and the SDK examples for your paid endpoint.
API Keys asks what the key is for. Choose I need to buy or I need to get paid. You do not need to know if you are a buyer.
Connect through stable ChatGPT callbacks. OAuth accepts the supported ChatGPT callback forms and preserves the requested resource through approval.
Read details for one tool without signing in. Agents and websites can request a public provider summary from /tools/:id.json.
Clearer search results. Default agent search excludes five StableSocial operations whose final results Weft cannot yet retrieve. Atlas Instagram descriptions state which fields observed responses contain. Contract links state when a Weft credential is required.
Reviewed RentCast discovery. The new catalog snapshot adds complete descriptions for ten property-data operations, including rental listings and rent estimates. Availability follows the separate production catalog update.

0.26.10

Agents can read the public tools catalog as JSON. Fetch /tools.json from the API origin without scraping HTML.

0.26.9

The connect command is set up https://weft.network/setup.md. That is the copy-paste line for agents. Workflow covers stay on the public pages.
Account email uses Weft branding. Confirmation, claim, password reset, and referral mail use the current Weft layout.

0.26.8

Reviewed tools can include the provider's own helper docs. Matching operations can show provider API-reference links.
API docs point to the current Weft Labs guides. The public API reference no longer sends you to GitHub-only copies.

0.26.7

Agents can find and pay more reviewed APIs. First-party x402 routes (Glassnode, NetIntel, GeoPrimitives, GovLaws, 2s.io, Oblique, OpenWeb Ninja) and reviewed Tempo MPP routes (OpenRouter, KicksDB, Aviationstack, SpyFu, Firecrawl) are available after a live payment challenge and a paid success.
Every buyer can pay Tempo MPP. Production no longer limits Tempo payments to one account. An empty Tempo pocket still funds from Base instead of charging.

0.26.6

See spending caps before you connect an agent. Connection approval and account-claim review show the current per-purchase, daily, and weekly limits when the connection can make purchases.
Workflow updates can reach the gallery without an app release. Valid workflow changes refresh the public gallery. Saved snapshots keep existing workflow links working.
First metered payments can approve token spend automatically. If the wallet needs an on-chain approval before a metered payment, Weft requests it before signing.
Payment approvals expire in 15 minutes. Rejected requests without a settlement receipt no longer leave the same open hold.

0.26.5

New seller API keys reach the payment edge. A quoted paid request can complete after you create or rotate a seller key.

0.26.4

Tempo balance checks use the configured token. Funding no longer accepts a different token with the same symbol.

0.26.3

Agents can collect paid-job results without paying again. After an async job is submitted, the saved result can be retrieved at no extra charge.

0.26.2

MPP funding requests include the required credentials. A funding transfer can start instead of stopping on an authorization refusal.

0.26.1

Flight requests can keep their search attribution. Requests from matching results can link to the returned flight operation when the provider uses named route parameters.

Earlier

The app has a clearer visual hierarchy across every main page. Dashboard, money, seller, account, setup, catalog, and public pages now use a calmer canvas, clearer panels, larger headings, quieter status treatment, and more reliable layouts for large amounts on narrow screens. Routes, payment values, and status meanings are unchanged. Profile initials are now centered in both header and sidebar avatars, with a compact header size on desktop and mobile.
Connections has one clear path to add another client. Buyer accounts can start a new connection from the Connections page. The redundant Activity hub is removed because Purchases and seller Transactions already have direct dashboard navigation. Old Activity bookmarks now return not found.
Find and understand reviewed tools more easily. Signed-in catalog cards now show reviewed price evidence and practical operation documentation with inputs, examples, and outputs. Public provider pages are also easier for search engines to find. Prices are indexed evidence; Weft still confirms the current price when a request runs.
Search has a stronger fallback and better purchase attribution. Search handles more natural wording when semantic ranking is unavailable, and a completed purchase keeps the selected result's position for later analysis.
Hosted setup gives safer account-claim instructions. It puts email verification before the welcome-grant balance check and avoids examples that can leak credentials into logs or project files.
Sellers get a clearer path from profile to revenue evidence. Settings now links to the seller profile editor and shows the storefront link only for a public profile. Product pages show payout destinations, destination changes, and each exact paid path with its payment distribution. Weft does not guess route templates from numeric or UUID values. These observations describe history; they are not public API documentation. Seller setup now consistently calls these items products and explains that seller API keys connect an API to Weft to accept payments but cannot spend buyer funds.
Saved paid results remain available after a merchant hostname retires. An exact retry can return the buyer's stored response without a new merchant call or payment. New or changed requests still pass full network-safety checks.
Merchant disconnects now give agents a clear recovery signal. Weft reports the request phase and whether payment might have moved instead of returning a generic server error. It does not retry the merchant or release reserved spend automatically.
Older open browser tabs recover once after a frontend deployment. If a tab asks for a stale code file, Weft reloads it once for the current release. A repeated failure stays visible for diagnosis.
Search no longer offers one unusable BlockRun market-volume request. The reviewed source used a literal placeholder without a value to substitute, so Weft keeps its evidence but does not present it as executable. Other BlockRun operations remain unchanged.
Limited MPP beta accounts can recover from a short Tempo balance. Weft can start a controlled Base-to-Tempo refill and ask the agent to retry the same request later.
Incomplete Locus operations no longer look ready to call in MCP search. The records remain available for review, but agents do not receive an incomplete paid contract as an immediately usable result.
Agent results are easier to read and recover. MCP fetches keep plain text, JSON, and XML readable; other formats stay lossless as Base64. If a result is too large for the client, an authorized agent can read saved sections without calling or paying the merchant again.
The catalog is easier to scan. Tools and Explore now share clearer categories, filters, cards, and reviewed local provider logos. Search still recognizes the original provider names and labels.
Card top-up errors are clearer and safer. The wallet page now uses the current page security token and can explain supported checkout failures without exposing card, billing, or provider details in diagnostics.
Expired payment approvals can leave pending spend correctly. Weft can now record a consumed variable payment at its settled amount, or expire an unused approval instead of leaving it pending indefinitely.
Catalog navigation stays inside your dashboard. Opening a provider from Explore now keeps the dashboard layout and Explore context, while public provider pages remain a separate experience.
Search better absorbs brief ranking-service slowdowns. Weft retries temporary rate limits and server errors before it falls back to keyword ranking, with a short fixed retry budget so requests do not hang.
New accounts can finish wallet setup again. Crossmint changed the field used to assign wallet recovery control. Weft now sends the current field, so affected buyers can receive their $3 welcome grant and make paid requests.
Browse and install optional workflows from your dashboard. The new workflow gallery gives buyers reviewed install commands and starter prompts for each workflow, with layouts that work on desktop and mobile.
Choose the API key type before you create it. The API Keys page now makes Buyer and Seller API keys separate choices and enables only the types that your account can use.
See spend for each buyer API key. Active and revoked buyer keys now show their pending commitments and settled spend for the last 30 days. Weft keeps the first credential attribution when it safely retries a request.
Signed-in users get the complete reviewed catalog. Signed-out visitors see a small tools showcase. Full provider details, operation contracts, unlisted storefront entries, and network activity require an account or an authorized API client.
Search avoids access methods with proven production failures. Weft now hides the affected APIToll, BlockRun, Clado, Apollo, and Locus methods when they fail, or when a paid response contains no useful result. Other working methods from the same providers stay available.
Purchase details show only your transaction facts. Weft no longer shows network-wide merchant activity counts or dates in the dashboard, REST API, or agent tools. Weft still records payment events internally so it can build useful merchant conclusions without publishing the underlying dataset.
The reviewed catalogue adds 81 paid operations. Massive adds 23 stock market data routes, and Strale adds 58 company, market, and utility routes. Weft verified each payment request without sending a payment or provider credential. Two free Reachpad wallet-recovery operations are also available.
Provider cards now open a full page inside Weft. You can inspect all reviewed operations for one provider and open each operation's contract before you visit the provider's own website.
Variable-price purchases show what you paid. Completed purchases and spend totals use the amount that settled. Pending purchases still reserve their full authorized amount, and wallet responses now show that policy use separately from settled spend.
Merchant timeouts now have a clear recovery path. REST and MCP clients receive a typed timeout response that says which request phase stopped and whether payment might have moved. A retry with the same idempotency key can safely recover the exact timed-out request.
Explore reviewed tools without signing in. The new public Tools page lets you search providers and filter them by category on desktop or mobile. Every listing comes from Weft's reviewed catalogue; the page does not make live calls while you browse.
Every eligible new account now gets the same $3 welcome grant. The amount shown during sign-up now matches the amount paid after confirmation, including agent-connected sign-ups. The offer is labelled in USD.
Pending purchases no longer inflate your total spend. Purchases now shows recent signed commitments as a separate pending amount, while Total spent continues to include only completed settlements.
Purchases are easier to search. Search now uses the full width of the ledger, while status filters and the matching result count have their own clear row. Screen readers also receive the result count as one complete update.
Search results now fit inside narrow app panels. The hosted search widget stays inside its frame from phone width through wide desktop layouts. Dense endpoint details shorten when space is tight, and you can expand them to see the full method, address, and price.
New confirmed buyers receive their signup grant while the treasury has funds. The campaign no longer stops at a separate accounting limit. Weft still watches the treasury balance and keeps an emergency campaign kill switch.
Small endpoint earnings no longer look like zero. Revenue below one cent now keeps its exact value in the seller dashboard instead of rounding to $0.00.
MPP payments are entering production beta. Weft can now pay compatible merchants from the Tempo balance in a buyer's existing wallet, with the same spending limits, denylist checks, receipts, and retry safety as other paid requests. The first production rollout is restricted to one test account while settlement and delivery are verified end to end.
Invitations now stay visible until your friend joins. The referrals page shows email invitations separately from friends who have created an account. When an invited address signs up through your referral, it moves automatically from pending to joined, even if the invite and signup happen at the same time.
A payment refused for low funds now shows another way forward. Alongside topping up, the dashboard can offer the referral reward when the campaign has funds available, with a direct link to invite a friend.
Email sign-up works again after the form redesign. The shorter sign-up form now submits the matching password confirmation that account validation expects, so creating an account no longer stops on a hidden confirmation error.
Payment details can now show the request metadata a merchant supplied. x402 payments preserve the merchant's structured extensions and resource URL, then show them under Request Metadata on the payment detail page. Older payments without this information continue to display normally.
A cleaner signed-out screen. The header you see before signing in is now plain text — the small decorative icons are gone and the labels read in sentence case.
A shorter, clearer sign-up page. Sign-up now fits into a responsive split-panel layout, leads with what Weft removes from an agent integration, and shows providers that Weft has verified as payable. The repeated password field is gone, field errors are linked to their inputs, and the provider animation respects reduced-motion settings.
Pay-as-you-go merchants now work. Some services charge per request and only tell you the final amount after the work is done — Grok's web search is one. Weft can now sign that kind of payment, and you are charged what actually moved, not the maximum you agreed to. Your spending limits and your dashboard both count the real amount.
A purchases page that leads with your purchases. The ledger is now at the top of the page. Click any row and its receipt opens beside it — no page change, and the list stays on screen. When nothing is selected, the side panel shows your 30-day spend, your chart, and your top merchants.
We only call a service payable when we have seen it ask for payment. A few listings said they accepted a payment method they never actually offered, so a request to them failed. Every payable listing is now backed by a real payment challenge we observed ourselves. Three Tomba operations lost their payable badge as a result — they stay searchable, and they will come back when their payment rail does.
Revoked keys really do stop working at the edge. Key and account access changes now settle to the correct final state even when two changes happen at once, so a revoked key cannot linger.
The catalogue doubled again — 865 services you can search. We imported the first-party vendors listed in the Coinbase Bazaar, probing all 465 of them before listing, across 63 providers. One vendor was deliberately left out: it sold hijacked social accounts and prepaid cards, which is not something an agent should be able to buy through us.
Honest prices. Some listings showed the merchant's minimum charge rather than what a normal request costs — in the worst case a service that advertised $0.001 actually charged $4.61. We re-checked all 533 payable services with a real example request; the 32 that were wrong now show the real amount and are marked as variable-priced.
Services that spend money are labelled as such. 52 listings that sell goods — wine, jewellery, print-and-mail — used to look like ordinary read-only lookups. They now carry a clear "this has real-world side effects" label.
Payments go to the right host. A listing could show one address to call and a different one that takes the payment. They now always match.
Setup instructions live at one address. Point your agent at /setup.md and it gets the current instructions, matched to the site it is talking to.
Nearly twice as many services you can pay for. We added a new first-party payment rail and made a large batch of Locus operations searchable and payable — the number of operations an agent can reach jumped from 226 to 434.
A clearer first step when you connect. Onboarding now leads with connecting your agent, so the one thing you need to do first is front and center, with honest wording about what your funding grant covers.
Closing your account cleans up fully. When you close your account, the related setup data tied to it is now removed too, so nothing is left behind.
Connect your agent before you have an account. An AI agent can now connect to Weft and start searching straight away with a temporary connection, then email you a link to claim it. When you claim it, the same connection keeps working and gains your balance and paid requests — nothing to reconnect.
Pick how you set up: MCP or CLI. The setup guide now offers both, and the Connect card on your dashboard leads with the copy button so the one line you need is easier to grab.
Invite a friend, earn credit. Share your invite link — when a friend signs up and makes their first payment to a merchant, you get $3 in credit. There's a new invite page to get your link and see how it works.
See where your balance came from. A new funding history page brings your card top-ups and any credits granted to you together in one clear, ordered list.
Nicer shared links. Links to Weft now show a branded preview and icon when you paste them into chats, docs, or social posts.
A faster dashboard that tells you what your wallet needs. The home page no longer waits on a live balance lookup, so it loads quickly and keeps working even when our wallet provider is slow. It now says plainly whether your wallet is still being set up, needs funds, or is ready to spend. Your exact balance is still on the wallet page, fetched fresh every time.
A bigger welcome credit. New accounts now start with real USDC in a range from $1 to $5 (we're testing which amount helps most), instead of a flat $1 — so your first paid requests go further.
Payments now run on a single network. All payments settle on Base. We removed the second payment network, which means fewer moving parts and more reliable settlement — you just pay, and it goes through.
Clearer messages when a payment can't complete. If a paid request fails to settle, Weft now tells you which payment network was involved instead of a vague failure.
Requests pick a payment method your wallet can cover. A paid request now chooses a settlement path your wallet can actually pay on, so fewer requests fail for using the wrong one.
Top-up failures now explain themselves. When adding funds by card fails, Weft now tells you what happened instead of a bare refusal — so you know whether to retry, use a different card, or wait.
We now watch money that left your card but never reached your wallet. If a card payment clears but the USDC never lands, Weft notices instead of staying silent.
New buyers get $1 in real USDC to start. Once you confirm your account, Weft funds your wallet with a real dollar so your very first paid request goes through instead of being blocked for lack of funds.
Tempo payments now work. Charges on the Tempo network were being rejected even when your wallet held funds; your wallet is now fully set up on Tempo, so those payments settle. This unblocks most services in the catalog.
More services are now payable. When a service offers a payment method Weft cannot complete, Weft now falls back to a method it can — so requests that used to fail with a valid alternative right there now succeed.
Clearer card top-up totals. The top-up screen no longer shows a misleading "no card fee" line when a fee is included; it always shows exactly what leaves your card and what lands in your wallet.
A better API key experience. Creating a key now suggests a name for your agent and lets you set an optional expiration; the key list gains search, an Expires column, and a last-used time, and API Keys is now a top-level item in the sidebar.
Form errors now show up. Sign-up and other forms that used to reload blank after a rejected submission now show the message explaining what to fix.
Top up your wallet with a card. You can now add funds with a debit or credit card, so you can get started even with no crypto on hand. You see exactly what you pay and what you receive before you enter any card details.
Card checkout now opens in a focused window. The payment steps moved into a clear, roomy dialog with your quote pinned in view the whole time, instead of squeezing into a narrow column beside your balance.
A short Tempo balance now tells you how to recover. When a payment can't go through because your Tempo balance is a little short, Weft starts topping it up for you and tells you when to try again, instead of just refusing.
Your Tempo balance now reads correctly. Weft reads your Tempo balance straight from the network, fixing a case where a funded wallet could show as empty and wrongly reject a payment.
Recover on your own if wallet setup gets stuck. If your wallet setup ever stalls, you can now retry it yourself from the dashboard instead of being stranded with no way forward.
Your wallet now sets itself up automatically. Weft creates your wallet and its smart account for you when your account is provisioned, so you no longer handle a separate key export or setup step.
Paid fetches and Tempo payments run on the new wallet. Both payment rails now go through Weft's updated wallet, and your spending limits, blocked-service rules, and balance checks all keep working exactly as before.
Search ranks results by meaning, not just keywords. Weft compares what you asked for against what each service does and surfaces the most relevant matches, even when you do not use the exact name.
The first search of a session is faster. We keep the search ranking ready behind the scenes, so you get results quickly without a slow startup wait.
Behind the scenes, search keeps a simple keyword fallback so results still come back whenever the smart ranking is unavailable.
Agents keep searching without interruption while you claim their account. A bootstrapped agent no longer loses its temporary search access during the moment it hands off to full credentials, so there is no gap in service while you review and approve the account.
Agents can now create a Weft account and hand it to you securely. An agent can start with a limited bootstrap credential, then ask you to review and claim the account before it receives normal access.
Search results now explain exactly how to call each service. Results can include the inputs, request method and URL, price and access details, output shape, and polling steps for work that finishes asynchronously.
Search can find more verified, ready-to-use paid services. The curated marketplace now includes web search, page extraction, deep research, and email intelligence from Parallel, Linkup, Tavily, and Tomba, with the same price and type filters.
Send structured data through MCP fetches. Object request bodies are now serialized correctly, so agents can call APIs that expect JSON payloads.
Weft tools are easier for agents to discover. Tool descriptions now lead with jobs such as scraping, enrichment, and people research instead of requiring agents to know the underlying payment method.
Curated search results are tuned for relevance. Search now favors services whose names match what you asked for and understands requests to check, extend, or stop a service, so results land closer to what you meant.
Behind the scenes, we strengthened sign-in protection, payment amount handling, paid-fetch network safety, and settlement webhook verification.
Adding funds by card now works correctly in production. We fixed a configuration bug that could route the card-based top-up flow through a test environment instead of the live one, so wallet top-ups behave as expected.
Spending limits now accept the same fine precision as your charges. You can set per-transaction, daily, and weekly caps down to the sub-cent amounts purchases are actually billed at, so a limit can precisely bound real spend.
Tighter handling of sign-in data. We hardened how sign-in and connected accounts are processed so sensitive credentials are never retained where they do not belong.
Learn how to charge for your own API. The docs now include a "Charge for your API" guide — create a key, drop in the Express or Hono middleware, and start getting paid per call. Read it at weft.network/docs.
Set spending caps on your wallet. The wallet page now lets you cap how much can be spent, with a single, clearer deposit button.
Forms save as you go. Settings and edit forms across the dashboard now save automatically — no more hunting for a Save button.
More reliable payments. We fixed how paid fetches are settled and attributed, so charges land against the right activity and nothing is dropped silently.
Tidier navigation. Removed the wallet balance pill from the sidebar for a cleaner layout.
One unified dashboard. Buyer and seller views now live in a single, tidier dashboard with a shared navigation shell, so everything you can do is in one place instead of switching between separate screens.
Explore your capabilities. A new Explore area lets you browse what you can do, and the whole row is now clickable — not just the small link — so it's easier to jump straight into a capability.
A clearer path to connect. The menu now shows only the sections that apply to your account, and the Connections page points you directly to MCP setup when you haven't connected anything yet.
Cleaner, more readable screens. Polished the sidebar and Explore layout and brought back the account-type card that had been hidden.
Improved developer docs. Refreshed the API reference and made the SDK examples self-contained so they're easier to copy, paste, and run.
Security update across the platform. We patched outstanding security advisories in Weft's dependencies. How you use Weft did not change.
Paid fetches that actually settled are no longer mislabeled as refused. A paid request to a service could complete — you received the content and the payment landed on-chain — while your history recorded it as a merchant refusal with no transaction. Weft now reads the full range of ways a service can return its payment receipt, and treats a successful, delivered response as exactly that. Your purchase history reflects what really happened.
Payments that were stuck "in limbo" now clear correctly. Some paid requests could sit unsettled forever — never counted as spend, yet still eating into your daily and weekly spending limits (in one case, a single stuck payment accounted for most of a buyer's reported weekly spend). Weft now sweeps and finalizes these across all request types, so your spending totals and limits stay accurate.
The purchases API now matches your dashboard. GET /api/v1/purchases returns the same set of transactions the dashboard's Purchases page shows, so you won't chase a payment that was never yours.
Retrying a request with a reused key gives a clear error instead of crashing. Reusing an idempotency key on a changed request now returns a proper, explained error over both the API and MCP, rather than failing with an unexpected server error.
Your wallet now shows one balance. Instead of a four-way breakdown across networks and tokens, the wallet leads with a single dollar amount and a deposit button — the number you actually manage. The per-network detail is still one click away under "Balance detail" if you need it. And if a balance can't be read reliably, Weft shows a dash and says which part failed rather than pretending the missing side is worth $0.
Purchases now shows what was paid, and with what. The Purchases page shows spend over the last 30 days and your top merchants. Those totals match the booked spend for that window.
Clearer, more honest purchase history. Pending rows that were never your purchases no longer sit forever as "Pending"; every summary now states its own time window; and empty trend indicators explain themselves ("no spend yet") instead of showing a blank dash.
Weft helps your agent discover and use it, on its own. Your agent now gets a built-in Weft usage skill over MCP and can surface what Weft can do proactively — less manual setup to get real work flowing.
New building blocks for the buyer API. A dedicated purchases API and expanded account endpoints, plus stricter handling so that retrying a paid request with the same idempotency key won't double-charge you.
Paying on the Tempo network now just works — no per-network funding. Weft keeps a small working balance topped up automatically from your main USDC and converts exactly what a payment needs inside the transaction itself, so your agent can pay on Tempo mainnet without you funding each network by hand. You fund one balance; Weft handles the rest in the background.
Search results now tell your agent how to call and how to pay for each endpoint. Every result carries the request shape (what to send) and its payment options, so your agent can call a service and settle with its own x402/MPP wallet instead of guessing. Prices are marked to show whether they were seen live from the endpoint or come from our index.
Connecting an agent is easier and clearer. Your dashboard now walks you through it with a card that stays put until an agent is actually connected — no more confusing "connected" message before anything is — and agents can now set themselves up.
The install page and dashboard install command just work. The setup page now leads with your agent, and the dashboard's copy-paste install command — which previously didn't work — is fixed, so there's far less manual fiddling to get connected.
Payments settle directly in the currency a service asks for. Purchases no longer route through an intermediary conversion, and a new push-payment path means more paid endpoints work out of the box.
Exact amounts everywhere your agent sees them. Balances and amounts on agent- and API-facing surfaces now show their exact value instead of being rounded to cents.
Sign up and you're ready to go — no setup wizard. Creating an account used to mean three more decisions before you could do anything: pick an account type, pick a network, and type in spending limits. Those are gone. Every new account is a ready-to-use buyer, and your wallet and sensible default spending limits are set up automatically in the background while your dashboard loads. Want to sell as well? Turn that on from Settings whenever you like.
Signing in with Google works reliably again. A stale security token could block the Google sign-in and sign-up flow; that's fixed.
Search in production now returns live results. The production search API and the weft_search tool are now backed by the real search platform instead of sample data.
Search results show where each price came from. Every priced hit now tells you whether its price was observed live from the endpoint or inferred from a spec, so you know how fresh it is.
The payments API no longer returns a fee field. The fee_amount field has been removed from payment records because its value couldn't be reliably verified.
Search results now tell you more about each matched endpoint. Every hit from the search API now includes the HTTP method it was matched on and a grouping id that ties together the different method variants of the same URL, so you can distinguish and group callable operations more precisely. Both fields are now documented in the API reference.
Paid fetches on Base mainnet now go through. Weft was announcing the network under a name no mainnet merchant recognizes, so every real-money payment was rejected before it ever reached the chain. It now uses the name the ecosystem actually expects.
Payments from upgraded wallets are accepted again. Once a wallet had been used for a sponsored transaction, some merchants' verification refused its signatures outright. Weft now detects those wallets and signs in the format they require, while ordinary wallets are unaffected.
A payment no longer fails just because a network check is slow or unreachable. If Weft can't reach the blockchain to inspect your wallet, it falls back to the standard signature instead of erroring out, and the check is now capped at a couple of seconds so it never holds up a payment.
Paid fetches now work against a whole class of merchants that used to fail. Some paid APIs send their payment request in a response header rather than in the response body. Weft previously ignored that header and gave up with a confusing "no payment options offered" error. It now reads the header whenever the body doesn't carry a usable payment request, so those merchants pay through normally.
Stuck payments now resolve to a clear final status. A payment left hanging because a merchant never confirmed it is automatically swept to an "expired" state, so your history shows an honest outcome instead of a charge frozen mid-flight.
A payment that really settled on-chain is never wrongly marked expired. Before finalizing a stalled payment, Weft now checks the blockchain directly, so a transaction that actually went through keeps its settled status.
More paid fetches succeed when a merchant doesn't cover the network fee. When a charge doesn't sponsor its own gas, Weft can now pay that fee itself (within your limits), so a fetch no longer fails just because the merchant left gas uncovered.
Your Tempo wallet now shows each token separately. The balance card breaks down your Tempo holdings per token — including USDC.e — so you can see exactly what you hold instead of a single lumped figure.
Your spending numbers now match everywhere you look. The 30-day spend, average cost per fetch, and balance figures on your home, purchases, and wallet pages are all computed the same way, so a number no longer differs from page to page. Internal wallet top-ups no longer get counted as purchases or skew your averages.
Tiny payments show their real amount instead of $0.00. Weft runs on micro-payments, so sub-cent charges now display down to the fraction of a cent (for example "$0.0005") across your purchase list, details, history, and dashboard — no non-zero amount rounds away to zero.
Transaction links always point to the right block explorer. Each on-chain link is now derived from the network the transaction actually happened on, so a Tempo payment opens the Tempo explorer and a Base payment opens Base — no more dead or wrong-network links.
A payment only shows as "Settled" once it's confirmed on-chain. Weft now verifies the settlement transaction on the blockchain before marking a merchant payment settled, and a background sweep finishes settling any that were still confirming — so your history reflects money that has actually moved, not just a merchant's claim.
Edit your spending limits directly from the policy page. Your per-transaction, daily, and weekly caps live on a self-serve edit form; if the page was open while something changed, saving now warns you cleanly instead of silently overwriting. Cap validation messages read in plain language ("the weekly limit") instead of raw field names.
Filter and search your purchases across all pages. Purchase filters and search now run on the server, so results and counts cover your full history — not just the current page — and the view stays in sync when you use browser Back and Forward.
Weft can now pay merchants in the exact token they ask for. When a paid resource demands a specific settlement token your wallet doesn't already hold, Weft converts your balance into that token at payment time and settles — no more payments that fail simply because the merchant named a different-but-equivalent currency.
Payments that can't succeed now fail instantly and clearly, instead of silently moving money first. If a merchant asks for a token Weft genuinely can't settle, the payment is refused up front with a clear reason and your funds stay put — previously a mismatch could move money that then couldn't be spent. Error messages also now name the exact asset that was checked.
Your balance is protected from bad exchange rates. Currency conversions during a payment are now capped on both sides: Weft refuses any conversion priced worse than a hair off fair value, and every conversion counts against the same daily and per-payment limits as a top-up, so an unusual market can never drain your wallet on a tiny charge.
Filter your search by price, type, and protocol. Search now takes a typed filters object on both the API and the MCP search tool — cap or bound the price, restrict to a resource type (API, agent, MCP tool, webpage, feed, document, dataset), or to a payment protocol (x402, a2a, mpp, erc-8004). Every filter is optional, and a filter Weft doesn't recognise is rejected outright rather than quietly ignored, so a typo can never silently widen your search.
Prices in filters are exact, by design. You give a price as a decimal string ("5.00") or as an integer in micro-dollars — one or the other, never both. Writing a plain number is refused on purpose: it's the difference between "under $5" and "under 5 micro-dollars", and that mistake used to return an empty result set with no explanation.
Search results carry richer provenance. Results now include the provider, capability, and endpoint behind each hit, plus a query trace id you can quote to us when something looks wrong.
Your wallet survives an interrupted setup. If wallet provisioning is cut off midway, Weft can now recover and adopt the wallet that was actually created, instead of leaving you stranded with a half-finished setup.
Your balance is honest across networks. A wallet is now clearly all-mainnet or all-testnet, and payments are checked against the wallet's environment before signing — so real funds can't accidentally settle on a test network, and vice versa. Real-money wallets that used to show "Tempo unavailable" now display their true balance.
Low balance on one network no longer blocks a payment. When you pay a Tempo charge you can't fully cover, Weft now moves just enough of your own Base USDC across for you automatically, bounded by the spending policy you already set. If the top-up can't complete, your original payment simply fails as before — nothing riskier.
Paid fetches are reliable even when requests overlap. If two identical paid requests run at the same moment, both callers now get the content they paid for. Previously a rare race could drop one caller's stored result.
You always get what you paid for. If our content storage ever hiccups right after a payment settles, Weft now still returns the content you fetched along with the on-chain receipt, instead of failing the request. The payment's transaction record is also saved before anything else runs, so a successful charge is never left without its receipt.
Clearer errors when a paid service misbehaves. When an upstream provider returns an unexpected response, Weft now replies with a precise error that keeps the full details (the reason, plus your policy and balance) intact, rather than a generic one that could get stripped in transit.
One balance for everything. Your wallet now shows a single total that combines your Base USDC and your Tempo dollars, so you see your real spending power at a glance instead of two separate numbers. The breakdown is still there, and if one network is temporarily unreachable Weft tells you plainly rather than showing a misleading total.
New account setup is reliable again. A recent change on our wallet provider's side had started rejecting new sign-ups mid-setup, so some new users hit an error when their wallet was being created. That's fixed — creating a fresh account and wallet now works end to end, and accounts created before the fix are automatically reconnected.
Streamlined payment networks. Weft now settles payments over Base USDC (x402) and Tempo (MPP). We retired the earlier Monad option to keep the paid-fetch experience simpler and more predictable.
Use Weft inside ChatGPT. Weft is now available as a ChatGPT app — search for paid services right in the chat and see results as a rich gallery showing each provider, price, payment network, and trust signals, with a one-tap "Use via Weft" to pay and fetch under your own spending rules.
A clearer way to connect your AI tools. The "Connect" page got a full redesign: pick your app from a sidebar — Claude Code, Claude Desktop, Claude.ai, ChatGPT, Cursor, and more — and get the exact one-click install, copy-paste link, or config tailored to it, instead of hunting through tabs.
Wallet setup is now more reliable. Wallet provisioning is now idempotent — if something goes wrong mid-setup and you try again, Weft will reconnect to your existing wallet instead of failing with a generic error.
Sign in with Google now links to your existing account. If you previously signed up with email and password, signing in with Google will now connect the two instead of showing an error. Your existing account, history, and wallet stay intact.
Signing out works consistently. A crash that could occur when signing out while already signed out has been fixed. Sign-out now behaves correctly in all cases.
Safer spending, always. We tightened the guards that enforce your spending caps and idempotency keys, so rapid or concurrent requests can never accidentally double-spend or return an error when a replay was the right answer.
Smoother sign-in. If you're already signed in and revisit the login page, you'll now see a calm, informational message instead of a red error.
More reliable payments and wallet checks. Behind the scenes we added timeouts to our wallet and edge-network connections, so a slow upstream service can no longer leave your requests hanging.
Tidied-up accounts. We removed the old public user-profile pages, which were no longer in use.
Pay for online services automatically. Give your AI agent a Weft wallet and it can buy data, API calls, and other paid services on your behalf — handling the payment handshake end to end so you don't have to wire up billing for each provider.
Stay in control of spending. Set per-transaction, daily, and weekly limits, and block specific recipients. Weft enforces these caps before any payment goes out, so an agent can never overspend.
Connect your AI tools in one click. Plug Claude.ai, Claude Code, or Cursor into Weft and they instantly gain three abilities: check your balance, search for services, and fetch-and-pay for content — all governed by your spending rules.
Fund and watch your wallet. Create a managed wallet across supported networks, top it up, and see live balance plus a full history of everything your agent has bought.
Sell your own agents and APIs. Set up a seller profile, list your agents, data APIs, MCP servers, datasets, or models, and price each operation. Buyers (human or AI) can discover and pay for them.
Track your earnings. A seller dashboard shows settled payments, daily volume, top payers, network breakdown, and revenue by endpoint — plus webhooks so your own systems get notified the moment a payment lands.
Publish a public storefront. Each agent, API, and provider gets a shareable public page with machine-readable API docs so others can find and integrate with what you offer.