fix(assistant): store the transcript before message.end so the panel shows the answer

The panel reloads the conversation the moment the stream ends. The
redacted transcript (message parts, pending actions, stored events) was
flushed only after message.end was sent, so that reload read the "[]"
placeholders and showed the newest question and answer as empty bubbles
until the panel was closed and reopened. Persist and flush first, then
emit message.end.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

.circleci/config.yml

@@ -1,12 +1,12 @@
 version: 2.1
 setup: true
 orbs:
-  qqq-orb: kingsrook/qqq-orb@0.7.12
+  qqq-orb: kingsrook/qqq-orb@0.8
 workflows:
   setup:
     jobs:
     - qqq-orb/generate_pipeline:
-        orb_version: 0.7.12
+        orb_version: '0.8'
         filters:
           branches:
             ignore: automation/qbit-rc-pending

.gitea/workflows/qrun-release.yaml

@@ -0,0 +1,25 @@
+# Standard QRun release caller (v2, thin contract): every push and manual dispatch delegates to the protected qrun-release-v1 channel. Do not edit per repository.
+name: qrun-release
+on:
+  push: {}
+  workflow_dispatch: {}
+permissions:
+  contents: read
+  actions: read
+jobs:
+  release:
+    uses: self:KofTwentyTwo/platform-ci/.gitea/workflows/java-docker.yaml@qrun-release-v1
+    with:
+      runner: linux-small
+      docker-runner: container-build
+      browsers: true
+      browser-runner: browser-build
+  postpublish:
+    if: ${{ needs.release.outputs.capture == 'true' }}
+    needs: release
+    permissions:
+      contents: write
+      actions: write
+    uses: self:KofTwentyTwo/platform-ci/.gitea/workflows/qrun-source-version.yaml@qrun-release-v1
+    with:
+      runner: linux-small

.gitea/workflows/release-rc.yaml

@@ -0,0 +1,72 @@
+name: Release candidate
+'on':
+  workflow_dispatch:
+    inputs:
+      dry_run:
+        description: Plan a cut without changing release state
+        type: boolean
+        default: false
+      finalize:
+        description: Retry finalization of an existing cut; do not cut another RC
+        type: boolean
+        default: false
+      recover:
+        description: Recover the same published RC after a deployment timeout; never
+          reupload
+        type: boolean
+        default: false
+      circle_build:
+        description: Original release run id (recovery only)
+        type: string
+        default: ''
+      merge_sha:
+        description: Existing cut merge_sha (finalize retry only)
+        type: string
+        default: ''
+      bump_sha:
+        description: Existing cut bump_sha (finalize retry only)
+        type: string
+        default: ''
+      branch:
+        description: Existing cut branch (finalize retry only)
+        type: string
+        default: ''
+      version:
+        description: Existing cut version (finalize retry only)
+        type: string
+        default: ''
+  pull_request:
+    types:
+    - closed
+    branches:
+    - develop
+permissions:
+  contents: read
+jobs:
+  release-rc:
+    # The old qbit-rc-published repository_dispatch is replaced by the already-proven workflow_dispatch
+    # completion: the publisher's own ActionsWRITE job dispatches this workflow with finalize=true plus
+    # merge_sha/bump_sha/branch/version (same inputs as the manual finalize retry).
+    permissions:
+      contents: write
+      actions: read
+      pull-requests: write
+    if: github.event_name == 'workflow_dispatch'
+      || (github.event_name == 'pull_request' && github.event.pull_request.merged
+      == true && github.event.pull_request.base.ref == 'develop' && contains(github.event.pull_request.labels.*.name,
+      'release:rc'))
+    uses: self:KofTwentyTwo/platform-ci/.gitea/workflows/qbit-release-rc.yaml@qrun-release-v1
+    with:
+      dry_run: ${{ github.event_name == 'workflow_dispatch' && inputs.dry_run || false
+        }}
+      finalize: ${{ github.event_name == 'workflow_dispatch' && inputs.finalize }}
+      recover: ${{ github.event_name == 'workflow_dispatch' && inputs.recover || false
+        }}
+      circle_build: ${{ github.event_name == 'workflow_dispatch' && inputs.circle_build
+        || '' }}
+      merge_sha: ${{ inputs.merge_sha || '' }}
+      bump_sha: ${{ inputs.bump_sha || '' }}
+      branch: ${{ inputs.branch || '' }}
+      version: ${{ inputs.version || '' }}
+    secrets:
+      QRUN_TAG_SIGNING_KEY: ${{ secrets.QRUN_TAG_SIGNING_KEY }}

.qqq-ci.yml

@@ -1,5 +1,5 @@
 profile: qbit
-orb_version: 0.7.12
+orb_version: '0.8'
 default_branch: develop
 java_version: '21'
 browsers: true

CHANGELOG.md

@@ -1,5 +1,16 @@
 # Changelog
 
+## Unreleased - Answer visible at the end of the stream
+
+- The turn's redacted transcript (the question and answer parts, pending actions, stored events) is now written before `message.end` is sent. The panel reloads the conversation the moment the stream ends, and used to read the empty placeholders (`[]`) the transcript redactor stores until its flush, so the newest question and answer showed as empty bubbles until the panel was closed and reopened.
+
+## Unreleased - Clear reason without a model provider (UI API extension 1.7.0)
+
+- `GET /config` checks the server-wide causes before the user's permission: `DISABLED`, then `NOT_CONFIGURED` (no usable model provider), then `NO_PERMISSION`, then `KILL_SWITCH`. A server with no provider no longer tells users to ask for access (#82).
+- UI API extension assistant 1.7.0: `GET /config` has `message` whenever `reason` is set: a sentence for the user. For `NOT_CONFIGURED` it says what an administrator configures (an enabled row on **Assistant > Providers**, for example Anthropic with `ANTHROPIC_API_KEY` set in the server environment, or a local OpenAI-compatible server such as Ollama).
+- Without a usable provider, calls from a user who also lacks `assistant.hasAccess` answer `503 NOT_CONFIGURED` with that message instead of `403 NO_PERMISSION`; users with access keep their conversations, and a new question answers `503 NOT_CONFIGURED`.
+- Verified the OpenAI-compatible adapter (kind `OLLAMA`) live against a local Ollama `llama3.3:70b-instruct-q4_K_M`, including a tool call (`LiveOpenAiCompatibleTest`).
+
 ## Unreleased - Conversation management (UI API extension 1.6.0)
 
 - `assistant_conversation` is open to every assistant user (`assistant.hasAccess` or `assistant.admin`, through the new `AssistantAccessPermissionChecker`) in the standard table UI. Rows stay locked to their owner by the `userId` record lock, so a user only ever sees, updates and deletes their own conversations; admins with the all-access key still see everyone's. The Assistant app (its nav) and every other assistant table stay admin-only.
@@ -27,6 +38,7 @@
 - Provider SPI: `ModelPart.Type.IMAGE` (`mediaType`, base64 `data`), mapped by the Anthropic, Bedrock and OpenAI-compatible providers. Claude models report vision. `ModelPart.toString()` never includes text or image data.
 - `GET /config`: `features.attachments`, `limits.maxAttachmentBytes`, `limits.maxAttachments`, `limits.attachmentTypes`.
 - System prompt: when `create_record` / `update_record` are offered, the model proposes the change instead of saying it cannot create records.
+||||||| parent of c5125f9 (fix(assistant): store the transcript before message.end so the panel shows the answer)
 
 ## Unreleased - Optional provider health
 

README.md

@@ -479,6 +479,17 @@ OLLAMA, LM_STUDIO, VLLM, LLAMA_CPP or OPENAI_COMPATIBLE as their kind.
   Registry prices override type/model prices, then configured prices. Unpriced
   local models cost zero. Capabilities can be adjusted per registered model.
 
+For a local Ollama with no key (the live test ran against `llama3.3:70b-instruct-q4_K_M`):
+
+```java
+config.withProviderType(new OpenAiCompatibleProviderType());
+// provider row: type openai-compatible, kind OLLAMA, model llama3.3:70b-instruct-q4_K_M,
+// base URL empty (http://localhost:11434/v1) or an allow-listed override such as http://127.0.0.1:11434/v1
+```
+
+Without any usable provider `GET /config` answers `enabled: false, reason: NOT_CONFIGURED` with a
+`message` that says what to configure, before any permission check.
+
 For Ollama behind a TLS proxy with a private CA and bearer authentication:
 
 ```java

docs/DESIGN.md

@@ -44,7 +44,7 @@ Guards:
 `ProviderService.resolve(policyModel)`:
 1. The enabled row marked default (marking one default unmarks the others), else the first enabled row.
 2. With no enabled row, the provider registered in code (`withModelProvider`) and the policy's main model: hosts that never add a row keep working.
-3. A row that cannot be used (no key, a disallowed name or URL, an uninstalled type) makes the assistant report `NOT_CONFIGURED` rather than fall back silently.
+3. A row that cannot be used (no key, a disallowed name or URL, an uninstalled type) makes the assistant report `NOT_CONFIGURED` rather than fall back silently. `GET /config` reports it before `NO_PERMISSION`, with a `message` naming what to configure (#82).
 
 The model is the row's, else the policy's main model. The rows and prices are cached for `policyCacheSeconds` and dropped on any change.
 

qbit-assistant/src/main/java/com/kingsrook/qbits/assistant/api/AssistantApi.java

@@ -86,6 +86,20 @@ public class AssistantApi
 
    public static final int TITLE_MAX_CHARS = 60;
 
+   /////////////////////////////////////////////////////////////////////////////
+   // GET /config reason (and error code) when no model provider is usable:   //
+   // no enabled provider row that can be built (a missing key, a disallowed  //
+   // variable or URL, an uninstalled type) and no provider registered in code //
+   /////////////////////////////////////////////////////////////////////////////
+   public static final String NOT_CONFIGURED = "NOT_CONFIGURED";
+
+   public static final String NO_MODEL_PROVIDER_MESSAGE = "No model provider is configured on this server. An administrator adds an enabled provider on "
+      + "Assistant > Providers (for example type Anthropic, whose API key variable ANTHROPIC_API_KEY must be set in the server environment, or a local "
+      + "OpenAI-compatible server such as Ollama).";
+   public static final String NO_PERMISSION_MESSAGE     = "You do not have access to the assistant.";
+   public static final String DISABLED_MESSAGE          = "The assistant is turned off for this application.";
+   public static final String KILL_SWITCH_MESSAGE       = "The assistant has been turned off by an administrator.";
+
    private final AssistantRuntime    runtime;
    private final AssistantQBitConfig config;
    private final AssistantStore      store;
@@ -134,27 +148,39 @@ public class AssistantApi
       Map<String, Object> result = new LinkedHashMap<>();
       String              reason = null;
 
+      String              message = null;
+
+      /////////////////////////////////////////////////////////////////////////
+      // server-wide causes come before the user's own permission, so a      //
+      // server without a usable model provider says so (QRun-IO/qbit-       //
+      // assistant#82) instead of sending the user to ask for access         //
+      /////////////////////////////////////////////////////////////////////////
       if(!config.isEffectivelyEnabled())
       {
          reason = "DISABLED";
+         message = DISABLED_MESSAGE;
+      }
+      else if(activeProvider() == null)
+      {
+         reason = NOT_CONFIGURED;
+         message = NO_MODEL_PROVIDER_MESSAGE;
       }
       else if(!hasAccess())
       {
          reason = "NO_PERMISSION";
-      }
-      else if(activeProvider() == null)
-      {
-         reason = "NOT_CONFIGURED";
+         message = NO_PERMISSION_MESSAGE;
       }
       else if(runtime.getPolicyService().current(false).killSwitch())
       {
          reason = "KILL_SWITCH";
+         message = KILL_SWITCH_MESSAGE;
       }
 
       result.put("enabled", reason == null);
       if(reason != null)
       {
          result.put("reason", reason);
+         result.put("message", message);
       }
       result.put("shortcut", config.getShortcut());
       result.put("model", reason == null ? activeProvider().model() : config.getModels().getMainModel());
@@ -425,11 +451,11 @@ public class AssistantApi
       ProviderService.Active        active = runtime.getProviderService().resolve(policy.mainModel());
       if(active == null)
       {
-         throw (new AssistantApiException(503, "NOT_CONFIGURED", "The assistant is not configured on this server."));
+         throw (new AssistantApiException(503, NOT_CONFIGURED, NO_MODEL_PROVIDER_MESSAGE));
       }
       if(policy.killSwitch())
       {
-         throw (new AssistantApiException(503, "KILL_SWITCH", "The assistant has been turned off by an administrator."));
+         throw (new AssistantApiException(503, "KILL_SWITCH", KILL_SWITCH_MESSAGE));
       }
 
       Long capMicros = ProviderService.capMicros(active.row());
@@ -1165,11 +1191,33 @@ public class AssistantApi
    {
       if(!config.isEffectivelyEnabled())
       {
-         throw (new AssistantApiException(503, "DISABLED", "The assistant is turned off for this application."));
+         throw (new AssistantApiException(503, "DISABLED", DISABLED_MESSAGE));
       }
       if(!hasAccess())
       {
-         throw (new AssistantApiException(403, "NO_PERMISSION", "You do not have access to the assistant."));
+         if(activeProviderOrNull() == null)
+         {
+            throw (new AssistantApiException(503, NOT_CONFIGURED, NO_MODEL_PROVIDER_MESSAGE));
+         }
+         throw (new AssistantApiException(403, "NO_PERMISSION", NO_PERMISSION_MESSAGE));
+      }
+   }
+
+
+
+   /*******************************************************************************
+    ** The active provider, or null when none is usable or it cannot be read
+    ** (only used to pick the clearer refusal).
+    *******************************************************************************/
+   private ProviderService.Active activeProviderOrNull()
+   {
+      try
+      {
+         return (activeProvider());
+      }
+      catch(QException | RuntimeException e)
+      {
+         return (null);
       }
    }
 

qbit-assistant/src/main/java/com/kingsrook/qbits/assistant/api/specs/AssistantSpecs.java

@@ -63,7 +63,7 @@ import io.javalin.http.HttpStatus;
 public final class AssistantSpecs
 {
    public static final String EXTENSION_NAME    = "assistant";
-   public static final String EXTENSION_VERSION = "1.6.0";
+   public static final String EXTENSION_VERSION = "1.7.0";
    public static final String PREFIX            = "/assistant";
 
 
@@ -198,7 +198,8 @@ public final class AssistantSpecs
          "starterPrompts", array(string(null), "Up to three suggested first questions in total: permitted configured questions first, then generated questions. Table, row and field access are checked on every request.")));
       rs.put("AssistantConfigResponseV1", object("Whether the assistant is available to this user, and how.",
          "enabled", bool("Whether the panel may be used."),
-         "reason", string("Why not: DISABLED, NO_PERMISSION, NOT_CONFIGURED or KILL_SWITCH.").withNullable(true),
+         "reason", string("Why not: DISABLED, NOT_CONFIGURED (no usable model provider on the server), NO_PERMISSION or KILL_SWITCH, checked in that order.").withNullable(true),
+         "message", string("Since 1.7.0: when reason is set, a sentence for the user that names the cause (for NOT_CONFIGURED, what an administrator configures).").withNullable(true),
          "shortcut", string("Keyboard shortcut, e.g. mod+j."),
          "model", string("The model that answers."),
          "limits", freeObject("maxTurnSeconds and maxToolCallsPerTurn. When features.attachments is true also maxAttachmentBytes (the largest file read, in bytes), "

qbit-assistant/src/main/java/com/kingsrook/qbits/assistant/engine/TurnEngine.java

@@ -1343,6 +1343,17 @@ public class TurnEngine implements Runnable
             String finalErrorText = errorText;
             safely(() -> emit(TurnEvent.ERROR, map("code", finalErrorCode, "message", finalErrorText)));
          }
+
+         ///////////////////////////////////////////////////////////////////////
+         // the stream ends at message.end and the panel reloads the          //
+         // conversation at once, so the redacted transcript (message parts,  //
+         // pending actions, events) must be stored before that event is sent //
+         ///////////////////////////////////////////////////////////////////////
+         if(!alreadyTerminal)
+         {
+            safely(this::persistPendingEvents);
+            safely(() -> transcript.flush(store));
+         }
          safely(() -> emit(TurnEvent.MESSAGE_END, map("messageId", assistantMessage.getId(), "status", finalStatus, "stopReason", finalStopReason)));
          if(!alreadyTerminal)
          {

qbit-assistant/src/main/resources/openapi/qqq-v1/assistant.yaml

@@ -175,7 +175,13 @@ components:
           description: "Whether the panel may be used."
           type: "boolean"
         reason:
-          description: "Why not: DISABLED, NO_PERMISSION, NOT_CONFIGURED or KILL_SWITCH."
+          description: "Why not: DISABLED, NOT_CONFIGURED (no usable model provider\
+            \ on the server), NO_PERMISSION or KILL_SWITCH, checked in that order."
+          nullable: true
+          type: "string"
+        message:
+          description: "Since 1.7.0: when reason is set, a sentence for the user that\
+            \ names the cause (for NOT_CONFIGURED, what an administrator configures)."
           nullable: true
           type: "string"
         shortcut:
@@ -305,7 +311,9 @@ components:
           nullable: true
           type: "string"
         starterPrompts:
-          description: "Up to three suggested first questions in total: permitted configured questions first, then generated questions. Table, row and field access are checked on every request."
+          description: "Up to three suggested first questions in total: permitted\
+            \ configured questions first, then generated questions. Table, row and\
+            \ field access are checked on every request."
           items:
             type: "string"
           type: "array"
@@ -468,7 +476,7 @@ info:
   description: "QQQ Assistant: conversations, streamed turns, feedback, pending actions\
     \ and data access explanations"
   title: "QQQ Middleware API extension: assistant"
-  version: "1.6.0"
+  version: "1.7.0"
 openapi: "3.0.3"
 paths:
   /qqq/v1/assistant/config:

qbit-assistant/src/test/java/com/kingsrook/qbits/assistant/api/AssistantRouteProviderTest.java

@@ -166,6 +166,45 @@ class AssistantRouteProviderTest extends BaseAssistantTest
 
 
 
+   /*******************************************************************************
+    ** No usable model provider (QRun-IO/qbit-assistant#82): the probe names the
+    ** server's configuration, for users with and without access, and calls
+    ** answer 503 NOT_CONFIGURED instead of 403 NO_PERMISSION.
+    *******************************************************************************/
+   @Test
+   void testNoModelProviderIsNotAPermissionProblem() throws Exception
+   {
+      String providerName = config.getProviderName();
+      config.setProviderName("no-such-provider");
+      try
+      {
+         for(String user : new String[] {EDITOR, NO_ACCESS})
+         {
+            Map<String, Object> probe = json(send("GET", "/config", user, null));
+            assertThat(probe).containsEntry("enabled", false).containsEntry("reason", "NOT_CONFIGURED");
+            assertThat((String) probe.get("message")).contains("No model provider").contains("Assistant > Providers").contains("ANTHROPIC_API_KEY");
+         }
+
+         HttpResponse<String> denied = send("GET", "/conversations", NO_ACCESS, null);
+         assertThat(denied.statusCode()).isEqualTo(503);
+         assertThat(json(denied)).containsEntry("code", "NOT_CONFIGURED");
+         assertThat((String) json(denied).get("message")).contains("No model provider");
+
+         assertThat(send("GET", "/conversations", EDITOR, null).statusCode()).isEqualTo(200);
+         HttpResponse<String> turn = send("POST", "/conversations", EDITOR, "{}");
+         Object id = json(turn).get("id");
+         HttpResponse<String> ask = send("POST", "/conversations/" + id + "/turns", EDITOR, "{\"text\":\"hi\"}");
+         assertThat(ask.statusCode()).isEqualTo(503);
+         assertThat(json(ask)).containsEntry("code", "NOT_CONFIGURED");
+      }
+      finally
+      {
+         config.setProviderName(providerName);
+      }
+   }
+
+
+
    /*******************************************************************************
     **
     *******************************************************************************/
@@ -178,7 +217,8 @@ class AssistantRouteProviderTest extends BaseAssistantTest
       assertThat((Map<String, Object>) config.get("limits")).containsEntry("maxTurnSeconds", 300).containsEntry("maxToolCallsPerTurn", 25);
       assertThat((Map<String, Object>) config.get("features")).containsEntry("writes", false);
 
-      assertThat(json(send("GET", "/config", NO_ACCESS, null))).containsEntry("enabled", false).containsEntry("reason", "NO_PERMISSION");
+      assertThat(json(send("GET", "/config", NO_ACCESS, null))).containsEntry("enabled", false).containsEntry("reason", "NO_PERMISSION")
+         .containsEntry("message", "You do not have access to the assistant.");
       HttpResponse<String> denied = send("GET", "/conversations", NO_ACCESS, null);
       assertThat(denied.statusCode()).isEqualTo(403);
       assertThat(json(denied)).containsEntry("code", "NO_PERMISSION")

qbit-assistant/src/test/java/com/kingsrook/qbits/assistant/engine/TurnEngineTest.java

@@ -141,6 +141,27 @@ class TurnEngineTest extends BaseAssistantTest
 
 
 
+   /*******************************************************************************
+    ** The panel reloads the conversation as soon as message.end arrives, so the
+    ** question and the answer must already be stored then, not after it.
+    *******************************************************************************/
+   @Test
+   void testMessagesAreStoredBeforeMessageEnd() throws Exception
+   {
+      provider.thenText("Four posts are published.");
+
+      TurnRunner.Outcome outcome = runner.run(EDITOR, "how many published posts?");
+      assertThat(outcome.end()).containsEntry("status", "COMPLETE");
+
+      List<AssistantMessage> stored = runner.messagesAtEnd();
+      assertThat(stored).hasSize(2);
+      assertThat(stored).extracting(AssistantMessage::getPartsJson).allSatisfy(parts -> assertThat(parts).isNotEqualTo("[]"));
+      assertThat(stored.get(0).getPartsJson()).contains("how many published posts?");
+      assertThat(stored.get(1).getPartsJson()).contains("Four posts are published.");
+   }
+
+
+
    /*******************************************************************************
     ** Parallel tool calls: all results in ONE user message, in order.
     *******************************************************************************/

qbit-assistant/src/test/java/com/kingsrook/qbits/assistant/testing/TurnRunner.java

@@ -27,6 +27,7 @@ import com.kingsrook.qbits.assistant.AssistantRuntime;
 import com.kingsrook.qbits.assistant.api.AssistantApi;
 import com.kingsrook.qbits.assistant.engine.RunningTurn;
 import com.kingsrook.qbits.assistant.engine.TurnEvent;
+import com.kingsrook.qbits.assistant.model.AssistantMessage;
 import com.kingsrook.qbits.assistant.model.AssistantToolCall;
 import com.kingsrook.qbits.assistant.persistence.AssistantStore;
 import com.kingsrook.qqq.backend.core.context.QContext;
@@ -47,6 +48,8 @@ public class TurnRunner
    private final QInstance           qInstance;
    private final AssistantQBitConfig config;
 
+   private List<AssistantMessage> messagesAtEnd = List.of();
+
 
 
    /*******************************************************************************
@@ -169,6 +172,10 @@ public class TurnRunner
       {
          List<TurnEvent> batch = running.getLog().after(last, 1000);
          events.addAll(batch);
+         if(batch.stream().anyMatch(e -> TurnEvent.MESSAGE_END.equals(e.getType())))
+         {
+            messagesAtEnd = messagesOf(userId, convId);
+         }
          if(!batch.isEmpty())
          {
             last = batch.get(batch.size() - 1).getSeq();
@@ -194,6 +201,36 @@ public class TurnRunner
 
 
 
+   /*******************************************************************************
+    ** The conversation's stored messages, read the moment the run's message.end
+    ** was seen: what the panel finds when it reloads the conversation at the end
+    ** of the stream.
+    *******************************************************************************/
+   public List<AssistantMessage> messagesAtEnd()
+   {
+      return (messagesAtEnd);
+   }
+
+
+
+   /*******************************************************************************
+    ** The stored messages of a conversation (read as the user).
+    *******************************************************************************/
+   private List<AssistantMessage> messagesOf(String userId, Integer conversationId) throws Exception
+   {
+      QContext.init(qInstance, AssistantTestFixture.ownerSession(userId, config));
+      try
+      {
+         return (new AssistantStore().listMessages(userId, conversationId, null, 50));
+      }
+      finally
+      {
+         QContext.clear();
+      }
+   }
+
+
+
    /*******************************************************************************
     ** The audited tool calls of a turn (read as the user).
     *******************************************************************************/