@byollm/protocol 0.1.0-alpha.62 → 0.1.0-alpha.64

package/ABOUT-SHORT.md

@@ -0,0 +1,7 @@
+BYOLLM – Bring Your Own LLM – lets you use your own AI on websites you authorize. A small program installed on your machine lets you use your own models and subscriptions on any BYOLLM-integrated site, including new models the moment you get access – no site updates required. BYOLLM Cloud connects sites to your devices with end-to-end encryption, so no one, including us, can see your data.
+
+<!-- lede ends here. Surfaces with room show the tail below; the welcome screen shows the lede only. -->
+
+Sites can charge you less because you bring your own – see why that matters →. Teams can optionally share the free or metered services on their devices with people they name. Personal subscriptions are never shared.
+
+**Learn more →**

package/ABOUT.md

@@ -0,0 +1,60 @@
+# About BYOLLM
+
+**What BYOLLM is**
+
+BYOLLM lets you use your own AI on websites. You install one small program on
+your computer. Then, websites that support BYOLLM can use the AI you already
+have — a free model running on your machine, or an AI service you already pay
+for — instead of the website paying for AI and passing the cost to you.
+
+**Why it matters**
+
+For you:
+
+- Your favorite model, everywhere you go.
+- New models the moment you get them – not when a site gets around to adding
+  them.
+- Encrypted end-to-end. Your prompts go to your own device; byollm.cloud can't
+  read them.
+- Sites never learn which model you use, and your subscriptions are never
+  shared.
+- Pay less. Sites that don't pay for AI can charge you less – or nothing.
+
+For sites and developers:
+
+- Zero AI bills. Your users bring their own compute.
+- No floating money – you don't pay LLM bills up front and hope to collect
+  later, and you never ask people to prepay just to try you.
+- Free trials that cost you nothing to offer.
+- Ship the AI features you kept private for fear of the API bill.
+- One small integration. Your users choose the models.
+
+**Your device**
+
+The `byollm` program runs on your computer. It knows which AI services you have
+set up: free open-source models on your machine, metered services you pay per
+use, or your own subscriptions like Claude Pro/Max. When a website you have
+enabled sends work, your device runs it with the service you chose. Your
+prompts are encrypted end-to-end to your own device. byollm.cloud passes them
+along and cannot read them.
+
+**Sites**
+
+A website that wants to use BYOLLM says what it needs — "writing help," "chat,"
+and so on. When you connect the site, you pick which of your services answers
+each one. The site never learns which model you use. You can turn a site off at
+any time, and it stops getting your work.
+
+**Teams (optional)**
+
+A team lets you share what runs on your devices with people you name — the free
+open-source models on your machine, or a metered service with a spending limit
+you set. Your subscription accounts (like Claude Pro/Max) are never shared with
+anyone. That is a rule, not a setting.
+
+**byollm.cloud (or your own relay)**
+
+Many sites, many devices, many people. byollm.cloud keeps track of who has
+allowed what and sends each job to the right device. It never sees your
+prompts. If you would rather run this part yourself, the relay is open source —
+you can run your own instead of using byollm.cloud.

package/README.md

@@ -1,5 +1,5 @@
 > [!WARNING]
-> **Alpha (`0.1.0-alpha.62`) — under active development. Don't use this yet.**
+> **Alpha (`0.1.0-alpha.64`) — under active development. Don't use this yet.**
 >
 > Install it deliberately: `npm install @byollm/protocol@alpha`.
 >
@@ -83,7 +83,7 @@
 > packages published and `@byollm/server` did not: a Sigstore
 > transparency-log 409 on its provenance attestation. The workflow's
 > "already published" guard correctly refuses to resume a partial publish,
-> so `0.1.0-alpha.62` is that release, whole.
+> so `0.1.0-alpha.64` is that release, whole.
 >
 > If you run the Supabase adapter, `alpha.21` needs
 > `20260819010000_completed_by_lease_id.sql`: alpha.19 shipped §3.6's

package/dist/index.d.ts

@@ -1,5 +1,19 @@
 import { z } from 'zod';
 
+/** The full description — five sections, plus why it matters. */
+declare const ABOUT = "# About BYOLLM\n\n**What BYOLLM is**\n\nBYOLLM lets you use your own AI on websites. You install one small program on\nyour computer. Then, websites that support BYOLLM can use the AI you already\nhave \u2014 a free model running on your machine, or an AI service you already pay\nfor \u2014 instead of the website paying for AI and passing the cost to you.\n\n**Why it matters**\n\nFor you:\n\n- Your favorite model, everywhere you go.\n- New models the moment you get them \u2013 not when a site gets around to adding\n  them.\n- Encrypted end-to-end. Your prompts go to your own device; byollm.cloud can't\n  read them.\n- Sites never learn which model you use, and your subscriptions are never\n  shared.\n- Pay less. Sites that don't pay for AI can charge you less \u2013 or nothing.\n\nFor sites and developers:\n\n- Zero AI bills. Your users bring their own compute.\n- No floating money \u2013 you don't pay LLM bills up front and hope to collect\n  later, and you never ask people to prepay just to try you.\n- Free trials that cost you nothing to offer.\n- Ship the AI features you kept private for fear of the API bill.\n- One small integration. Your users choose the models.\n\n**Your device**\n\nThe `byollm` program runs on your computer. It knows which AI services you have\nset up: free open-source models on your machine, metered services you pay per\nuse, or your own subscriptions like Claude Pro/Max. When a website you have\nenabled sends work, your device runs it with the service you chose. Your\nprompts are encrypted end-to-end to your own device. byollm.cloud passes them\nalong and cannot read them.\n\n**Sites**\n\nA website that wants to use BYOLLM says what it needs \u2014 \"writing help,\" \"chat,\"\nand so on. When you connect the site, you pick which of your services answers\neach one. The site never learns which model you use. You can turn a site off at\nany time, and it stops getting your work.\n\n**Teams (optional)**\n\nA team lets you share what runs on your devices with people you name \u2014 the free\nopen-source models on your machine, or a metered service with a spending limit\nyou set. Your subscription accounts (like Claude Pro/Max) are never shared with\nanyone. That is a rule, not a setting.\n\n**byollm.cloud (or your own relay)**\n\nMany sites, many devices, many people. byollm.cloud keeps track of who has\nallowed what and sends each job to the right device. It never sees your\nprompts. If you would rather run this part yourself, the relay is open source \u2014\nyou can run your own instead of using byollm.cloud.";
+/**
+ * The first paragraph, which stands alone.
+ *
+ * What the welcome screen shows: somebody deciding whether to trust a site's
+ * button needs the whole idea in one breath, not a page.
+ */
+declare const ABOUT_SHORT_LEDE = "BYOLLM \u2013 Bring Your Own LLM \u2013 lets you use your own AI on websites you authorize. A small program installed on your machine lets you use your own models and subscriptions on any BYOLLM-integrated site, including new models the moment you get access \u2013 no site updates required. BYOLLM Cloud connects sites to your devices with end-to-end encryption, so no one, including us, can see your data.";
+/** The rest, for surfaces with room. Shown before "Learn more →". */
+declare const ABOUT_SHORT_TAIL = "Sites can charge you less because you bring your own \u2013 see why that matters \u2192. Teams can optionally share the free or metered services on their devices with people they name. Personal subscriptions are never shared.";
+/** Both halves, for a surface that wants the paragraph entire. */
+declare const ABOUT_SHORT = "BYOLLM \u2013 Bring Your Own LLM \u2013 lets you use your own AI on websites you authorize. A small program installed on your machine lets you use your own models and subscriptions on any BYOLLM-integrated site, including new models the moment you get access \u2013 no site updates required. BYOLLM Cloud connects sites to your devices with end-to-end encryption, so no one, including us, can see your data.\n\nSites can charge you less because you bring your own \u2013 see why that matters \u2192. Teams can optionally share the free or metered services on their devices with people they name. Personal subscriptions are never shared.";
+
 /**
  * Its members, for anything that has to report what it accepts.
  *
@@ -880,6 +894,16 @@ declare const SIZE_CLASSES: readonly ("small" | "medium" | "large" | "unbounded"
  * bucket survives compression and encoding changes that an exact count does
  * not.
  *
+ * **Two grains, on purpose — ratified 2026-08-28.** Metering a
+ * GB-denominated plan needs real totals, and this is deliberately not where
+ * they come from: exact bytes exist only as increment-only *monthly*
+ * aggregates, and no per-job byte figure is ever persisted anywhere. The
+ * record needs vagueness and the meter needs totals; neither borrows the
+ * other's grain, which is why the consent screen's "roughly how big" stays
+ * exactly true of everything retained about a job. If you are here because
+ * you need a number, the aggregate is the one to reach for — adding a byte
+ * count to this envelope would trade a promise for a convenience.
+ *
  * `unbounded` exists for streamed jobs, which have no size when they start.
  * It is reserved now rather than added later: byollm_009 §8.1 — adding a
  * field to a published envelope is the v2 break all over again.
@@ -891,6 +915,41 @@ declare const SizeClass: z.ZodEnum<{
     unbounded: "unbounded";
 }>;
 type SizeClass = z.infer<typeof SizeClass>;
+/**
+ * The most one envelope may be, in bytes — ratified 2026-08-28.
+ *
+ * A **relay-memory safety rail**, not a plan feature: every tier has the same
+ * ceiling, and differentiating tiers on it would be selling a safety limit as
+ * a benefit. What it bounds is any single job, so no one message can make the
+ * relay hold an unbounded amount of somebody else's memory.
+ *
+ * ## It stores nothing, and that is the design
+ *
+ * Enforced at ingress by refusing before acceptance, in both directions. A
+ * ceiling on what the relay already has in hand needs no schema and no record:
+ * the size is known for the length of the check and then it is gone. This
+ * matters because the alternative — recording a size to enforce a limit
+ * against — is precisely the per-job byte figure the metering ruling exists to
+ * not have.
+ *
+ * ## Measured on the serialised envelope
+ *
+ * The same quantity the monthly rollup counts, deliberately. The relay stores
+ * the serialised envelope and the meter measures what it stored, so a cap on
+ * anything else — the ciphertext alone, the decoded length — would mean the
+ * limit and the bill disagreed about what a byte is, and a job could be small
+ * enough to accept and larger than it was charged as.
+ */
+declare const MAX_ENVELOPE_BYTES: number;
+/**
+ * How big an envelope is, by the one measure that counts it.
+ *
+ * `JSON.stringify` because that is what the store persists and therefore what
+ * the meter measures. Length in UTF-16 code units rather than encoded bytes:
+ * it is the same number the store's own `HSTRLEN` reports, and the point of
+ * this function is that one number answers both questions.
+ */
+declare function envelopeBytes(envelope: unknown): number;
 /** Where the bucket boundaries sit, in characters of payload text. */
 declare const SIZE_CLASS_LIMITS: Readonly<{
     small: 4000;
@@ -2605,4 +2664,4 @@ declare const FetchResponse: z.ZodObject<{
 }, z.core.$strict>;
 type FetchResponse = z.infer<typeof FetchResponse>;
 
-export { AUDIENCES, Audience, BACKENDS, BACKEND_CLASSES, BACKEND_IDS, BackendClass, BackendCost, type BackendDescriptor, type BackendId, BackendIdSchema, CLOCK_ATTRIBUTION_MS, CLOCK_SKEW_WARN_MS, Capability, CapabilityMatrix, ChatMessage, ChatPayload, ClaimRequest, ClaimResponse, ClaimedJob, ClaimedStub, DeliveredResult, ENCRYPTION_KEY_CONTEXT, ENDPOINTS, ENVELOPE_MAX_AGE_MS, ERROR_STATUS, type Endpoint, type EnvelopeContext, EnvelopeDirection, type EnvelopeFailure, FetchRequest, FetchResponse, GRANT_CONTEXT, GRANT_MAX_AGE_MS, GRANT_SIGNED_FIELDS, GeneratePayload, type GrantClaims, GrantRef, type GrantRefusal, HeartbeatRequest, HeartbeatResponse, JOB_KINDS, JobKind, JobOutcome, JobPayload, JobRefused, JobResultCanceled, JobResultError, JobResultOk, JobState, JobStub, KindedPayload, Lease, MAX_CLOCK_SKEW_MS, MAX_PURPOSES, MAX_SUCCESSION_CHAIN, MIN_PROTOCOL_VERSION, MUSTS, MUST_IDS, Manifest, type MatchDaemon, type MatchJob, MatchRefusal, type MatchResult, type Must, type MustEnforcer, type MustId, type MustVerification, type MustVerifiedBy, OFFER_SCOPES, OfferScope, type OpenResult, PAYLOAD_LIMITS, PROTOCOL_PREFIX, PROTOCOL_VERSION, PairPollRequest, PairPollResponse, PairRequest, PairStartRequest, PairStartResponse, type PayloadFor, PublicIdentity, Purpose, REFUSAL_MESSAGES, RESERVED_PURPOSE, RETIREMENT_WINDOW_MS, RefusalReason, ReleaseRequest, ReleaseResponse, RequestSignature, ResultDisposition, ResultProvenance, ResultRequest, ResultResponse, RunMetadata, SIZE_CLASSES, SIZE_CLASS_LIMITS, SUCCESSION_CONTEXT, SUPPORTED_PROTOCOL_VERSIONS, SealedEnvelope, SealedOutcome, type SignatureFailure, SignedGrant, SizeClass, type SpendConsent, StoredKeys, Succession, type SuccessionFailure, type SuccessionWalk, TERMINAL_STATES, type VersionRefusal, WireError, WireErrorCode, WithheldKind, backendDescriptor, backendName, canTransition, canonicalRequest, checkProtocolVersion, classifyCost, cryptoReady, declaredVersion, effectiveOfferScope, fingerprint, generateKeys, grantStatement, isBackendId, isCloudTaggedModel, isJobKind, isLocalHost, isTerminal, keyId, kindsOf, matchAudience, mustsVerifiedBy, open, payloadTextLength, provenanceFor, publicIdentityOf, resolveCost, seal, signGrant, signRequest, signSiteRequest, signSuccession, signWith, singlePurposeManifest, sizeClassCeiling, sizeClassOf, successionStatement, verifyGrant, verifyLink, verifyPublicIdentity, verifyRequest, verifySiteRequest, verifyWith, walkSuccession };
+export { ABOUT, ABOUT_SHORT, ABOUT_SHORT_LEDE, ABOUT_SHORT_TAIL, AUDIENCES, Audience, BACKENDS, BACKEND_CLASSES, BACKEND_IDS, BackendClass, BackendCost, type BackendDescriptor, type BackendId, BackendIdSchema, CLOCK_ATTRIBUTION_MS, CLOCK_SKEW_WARN_MS, Capability, CapabilityMatrix, ChatMessage, ChatPayload, ClaimRequest, ClaimResponse, ClaimedJob, ClaimedStub, DeliveredResult, ENCRYPTION_KEY_CONTEXT, ENDPOINTS, ENVELOPE_MAX_AGE_MS, ERROR_STATUS, type Endpoint, type EnvelopeContext, EnvelopeDirection, type EnvelopeFailure, FetchRequest, FetchResponse, GRANT_CONTEXT, GRANT_MAX_AGE_MS, GRANT_SIGNED_FIELDS, GeneratePayload, type GrantClaims, GrantRef, type GrantRefusal, HeartbeatRequest, HeartbeatResponse, JOB_KINDS, JobKind, JobOutcome, JobPayload, JobRefused, JobResultCanceled, JobResultError, JobResultOk, JobState, JobStub, KindedPayload, Lease, MAX_CLOCK_SKEW_MS, MAX_ENVELOPE_BYTES, MAX_PURPOSES, MAX_SUCCESSION_CHAIN, MIN_PROTOCOL_VERSION, MUSTS, MUST_IDS, Manifest, type MatchDaemon, type MatchJob, MatchRefusal, type MatchResult, type Must, type MustEnforcer, type MustId, type MustVerification, type MustVerifiedBy, OFFER_SCOPES, OfferScope, type OpenResult, PAYLOAD_LIMITS, PROTOCOL_PREFIX, PROTOCOL_VERSION, PairPollRequest, PairPollResponse, PairRequest, PairStartRequest, PairStartResponse, type PayloadFor, PublicIdentity, Purpose, REFUSAL_MESSAGES, RESERVED_PURPOSE, RETIREMENT_WINDOW_MS, RefusalReason, ReleaseRequest, ReleaseResponse, RequestSignature, ResultDisposition, ResultProvenance, ResultRequest, ResultResponse, RunMetadata, SIZE_CLASSES, SIZE_CLASS_LIMITS, SUCCESSION_CONTEXT, SUPPORTED_PROTOCOL_VERSIONS, SealedEnvelope, SealedOutcome, type SignatureFailure, SignedGrant, SizeClass, type SpendConsent, StoredKeys, Succession, type SuccessionFailure, type SuccessionWalk, TERMINAL_STATES, type VersionRefusal, WireError, WireErrorCode, WithheldKind, backendDescriptor, backendName, canTransition, canonicalRequest, checkProtocolVersion, classifyCost, cryptoReady, declaredVersion, effectiveOfferScope, envelopeBytes, fingerprint, generateKeys, grantStatement, isBackendId, isCloudTaggedModel, isJobKind, isLocalHost, isTerminal, keyId, kindsOf, matchAudience, mustsVerifiedBy, open, payloadTextLength, provenanceFor, publicIdentityOf, resolveCost, seal, signGrant, signRequest, signSiteRequest, signSuccession, signWith, singlePurposeManifest, sizeClassCeiling, sizeClassOf, successionStatement, verifyGrant, verifyLink, verifyPublicIdentity, verifyRequest, verifySiteRequest, verifyWith, walkSuccession };

package/dist/index.js

@@ -1,3 +1,70 @@
+// src/about.ts
+var ABOUT = `# About BYOLLM
+
+**What BYOLLM is**
+
+BYOLLM lets you use your own AI on websites. You install one small program on
+your computer. Then, websites that support BYOLLM can use the AI you already
+have \u2014 a free model running on your machine, or an AI service you already pay
+for \u2014 instead of the website paying for AI and passing the cost to you.
+
+**Why it matters**
+
+For you:
+
+- Your favorite model, everywhere you go.
+- New models the moment you get them \u2013 not when a site gets around to adding
+  them.
+- Encrypted end-to-end. Your prompts go to your own device; byollm.cloud can't
+  read them.
+- Sites never learn which model you use, and your subscriptions are never
+  shared.
+- Pay less. Sites that don't pay for AI can charge you less \u2013 or nothing.
+
+For sites and developers:
+
+- Zero AI bills. Your users bring their own compute.
+- No floating money \u2013 you don't pay LLM bills up front and hope to collect
+  later, and you never ask people to prepay just to try you.
+- Free trials that cost you nothing to offer.
+- Ship the AI features you kept private for fear of the API bill.
+- One small integration. Your users choose the models.
+
+**Your device**
+
+The \`byollm\` program runs on your computer. It knows which AI services you have
+set up: free open-source models on your machine, metered services you pay per
+use, or your own subscriptions like Claude Pro/Max. When a website you have
+enabled sends work, your device runs it with the service you chose. Your
+prompts are encrypted end-to-end to your own device. byollm.cloud passes them
+along and cannot read them.
+
+**Sites**
+
+A website that wants to use BYOLLM says what it needs \u2014 "writing help," "chat,"
+and so on. When you connect the site, you pick which of your services answers
+each one. The site never learns which model you use. You can turn a site off at
+any time, and it stops getting your work.
+
+**Teams (optional)**
+
+A team lets you share what runs on your devices with people you name \u2014 the free
+open-source models on your machine, or a metered service with a spending limit
+you set. Your subscription accounts (like Claude Pro/Max) are never shared with
+anyone. That is a rule, not a setting.
+
+**byollm.cloud (or your own relay)**
+
+Many sites, many devices, many people. byollm.cloud keeps track of who has
+allowed what and sends each job to the right device. It never sees your
+prompts. If you would rather run this part yourself, the relay is open source \u2014
+you can run your own instead of using byollm.cloud.`;
+var ABOUT_SHORT_LEDE = "BYOLLM \u2013 Bring Your Own LLM \u2013 lets you use your own AI on websites you authorize. A small program installed on your machine lets you use your own models and subscriptions on any BYOLLM-integrated site, including new models the moment you get access \u2013 no site updates required. BYOLLM Cloud connects sites to your devices with end-to-end encryption, so no one, including us, can see your data.";
+var ABOUT_SHORT_TAIL = "Sites can charge you less because you bring your own \u2013 see why that matters \u2192. Teams can optionally share the free or metered services on their devices with people they name. Personal subscriptions are never shared.";
+var ABOUT_SHORT = `${ABOUT_SHORT_LEDE}
+
+${ABOUT_SHORT_TAIL}`;
+
 // src/audience.ts
 import { z as z2 } from "zod";
 
@@ -840,6 +907,11 @@ var DeliveredResult = z6.object({
 }).strict();
 var SizeClass = z6.enum(["small", "medium", "large", "unbounded"]);
 var SIZE_CLASSES = Object.freeze(SizeClass.options);
+var MAX_ENVELOPE_BYTES = 10 * 1024 * 1024;
+function envelopeBytes(envelope) {
+  const serialised = JSON.stringify(envelope);
+  return serialised === void 0 ? 0 : serialised.length;
+}
 var SIZE_CLASS_LIMITS = Object.freeze({
   small: 4e3,
   medium: 64e3,
@@ -2287,6 +2359,10 @@ var FetchResponse = z11.object({
   envelope: SealedEnvelope
 }).strict();
 export {
+  ABOUT,
+  ABOUT_SHORT,
+  ABOUT_SHORT_LEDE,
+  ABOUT_SHORT_TAIL,
   AUDIENCES,
   Audience,
   BACKENDS,
@@ -2333,6 +2409,7 @@ export {
   KindedPayload,
   Lease,
   MAX_CLOCK_SKEW_MS,
+  MAX_ENVELOPE_BYTES,
   MAX_PURPOSES,
   MAX_SUCCESSION_CHAIN,
   MIN_PROTOCOL_VERSION,
@@ -2387,6 +2464,7 @@ export {
   cryptoReady,
   declaredVersion,
   effectiveOfferScope,
+  envelopeBytes,
   fingerprint,
   generateKeys,
   grantStatement,