@aglyn/plugins-ai 1.0.0-beta.229 → 1.0.0-beta.231

package/src/lib/server/assist-chat.js

@@ -31,6 +31,8 @@ import { docsGroundingBlock, retrieveDocsSections } from "@aglyn/aglyn/app-utils
 import { composeDocsLinksAnswer, deflectToDocs, questionStandsAlone, sectionLabel, sectionUrl } from "@aglyn/aglyn/app-utils/docs-deflection";
 import { ASSIST_EDIT_TOOL_NAME, assistEditDocumentOf } from "../model/assist-edit.js";
 import { ASSIST_EDIT_MAX_OUTPUT_TOKENS, assistEditTool, editCanvasBlock, editSelectionBlock, parseAssistEditContext, resolveAssistEdit } from "./assist-edit.js";
+import { ASSIST_BUILD_TOOL_NAME } from "../model/assist-build.js";
+import { assistBuildBlock, assistBuildIntents, assistBuildIntentsBlock, assistBuildTool, resolveAssistBuild } from "./assist-build.js";
 import { describeView, extractAssistAction, finalAssistText, resolveAssistProposal, safeOrgFacts, sanitiseId, sanitiseRoute, viewFactsBlock, viewScreenBlock, visibleAssistText } from "./assist-view-context.js";
 import { invalidIdTokenResponse } from "@aglyn/tenant-data-admin/server/id-token-refusal";
 // By its own entry point rather than the barrel (AGL-2903), for the reason
@@ -519,6 +521,56 @@ ${AI_ACCEPTABLE_USE_BLOCK}`;
         context
     };
 }
+/**
+ * Whether this turn may PROPOSE a build (AGL-3616) — one request turned into
+ * one `build` job whose plan card the person confirms — and on which site,
+ * or null and the turn is answered as it would be without one.
+ *
+ * The gates are the generation doors' own, as the edit rung's: the org's
+ * `aiGenerative`, `release_ai_generative` (staff preview), the caller's
+ * `ai.generate` on the site the page named, and the `ai-generate` switch.
+ * No canvas is needed: a build is asked from anywhere on a site. Last, a
+ * deployment that loaded the build kind and a site where some operation
+ * passes its owner's gates, which is also what the model is told it may
+ * offer. The job door checks every one of these again when the panel starts
+ * the job; this rung only decides whether the model may propose it.
+ */ async function assistBuildRung(input) {
+    var _ref;
+    var _body_context;
+    const { body, org, staff } = input;
+    const hostId = sanitiseId((_ref = (_body_context = body.context) == null ? void 0 : _body_context.hostId) != null ? _ref : '');
+    if (!hostId) return null;
+    if (!checkEntitlement(org, 'aiGenerative')) return null;
+    if (!staff && !await isServerReleaseFlagOnForOrg('release_ai_generative', body.orgId)) {
+        return null;
+    }
+    if (!staff && !await memberHasPermissionOnHost(body.orgId, hostId, input.member, 'ai.generate')) {
+        return null;
+    }
+    if (await featureLockdownRefusal({
+        feature: 'ai-generate',
+        staff,
+        orgId: body.orgId
+    })) {
+        return null;
+    }
+    const intents = await assistBuildIntents(input.firestore, {
+        orgId: body.orgId,
+        hostId,
+        org
+    }).catch((error)=>{
+        // A rung that cannot be decided stays closed: the chat still answers.
+        console.error('assist build rung failed', {
+            orgId: body.orgId,
+            error
+        });
+        return null;
+    });
+    return intents ? {
+        hostId,
+        intents
+    } : null;
+}
 /**
  * The level-2 view block (entitled orgs only) and the scope the proposal
  * channel resolves against.
@@ -824,6 +876,15 @@ async function handler(request) {
             staff,
             member: resolved.member
         });
+        // The build rung (AGL-3616), decided beside the edit rung for the same
+        // reason: a request to make things is not a question the docs answer.
+        const buildRung = await assistBuildRung({
+            body,
+            org: org,
+            staff,
+            member: resolved.member,
+            firestore
+        });
         // ── Retrieval FIRST (AGL-2486) ────────────────────────────────────────
         // Level-1 grounding for everyone. Hoisted above the reservation and the
         // model call because it is now a candidate ANSWER, not only a prompt
@@ -843,7 +904,7 @@ async function handler(request) {
         // from the docs there. `questionStandsAlone` is the deflector's own test
         // for a question that needs nothing around it, and the open canvas is
         // exactly what an edit request leans on.
-        const deflection = editRung && !questionStandsAlone(body.question) ? null : deflectToDocs(body.question, scored, body.history.length > 0);
+        const deflection = (editRung || buildRung) && !questionStandsAlone(body.question) ? null : deflectToDocs(body.question, scored, body.history.length > 0);
         if (deflection == null ? void 0 : deflection.answered) {
             var _ref4;
             var _body_context4, _body_context5;
@@ -946,7 +1007,7 @@ async function handler(request) {
         // was written by whichever model Auto chose. Nor does a turn on the edit
         // rung (AGL-2906): that answer is composed against the canvas the request
         // described, which the key cannot describe either.
-        const cacheKey = body.history.length || body.model || editRung ? '' : assistAnswerCacheKey({
+        const cacheKey = body.history.length || body.model || editRung || buildRung ? '' : assistAnswerCacheKey({
             question: body.question,
             tier: entitled ? 'entitled' : 'free',
             route: (_ref = (_body_context2 = body.context) == null ? void 0 : _body_context2.route) != null ? _ref : '',
@@ -1133,6 +1194,13 @@ async function handler(request) {
                     cacheBreakpoint: true
                 }
             ] : [],
+            // The build protocol (AGL-3616): the same for every site, so cached.
+            ...buildRung ? [
+                {
+                    text: assistBuildBlock(),
+                    cacheBreakpoint: true
+                }
+            ] : [],
             // Per-org, and therefore AFTER every breakpoint — see
             // `assistBrandBlock`. Unconditional: a free workspace assembles no
             // view block, and it must still be told what the product is called.
@@ -1153,6 +1221,13 @@ async function handler(request) {
                     volatile: true
                 }
             ] : [],
+            // What a build can make on this site: per site, so volatile.
+            ...buildRung ? [
+                {
+                    text: assistBuildIntentsBlock(buildRung.intents),
+                    volatile: true
+                }
+            ] : [],
             ...docsBlock ? [
                 {
                     text: docsBlock,
@@ -1166,11 +1241,17 @@ async function handler(request) {
                 model,
                 // An edit proposal writes its operations as a tool call, which needs
                 // room an answer does not.
-                maxTokens: editRung ? ASSIST_EDIT_MAX_OUTPUT_TOKENS : MAX_OUTPUT_TOKENS,
+                // A build's brief is a tool call too, restating the whole request.
+                maxTokens: editRung || buildRung ? ASSIST_EDIT_MAX_OUTPUT_TOKENS : MAX_OUTPUT_TOKENS,
                 stream: true
-            }, editRung ? {
+            }, editRung || buildRung ? {
                 tools: [
-                    assistEditTool(editRung.target.kind)
+                    ...editRung ? [
+                        assistEditTool(editRung.target.kind)
+                    ] : [],
+                    ...buildRung ? [
+                        assistBuildTool()
+                    ] : []
                 ]
             } : {}, CHAT_ROUTE.thinking ? {
                 thinking: CHAT_ROUTE.thinking
@@ -1213,6 +1294,7 @@ async function handler(request) {
                 let emitted = 0;
                 let stopReason = null;
                 /** The first call of the edit tool, as the stream delivered it. */ let editInput = null;
+                /** The first call of the build tool (AGL-3616). */ let buildInput = null;
                 /**
          * What the model wrote into its tool calls, when the stream stopped
          * at its ceiling (AGL-3143). Held here only long enough to be
@@ -1226,7 +1308,8 @@ async function handler(request) {
                     cacheWriteTokens: 0
                 };
                 try {
-                    var _ref;
+                    var _ref, _ref1, _ref2, _ref3;
+                    var _guide_view, _describeView, _body_context;
                     for await (const event of upstream){
                         if (event.type === 'delta') {
                             raw += event.text;
@@ -1251,6 +1334,9 @@ async function handler(request) {
                             if (editRung && !editInput && event.name === ASSIST_EDIT_TOOL_NAME) {
                                 editInput = event.input;
                             }
+                            if (buildRung && !buildInput && event.name === ASSIST_BUILD_TOOL_NAME) {
+                                buildInput = event.input;
+                            }
                         } else if (event.type === 'done') {
                             var _event_rawOutput;
                             usage = event.usage;
@@ -1295,6 +1381,13 @@ async function handler(request) {
                             });
                         }
                     }
+                    // The build proposal (AGL-3616): the request, held to the site the
+                    // turn was asked on. Still nothing is made: the panel starts a
+                    // build job whose plan the person confirms.
+                    const build = buildRung && stopReason !== 'max_tokens' ? resolveAssistBuild(buildInput, {
+                        hostId: buildRung.hostId,
+                        screen: (_ref = (_ref1 = guide == null ? void 0 : (_guide_view = guide.view) == null ? void 0 : _guide_view.screen) != null ? _ref1 : (_describeView = describeView(sanitiseRoute((_ref2 = (_body_context = body.context) == null ? void 0 : _body_context.route) != null ? _ref2 : ''))) == null ? void 0 : _describeView.screen) != null ? _ref : null
+                    }) : null;
                     // A refusal is an HTTP 200 with an empty or partial answer, not an
                     // error — so without this the user watches the spinner stop and
                     // gets nothing back. Truncation is the same shape: the tokens are
@@ -1333,14 +1426,14 @@ async function handler(request) {
                     // tokens were spent either way and the miner wants the question.
                     let exchangeId = null;
                     try {
-                        var _ref1;
-                        var _body_context, _body_context1;
+                        var _ref4;
+                        var _body_context1, _body_context2;
                         exchangeId = await recordAssistExchange(firestore, body.orgId, _extends({
                             uid: decoded.uid,
                             question: body.question,
                             answer: answer.slice(0, MAX_STORED_ANSWER_CHARS),
-                            route: (_ref1 = (_body_context = body.context) == null ? void 0 : _body_context.route) != null ? _ref1 : '',
-                            hostId: ((_body_context1 = body.context) == null ? void 0 : _body_context1.hostId) || null,
+                            route: (_ref4 = (_body_context1 = body.context) == null ? void 0 : _body_context1.route) != null ? _ref4 : '',
+                            hostId: ((_body_context2 = body.context) == null ? void 0 : _body_context2.hostId) || null,
                             model,
                             tier,
                             usage,
@@ -1372,7 +1465,7 @@ async function handler(request) {
                     //  - an answer from a model other than the one the key names — an
                     //    allowlist moved Auto (AGL-2942) — would be served to readers
                     //    the list does not bind.
-                    if (cacheKey && answer && !proposal && stopReason !== 'refusal' && stopReason !== 'max_tokens' && model === assistModel()) {
+                    if (cacheKey && answer && !proposal && !build && stopReason !== 'refusal' && stopReason !== 'max_tokens' && model === assistModel()) {
                         await writeAssistAnswerCache(firestore, body.orgId, cacheKey, {
                             answer,
                             docs
@@ -1390,6 +1483,9 @@ async function handler(request) {
                         // The edit proposal (AGL-2906): ops the panel applies in the
                         // author's editor when they press Apply, and never before.
                         edit,
+                        // The build proposal (AGL-3616): the panel starts its job and
+                        // shows the plan card; nothing is built before it is confirmed.
+                        build,
                         // The reservation VERBATIM (AGL-2238). It already describes the
                         // standing after the message, because it is what moved the
                         // counter — `reserveAssistMessage` returns `used + 1` and
@@ -1406,7 +1502,7 @@ async function handler(request) {
                             lastCredits: assistCreditsFromUsd(estimateAssistCostUsd(usage, model)),
                             model: {
                                 id: model,
-                                auto: (_ref = choice == null ? void 0 : choice.auto) != null ? _ref : true
+                                auto: (_ref3 = choice == null ? void 0 : choice.auto) != null ? _ref3 : true
                             }
                         })
                     });