Release 0.76.0
The browser SDK, server SDK and VoiceID plugin use 0.76.0. Apply the D1 migration 0091_linked_device_email_otp_own_factor.sql, then deploy the server and the Wallet host together, then the apps.
Since 0.75.0:
- The Email OTP enrollment of a wallet keeps one factor for each Email OTP method, each with its own nonce, wrapped factor and pinned unlock key. The release picks the factor whose nonce the device sends. A revoke removes the method's factor in the same commit, also when a linked device is revoked.
- Google recovery works on a new device again. The recovery code opens the seed, and the device enrolls its own Email OTP factor for the recovered method. The server releases no factor, and the factors of the wallet's other devices stay.
- A linked device enrolls its own Email OTP factor, also on a wallet that has Email OTP. The code goes to the wallet's Email OTP address, which the new device must enter, and the link adds the device's factor to the enrollment.
- On a device that holds no Email OTP factor for the wallet, a Google or email sign-in sends no code. It fails with the new code
email_otp_device_factor_missing, which carries the wallet ID and the email hint inaccount. The hosted menu then offers "Link this device" and "Recover wallet". - An Email OTP owner sees the Wallet's email code prompt when removing a sign-in method or a linked device, and when moving a wallet to another region. The prompt stayed hidden, so the request failed after its timeout.
Breaking changes:
/wallets/recovery/email-otp/releaseis removed./wallets/recovery/email-otp/verifyalso returnsproviderSubjectandverifiedEmail./wallets/recovery/google-email-otp/finalizerequiresemailOtpEnrollment, which is now the enrollment material itself, with nokindwrapper.- The device-link route
/email-otp-base-factoris removed, with the base method chooser.QRCodeScanneranduseDeviceLinkingno longer takeonEmailOtpBaseFactorRequired. - A device-link approval and target preparation for Email OTP carry only
targetEmail. The challenge start has noworkerEphemeralPublicKey65B64u, the verification result has nofactorRelease, and an Email OTP target registration always carriesemailOtpEnrollment. The add-methodemailOtpTargetof kindexisting_enrollmentcarriesenrollmenttoo. - Email OTP wallets made by 0.75.0 do not open after the upgrade, because the stored enrollment changed.
Compatibility:
- Apply the D1 migration
0091_linked_device_email_otp_own_factor.sqlbefore the server. It drops the open device-link Email OTP grants, so such a link asks for a new code. An upgrade from 0.74 applies0090_email_otp_code_seal.sqlfirst. - A Wallet host at 0.76.0 needs a server at 0.76.0, and the reverse: the Email OTP recovery and device-link contracts changed on both sides.
- An app at 0.76.0 needs a Wallet host at 0.76.0. A Wallet host at 0.76.0 accepts an app at 0.74 or 0.75.
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signstay at 0.72.0. They have no change.- The Wallet iframe protocol stays 25.0.0, and the VoiceID host API stays 11. Install all three packages at 0.76.0.
Release 0.75.0
The browser SDK, server SDK and VoiceID plugin use 0.75.0. Apply the D1 migration 0090_email_otp_code_seal.sql, then deploy the server and the Wallet host together, then the apps.
Since 0.74.0:
- Email OTP is a device factor. The device keeps a random input. The server keeps a factor nonce and the factor, encrypted under the RFC 9497 POPRF output (ristretto255-SHA512) for that input. The server cannot compute that output without the device, so neither the server nor the device alone can open an Email OTP wallet. The POPRF
infonames the wallet, the enrollment and a nonce that the server picks. - The server stores each Email OTP code encrypted (AES-256-GCM, bound to its challenge), never in plain text. The key comes from the existing seal root (
SIGNING_SESSION_SEAL_ROOT_SECRET_B64U), so no new secret is needed. - The warm session that survives a page reload is sealed with the same POPRF in place of Shamir three-pass. The device keeps a random input and the session secret under AES-256-GCM, with a key from the POPRF output. The POPRF
infonames the wallet and the threshold session. A restore is one server call. A new WASM module,oprf_runtime, serves the Wallet workers and the Gateway, and theshamir3pass_runtimemodule and its nested worker are gone. - The server and the SDK refuse a bad signing count, signer slot, participant ID, step-up expiry or key-envelope binding instead of filling in a default. A NEAR RPC error such as
{"InvalidTxError":"Expired"}keeps its name. A Wallet Session whose authority row is gone reads as ended, not corrupt. - The Email OTP export and rotation no longer send the seal group as
orgIdin/wallet/unlock/challenge.
Breaking changes:
/wallet/email-otp/registration/sealis now/wallet/email-otp/registration/factor, which takes a blinded element. The factor release, the Ed25519 step-up material recovery and the Google recovery release takedeviceFactorsin place of a worker public key, and they return anemail_otp_factor_evaluation_v1./wallet-session/seal/apply-server-sealand/wallet-session/seal/remove-server-sealare now one route,POST /wallet-session/seal/evaluate, which takes a blinded element. The server allows 30 evaluations a minute for each wallet and threshold session. The seal protocol in/.well-known/webauthnis{ algorithm: 'ristretto255-sha512-poprf-v1' }with nogroupId. A deploy config that checkssigningSessionSeal.algorithmmust change it and dropgroupId.@seams/wallet-serverexportscreateSigningSessionSealEvaluatorin place of the seal cipher adapters and their Shamir three-pass types.parseSigningSessionSealRootConfigis removed.createSigningSessionSealOptionskeeps its input.- The Email OTP login options have no
groupId. - A device-link target cannot reuse the wallet's Email OTP factor, and Google recovery of an Email OTP wallet works only on a device that holds the factor. Choose a passkey or VoiceID for the new device, or use the recovery code. A later release gives each device its own factor.
- Email OTP wallets from an earlier version do not open after the upgrade. The Wallet deletes the sealed warm sessions of an earlier version when it reads them, so each open session needs one unlock.
Compatibility:
- Apply the D1 migration
0090_email_otp_code_seal.sqlbefore the server. It drops the open Email OTP challenges, so a code that is open at the deploy fails and the user asks for a new one. - A Wallet host at 0.75.0 needs a server at 0.75.0, and the reverse: a 0.75.0 Wallet host refuses a server with another seal protocol (
sealed_refresh_parity_mismatch), and a 0.75.0 server has no route for the Email OTP registration and the seal of a 0.74 Wallet host. - An app at 0.75.0 needs a Wallet host at 0.75.0. A Wallet host at 0.75.0 accepts an app at 0.74: the app config names the seal mode only, and the host ignores the seal protocol that an older app sends.
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signstay at 0.72.0. They have no change.- The Wallet iframe protocol stays 25.0.0. Install all three packages at 0.75.0.
Release 0.74.0
The browser SDK, server SDK and VoiceID plugin use 0.74.0. Deploy in this order: the server, then the Wallet host, then the apps.
Since 0.73.1:
- A passkey account sync asks for the app's Wallet Session budget (
signingSessionDefaults), as unlock does. The client sendswalletSessionBudgetto/sync-account/verify. Before, a sync always got 5 signatures and 24 hours. - A linked device reads the count of its session from the server after the link. Before, it recorded 5. The server grants a linked device 100 signatures and 15 minutes, capped by the tenant's maximum.
- The server picks the budget profile from the authority. A linked device gets the linked-device budget on every path: unlock, budget refresh, ECDSA activation and VoiceID.
- A retry of a session issue replays after the tenant changed the maximum. A stored maximum that does not parse changes only the grant (5 signatures and 24 hours, with a log line), not the rest of the environment.
signingSessionDefaultsandsigningSessionvalues must be non-negative integers. Another value fails the config or the login. Before, it disabled the warm session with no error.- The client checks every grant with one rule: 1 ≤ granted ≤ requested. VoiceID and the ECDSA session now check it too.
- Email OTP counts each attempt before it compares the code. Before, parallel guesses on one challenge could pass the limit of attempts.
- NEP-413 signs the app's exact message, recipient and callback URL. Before, the SDK trimmed the whitespace, so a message with a leading or trailing space failed.
- An ECDSA step-up is admitted only while the signer rows that it read still hold. A failed Email OTP unlock names a wallet in its webhook only when the request has a valid wallet ID.
- VoiceID unlock answers 402 for a billing failure. A NEAR lookup failure after the session is issued no longer loses the session: the response says
near: unavailable. - The lock after a signing failure logs each step that failed.
Breaking changes:
- A server that composes
AuthorizationServiceitself passes awalletSessionLimitsport. The service refuses a composition with none.@seams/wallet-server/cloud-hostexportscreateProjectEnvironmentWalletSessionLimitsandUNAVAILABLE_WALLET_SESSION_LIMITS(for a server with no Console). provisioningDefaults.{tempo,evm}andRegistrationHooksOptions.signerOptionshold onlyenabled. Nothing read theirsigningSession. Set the budget withsigningSessionDefaults. A Wallet host at 0.74.0 ignores thesigningSessionthat an older app sends.- The unused types
EcdsaSessionPolicyandThresholdEcdsaBootstrapSessionPolicyare removed from@seams/wallet-server. @seams/wallet/advancedexportsDEFAULT_TENANT_WALLET_SESSION_LIMITS,parseWalletSessionLimitsandWalletSessionLimits.
Compatibility:
- A Wallet host at 0.74.0 needs a server at 0.74.0: a 0.73 server refuses the
walletSessionBudgetof a sync. A server at 0.74.0 accepts a Wallet host at 0.73. - An app at 0.74.0 needs a Wallet host at 0.74.0: a 0.73 Wallet host refuses a config with no
provisioningDefaults.signingSession. A Wallet host at 0.74.0 accepts an app at 0.73 or 0.72. @seams/voice-verdict,@seams/voice-captureand@seams/voice-signstay at 0.72.0. They have no change.- The Wallet iframe protocol stays 25.0.0. Install all three packages at 0.74.0. No database migrations.
Release 0.73.1
The browser SDK, server SDK and VoiceID plugin use 0.73.1. It has no breaking change since 0.73.0.
Since 0.73.0:
- The transaction receipt keeps the heading and description of each stage on one line on a phone. Before, a stage that took two lines followed a stage that took one, and the card changed height while it signed and sent the transaction.
- The Japanese text of the stages is shorter, for example 署名中 and 送信中 in the headings. Three English descriptions are shorter: "Sending it to the network.", "Sent. Waiting for the receipt." and, for a signed message, "Nothing was sent to the network."
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signstay at 0.72.0. They have no change.- No server or protocol change. The Wallet iframe protocol stays 25.0.0, and a Wallet host at 0.73.1 works with a server at 0.73.0.
- Install all three packages at 0.73.1. No database migrations.
Release 0.73.0
The browser SDK, server SDK and VoiceID plugin use 0.73.0. A Wallet host at 0.73.0 needs a server at 0.73.0, so deploy the server first.
Since 0.72.0:
- The app sets the Wallet Session budget:
signingSessionDefaults: { ttlMs, remainingUses }in the provider config, orsigningSessionin the options of one login. Before, an app could only lower the count below 5. See Set the session budget. - The server grants the request up to the maximum of the project environment. A tenant sets that maximum in the Seams Console. With no setting, the maximum is 100 signatures and 30 days. The highest maximum is 1,000,000 signatures and 30 days. With no request, a session gives 5 signatures and lasts 24 hours, as before.
- Read the granted budget from the unlock result (
signingSession.remainingUsesandsigningSession.expiresAtMs). It can be lower than the request. - The budget goes to the server as
walletSessionBudgeton the passkey, Email OTP and VoiceID unlock and on registration setup. The session that registration gives lasts the granted time. Before, its ECDSA session lasted 5 minutes. - A linked device gets its own session of 100 signatures and 15 minutes, also capped by the tenant's maximum.
- A server that you host reads the maximum through the Console binding. The static Console binding takes the optional
deployment.walletSessionLimits({ maxTtlMs, maxRemainingUses }). - The Wallet lists JPYC (JPY Coin) on Ethereum, with JPYC's coin mark. Like WBTC and cbBTC, it shows only while the wallet holds it.
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signstay at 0.72.0. They have no change.- The Wallet iframe protocol stays 25.0.0. A server at 0.73.0 accepts the requests of a 0.72.0 Wallet host and ignores their old counts. A Wallet host at 0.73.0 does not work with a server at 0.72.0.
- Install all three packages at 0.73.0. No database migrations.
Release 0.72.0
The browser SDK, server SDK and VoiceID plugin use 0.72.0. It has no breaking change since 0.71.0.
Since 0.71.0:
- One unlock gives a Wallet Session of 5 signatures. Before, it gave 3. After the last one, the Wallet asks for a step-up, as before. A session that exists keeps its count until the next unlock.
- The default is one constant now,
DEFAULT_WALLET_SESSION_REMAINING_USES. The client defaults and the server read it.signingSessionDefaults.remainingUsescan still lower the count for an app, and it cannot raise it above the default. @seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.72.0. They have no change.- The Wallet iframe protocol stays 25.0.0. The count comes from the Wallet host and the server, so an app at 0.71.0 gets 5 signatures from a host at 0.72.0.
- Install all three packages at 0.72.0. No database migrations.
Release 0.71.0
The browser SDK, server SDK and VoiceID plugin use 0.71.0. It has no breaking change since 0.70.6.
Since 0.70.6:
- A wallet with no VoiceID can set up its voice from an app again. Since 0.70.1, a VoiceID activation starts when VoiceID has loaded, with no press on "Activate VoiceID". But the activation still needed a press in the Wallet frame, and a press in the app does not count there. So the first voice setup of a new wallet failed with "Open the Wallet and select Record to use VoiceID." Now the enrollment card opens and waits for "Record sample 1". For Speak-to-Sign, the three samples are still the enrollment of VoiceID and of the verifier.
- A VoiceID activation that records at once shows "Activate VoiceID" and waits, if the Wallet frame has no press. This is the case of a browser that already holds a voice enrollment of the wallet. Before, it failed with the same message.
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.71.0. Their behavior has no change. Their documents and tests follow verifier 0.15.0 of Speak-to-Sign: a verdict that passes has no heard words.- The Wallet iframe protocol stays 25.0.0. The fix is in the Wallet host, so an app at 0.70.6 gets it from a Wallet host at 0.71.0.
- Install all three packages at 0.71.0. No database migrations.
Release 0.70.6
The browser SDK, server SDK and VoiceID plugin use 0.70.6. It has no breaking change since 0.70.5.
Since 0.70.5:
- An app can press a control of a Wallet card in its dialog for the person, for example from a voice command. The new step event
onControlsgives the apppress, which takesretry,minimize,backorclose.retryis the card's button while it starts a recording: Record or "Try again". The Wallet never presses another action for the app, such as "Confirm with passkey" or "Done". A press counts as a click does: only on a card that is whole in view and not covered. - The card's stage has the new value
waiting: the card waits for a press on Record or "Try again", and its microphone is off. Before, the card reportedrecordingthere. An app that pauses its own speech recognition while the Wallet records can now listen again while the card waits. - After "Try again" on a failed check, the card records at once. Before, it showed Record, and the person pressed a second time.
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.70.6. Only@seams/voice-capturehas a change.- The Wallet iframe protocol stays 25.0.0. An app at 0.70.5 works with a Wallet host at 0.70.6: it does not see the stage
waiting. A press of an app at 0.70.6 needs a Wallet host at 0.70.6. - Install all three packages at 0.70.6. No database migrations.
Release 0.70.5
The browser SDK, server SDK and VoiceID plugin use 0.70.5. It has no breaking change since 0.70.4.
Since 0.70.4:
- The result card of a spoken request stays whole in the Wallet frame. Before, a result with a TEE receipt showed about 10 px too high: its top was cut off, and an empty band showed under it. The result is taller than the view before it, and it shows while the frame still grows. The card put the focus on "Done" with a plain focus, so the browser scrolled the frame. Each focus of the card now keeps the scroll position. The VoiceID cards have the same change.
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.70.5. Only@seams/voice-capturehas a change.- The Wallet iframe protocol stays 25.0.0. An app at 0.70.4 works with a Wallet host at 0.70.5.
- Install all three packages at 0.70.5. No database migrations.
Release 0.70.4
The browser SDK, server SDK and VoiceID plugin use 0.70.4. It has no breaking change since 0.70.2. No Wallet package has the version 0.70.3; the voice libraries have it.
Since 0.70.2:
- When the owner edits a note, the Wallet ends the summary request of the older text, if only
prepareasked for it. The summary route then ends its request to the model. Before, each pause in the typing left one model call. - The Wallet does not keep a summary that breaks a rule of
summaryRefusal. Before, such a summary stayed for 5 minutes. When the server's model gives no summary that a person can say, the result ofspoken.noteis now the refusalsummary_not_speakable. Before, it wasunavailablewithsummary_unavailable. - The second request of the summary writer asks the model to leave out a number that is inside a name, such as "3D". Before, it asked for the number as words. Speech-to-text writes a said "three D" as "3D" again, and the verifier reads such a word digit by digit, so the words did not match.
- A sent or signed result of a spoken request has
attestationUrl: the OutLayer record of the verifier call, with its TDX quote, or null when the Wallet got none. An app can show the proof with its own result. - The last repair of the summary writer keeps a quantity. A word that starts with a digit or a currency sign ("10km", "$40m", "2nd") is not cut to its letters: the result is
summary_not_speakable. @seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.70.4.writeSummarytakes{ signal }, andcreateSummarySourcehas the two changes above. The docs ofwriteSummarysay that a summary is a draft: only a community note has the verifier's check that the summary fits the text.- The Wallet iframe protocol stays 25.0.0. Upgrade the app before the Wallet host: an app at 0.70.2 or older reads a result with exact keys, so it refuses the success result of a spoken request from a Wallet host at 0.70.4, which has
attestationUrl("Speak-to-Sign … result is invalid", after the transaction was sent). An app at 0.70.4 reads the result of an older Wallet host too. - Install all three packages at 0.70.4. No new server setting, and no database migrations.
Release 0.70.2
The browser SDK, server SDK and VoiceID plugin use 0.70.2. It has no breaking change since 0.70.1.
Since 0.70.1:
- New call
seams.near.spoken.prepare({ note? }). Call it when your form is on the screen. The Wallet loads its Speak-to-Sign code, and withnoteit asks the Seams server for the summary of that text. Thespoken.notecall that follows shows its card sooner: before, the card waited 2.5 to 3 s for these steps. The call shows no card and signs nothing. See Prepare a spoken request. - The Wallet keeps one summary for each note text for 5 minutes. A second try of a note shows the same sentence. Before, each try asked the model again, and the sentence could change.
- A note request asks for its summary while the Speak-to-Sign code loads. Before, it asked after the load.
- The summary route writes a name with a number in it with one request to the model: "COP31" becomes "COP". Since 0.70.1 such a name got a second request, which added about 0.5 s. A number that stands alone still gets the second request.
- The summary code is now a general part of the voice libraries, and the Wallet uses it:
@seams/voice-verdict0.70.2 hassummaryRefusal, the rules of a spoken summary.@seams/voice-sign/server0.70.2 haswriteSummary. The Seams summary route calls it.@seams/voice-sign/summaryis a new entry withcreateSummarySource. The Wallet keeps its summaries with it.@seams/walletnow depends on@seams/voice-sign.@seams/voice-capture0.70.2 has no change.
- The Wallet iframe protocol stays 25.0.0. The new message
PM_SPEAK_TO_SIGN_PREPAREis optional: an app at 0.70.1 works with a Wallet host at 0.70.2. An app at 0.70.2 that callsprepareneeds a Wallet host at 0.70.2. - Install all three packages at 0.70.2. No new server setting, and no database migrations.
Release 0.70.1
The browser SDK, server SDK and VoiceID plugin use 0.70.1. It has no breaking change since 0.70.0.
Since 0.70.0:
- The result card of a transfer, a transaction or a message names a long receiver in its short form: an implicit account of 64 characters shows as "To 3e2210…7e55". A named account stays in full. The request that the person approves still names the full account. A request with no receiver shows its own text; before, the caption said "To null".
- The summary route of a note no longer fails for a name with a number in it, such as "COP31". The second request to the model names the word and says to write its number as words or to leave it out. If both replies still have such a word, the route keeps the letters of each word ("COP31" becomes "COP"). A number that stands alone still gives
summary_not_speakable. The verifier still checks that the summary fits the note. @seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.70.1. In@seams/voice-capture, a word that is longer than the line breaks inside the card. Before, such a word made the receipt and the button wider than the card.- The Wallet iframe protocol stays 25.0.0. An app at 0.70.0 works with a Wallet host at 0.70.1.
- Install all three packages at 0.70.1. No database migrations.
Release 0.70.0
The browser SDK, server SDK and VoiceID plugin use 0.70.0. An app has no breaking change since 0.69.2. A server operator must read the first item.
Since 0.69.2:
- The Speak-to-Sign verify route checks the TDX quote of each verifier call before it sends a verdict. The verdict key is bound to the OutLayer project, so every version of the project signs with it. The quote names the WASM that ran.
- The quote must bind the output to the WASM hash
SPEAK_TO_SIGN_OUTLAYER_VERSION_KEYand to the project. Intel must sign the quote, andworker.outlayer.testnetmust approve its five measurements. - The route reads the record of the call from OutLayer. It tries once each second for up to 15 s.
- If a check fails, or the record does not come in time, the route sends
failedwithattestation_<code>, for exampleattestation_wasm_not_acceptedorattestation_unavailable. - A
verdicthasattestationUrl, the URL of the OutLayer record. ItstiminghasattestationMs. - Set
SPEAK_TO_SIGN_OUTLAYER_VERSION_KEYto the WASM hash of the verifier version that the server calls. The route refuses a quote of another WASM. See Experimental: Speak-to-Sign.
- The quote must bind the output to the WASM hash
- Optional
SPEAK_TO_SIGN_OUTLAYER_VENICE_VERSION_KEY: the WASM hash of thevenicebuild of the verifier (0.14.0 and later), which sends the clip to Venice AI. With it andSPEAK_TO_SIGN_NEAR_AI_API_KEY, the server checks NEAR AI once a minute. Venice serves a call only while NEAR AI or its Whisper is down. Only the server chooses; the page cannot.- Venice AI runs in no TEE, so it can hear the recordings that it gets. The receipt of such a call shows "Speech to text: Venice AI, outside a TEE".
- The quote check accepts only the WASM hash of the build that the route called. The verdict event has
transcriber(nearaiorvenice). - The Wallet card reads a verifier version with a build name, such as
0.14.0+venice.
- New refusal
transcriber_not_attested(verifier 0.14.0 and later): the verifier could not check the TEE of NEAR AI's gateway, or Whisper's reply did not open with the checked key. No audio went out after a failed check. Before, the Wallet read this refusal asverdict_mismatch. The card says that the recording is not the cause, and offers Try again. An app before 0.70.0 cannot read this reason. - The verify route gives each verifier run 40 s (was 15 s) and waits 60 s for the answer (was 30 s). The verifier waits up to 30 s for each model. A slow NEAR AI Whisper now gives a signed refusal (
transcriber_unavailable), not a run that OutLayer stopped. - The paid and signed Speak-to-Sign cards show a TEE receipt: "TEE receipt · Intel TDX · OutLayer", the voice of the account, the words, the summary check of a note, the verifier version of the verdict, and "View TDX quote". If the server sends no
attestationUrl, the receipt has no link. - VoiceID activation goes to the first recording as soon as VoiceID has loaded. Before, the person pressed "Activate VoiceID" and then "Record sample 1".
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.70.0. A result card of@seams/voice-capturetakesreceipt(PhraseReceipt): a title, a source, checked rows and a link, between the caption and the detail line. In@seams/voice-sign, the default run limit is 40 s (was 15 s) and the defaulttimeoutMsis 60,000 (was 20,000).- The Wallet iframe protocol stays 25.0.0. An app at 0.69.x works with a Wallet host at 0.70.0. This release needs no new verifier version.
- Deploy the server and the hosted Wallet together. Install all three packages at 0.70.0. No database migrations.
Release 0.69.2
The browser SDK, server SDK and VoiceID plugin use 0.69.2. It has no breaking change since 0.69.0.
Since 0.69.0:
- The Speak-to-Sign card's check reads "Checking what you said". With the verifier step after it, the line now fits the card.
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.69.2. In a form card (a card with a context, such as a note), the check's line slides into the microphone's circle, and the microphone turns into a tick. A card with no context keeps the ring and the seal.- The Wallet iframe protocol stays 25.0.0. The server and the verifier (0.13.0) do not change. An app at 0.69.0 works with a Wallet host at 0.69.2.
- Install all three packages at 0.69.2. No database migrations.
Release 0.69.0
The browser SDK, server SDK and VoiceID plugin use 0.69.0. This release has breaking changes for an app that signs community notes with Speak-to-Sign.
spoken.notetakes nosummary. The app gives only the note. The Seams server writes the summary that the author reads aloud, so an app cannot put other words in the author's mouth.- New route
POST /speak-to-sign/v1/summaryon the Seams server. It reads the Wallet Session as the verify route does, takes{ text }(1 to 280 characters), and asksQwen/Qwen3.8-27Bon NEAR AI for one plain sentence. It checks the reply with the spoken rules and asks once more if a rule breaks. Failures aresummary_unreachable,summary_failedandsummary_not_speakable. - The route exists only when the server has
SPEAK_TO_SIGN_NEAR_AI_API_KEY.SPEAK_TO_SIGN_SUMMARY_MODELis optional. - Without a summary, the result is unavailable with
summary_unavailable, and no card opens. - The voice card shows the note above the summary.
- New route
- The verifier is 0.13.0 (
1b7771544f3bcf5b3ba9e9bf3ac611f6abad21e7538f67b2666fd8aafc8cf930). The Wallet sendsnote_textin place ofnote_sha256_hex. The verifier hashes the note and checks that the summary gives the main point of the note. A summary that does not fit is refused withsummary_mismatch, and no answer withsummary_check_unavailable. Set the 0.13.0 version key on the server before this Wallet ships. add_noteandrate_notecarry their verdict (verdict_json,signature_b64) to the notes contract, as a bet carries it in itsmsg. The co-signer refuses a call that carries another verdict. A note call without the two arguments, from an older Wallet, is still admitted.- Wallet steps: an app can ask for Minimize on an anchored card (
minimize: trueon the presentation, or theminimizeprop ofWalletStep). The card draws Minimize beside Close, also during the check.- A press hides the card and keeps its request. The SDK runs
onMinimize(restore).restore()shows the card again, and the card also shows again when it needs the person. Both runonRestore. - The card reports its stage (
recording,checkingwith done and total,result) withonProgress. It carries no approval data. - A Wallet host before 0.69.0 refuses
minimize. Deploy the hosted Wallet before an app sends it.
- A press hides the card and keeps its request. The SDK runs
@seams/voice-verdict,@seams/voice-captureand@seams/voice-signare 0.69.1. The voice card:- shows a context (the note) above its sentence, and a form card shows Record in place of the voice line until it records;
- takes a
colorsprop (withsurface3, the colour of the check segments that have not run yet), and the voice line can draw pixels in place of sine waves; - turns the Beam off with
voiceBeam: 'none', and takes the theme's Beam from the first frame. Before, it showed the default Beam until recording started; - shows a result as one group: the seal, the headline, the details and the button. The seal takes
voiceSeal, elsevoiceLine, ofappearance.theme.colors; - places its grid rows by name. A host that sets its own rows must use the names.
- The Wallet iframe protocol is 25.0.0: the note message has no summary. An app at 0.68.1 does not work with a Wallet host at 0.69.0, or the reverse.
- Deploy in this order: verifier 0.13.0 on OutLayer, then the server with its version key and the NEAR AI key, then the SDK and the hosted Wallet together.
- Install all three packages at 0.69.0. Speak-to-Sign stays experimental and NEAR testnet only. Do not use it with real funds. No database migrations.