Q77: a session keeps the model and account it started with

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

src/main/database/migrations/076-run-pins.ts

@@ -0,0 +1,7 @@
+export const runPinsSchema = `
+  alter table task_runs add column run_model text;
+  alter table task_runs add column run_credential_id text;
+
+  insert into schema_migrations(version, name, applied_at)
+  values (76, 'run_pins', datetime('now'));
+`;

src/main/database/schema.ts

@@ -74,6 +74,7 @@ import { spendSchema } from "./migrations/072-spend";
 import { usageWindowTokensSchema } from "./migrations/073-usage-window-tokens";
 import { runFailureKindSchema } from "./migrations/074-run-failure-kind";
 import { providerCooldownsSchema } from "./migrations/075-provider-cooldowns";
+import { runPinsSchema } from "./migrations/076-run-pins";
 
 interface Migration {
   version: number;
@@ -157,6 +158,7 @@ const MIGRATIONS: Migration[] = [
   { version: 73, name: "usage_window_tokens", sql: usageWindowTokensSchema },
   { version: 74, name: "run_failure_kind", sql: runFailureKindSchema },
   { version: 75, name: "provider_cooldowns", sql: providerCooldownsSchema },
+  { version: 76, name: "run_pins", sql: runPinsSchema },
 ];
 
 const SCHEMA_VERSION = MIGRATIONS[MIGRATIONS.length - 1].version;

src/main/services/provider-response-runner.ts

@@ -17,7 +17,7 @@ import { selectToolTurn, supportsToolTurn } from "./provider-tool-turns/router";
 import { resolveTurnCredential } from "./provider-tool-turns/credential-resolver";
 import { currentRunId } from "./run-control";
 import { budgetBlock, isPriced, promptChars, providerPrices } from "./spend";
-import { effectiveModel } from "./task-run-preparation";
+import { assertModelSelectable, effectiveModel } from "./task-run-preparation";
 import {
   HEALTH_STATE_HEALTHY,
   WORKSPACE_SCOPES,
@@ -122,11 +122,15 @@ export class ProviderResponseRunner {
     const agent = taskRunId
       ? this.deps.db.prepare("select a.id, a.department, a.model_id from task_runs r join agents a on a.id = r.agent_id where r.id = ?").get(taskRunId) as { id: string; department: string; model_id: string | null }
       : null;
-    // Azure calls the deployment stored with its credential, so an agent there cannot pick another model.
-    if (provider.provider_key === "azure-openai" && agent?.model_id && agent.model_id !== provider.model) {
-      throw new Error(`Azure OpenAI calls the deployment configured on the provider; this agent asks for "${agent.model_id}". Clear the agent's model or set it to "${provider.model}".`);
+    const pins = taskRunId
+      ? this.deps.db.prepare("select run_model, run_credential_id from task_runs where id = ?").get(taskRunId) as { run_model: string | null; run_credential_id: string | null }
+      : null;
+    if (pins?.run_credential_id && pins.run_credential_id !== provider.credential_id) {
+      throw new Error("The provider's account was reconnected during this session; the next session continues with the new account.");
     }
-    const model = effectiveModel(agent?.model_id ?? null, provider.model);
+    // Every turn of a session calls the model the session started with (checked then), whatever the configuration says now.
+    if (!pins?.run_model) assertModelSelectable(provider.provider_key, agent?.model_id ?? null, provider.model);
+    const model = pins?.run_model ?? effectiveModel(agent?.model_id ?? null, provider.model);
     if (!model) throw new Error("Provider model is required.");
     const owner: ProviderRow & { model: string } = { ...provider, model };
     const turn = this.deps.toolTurn ?? selectToolTurn(provider.provider_key);

src/main/services/task-execution-service.ts

@@ -37,7 +37,7 @@ import { ConnectorGrantsService } from "./composio/connector-grants";
 import { ProviderCooldowns } from "./provider-cooldowns";
 import { HeartbeatScheduler } from "./heartbeat-scheduler";
 import {
-  assertExecutionApproval, createIsolationContext, effectiveModel, executionApprovalProblem, loadGoal, loadProject, loadWorkspace,
+  assertExecutionApproval, assertModelSelectable, createIsolationContext, effectiveModel, executionApprovalProblem, loadGoal, loadProject, loadWorkspace,
   requireAgentProvider, requireAssignedAgent, requireTask, sameWorkspaceAccess, workspaceAccess,
 } from "./task-run-preparation";
 
@@ -175,7 +175,15 @@ export class TaskExecutionService {
     let providerRequestId: string | null = null;
     let skillRuntimeSnapshot: SkillRuntimeSnapshotRow | null = null;
     try {
-      skillRuntimeSnapshot = this.resolveSkillRuntimeSnapshot(orgId, agent.id, runId, skillBudgetTokens(provider.context_window_tokens), `${task.title}\n${task.description}`, { providerKey: provider.provider_key, model: effectiveModel(agent.model_id, provider.model) }, task.writing_mode === WRITING_MODE.CONVERSATIONAL);
+      // The session keeps the model and account it started with; a resumed run keeps its original pins.
+      const pinned = (this.db.prepare("select run_model from task_runs where id = ?").get(runId) as { run_model: string | null }).run_model;
+      if (pinned === null) {
+        assertModelSelectable(provider.provider_key, agent.model_id, provider.model);
+        this.db.prepare("update task_runs set run_model = ?, run_credential_id = ? where id = ?")
+          .run(effectiveModel(agent.model_id, provider.model), provider.credential_id, runId);
+      }
+      const runModel = pinned ?? effectiveModel(agent.model_id, provider.model);
+      skillRuntimeSnapshot = this.resolveSkillRuntimeSnapshot(orgId, agent.id, runId, skillBudgetTokens(provider.context_window_tokens), `${task.title}\n${task.description}`, { providerKey: provider.provider_key, model: runModel }, task.writing_mode === WRITING_MODE.CONVERSATIONAL);
       // Watermark for reply delivery and mid-run message polling.
       const runStartedAt = new Date().toISOString();
       const chatContext = loadChatWakeContext(this.db, orgId, agent.id, isolation.secretScopeJson);
@@ -227,6 +235,7 @@ export class TaskExecutionService {
             handoffLines: renderTaskHandoff(this.db, orgId, task.handoff_json),
             knowledgeLines: relevantKnowledgeLines(this.db, { orgId, agentId: agent.id, projectId: task.project_id, goalId: task.goal_id, taskId: task.id }, `${task.title}\n${task.description}`),
             priorCompletedRuns: this.completedRunCount(task.id, runId),
+            runModel,
             journalLines: runJournalLines(this.db, task.id),
             subtasks: (this.db.prepare("select title, status = 'done' as done from tasks where parent_task_id = ? order by created_at").all(task.id) as Array<{ title: string; done: number }>)
               .map((row) => ({ title: row.title, done: row.done === 1 })),

src/main/services/task-prompts.ts

@@ -1,6 +1,5 @@
 import { WRITING_MODE, type WritingMode } from "../../shared/domain";
 import { CEO_FIRST_HIRES, firstHiresFor, leadsPeople } from "./org-building";
-import { effectiveModel } from "./task-run-preparation";
 import { canonicalMentionHandle } from "../../shared/mention-handles";
 
 const RECORD_LEARNINGS_INSTRUCTION =
@@ -41,6 +40,7 @@ export interface TaskExecutionProviderRow {
   label: string;
   provider_key: string;
   model: string | null;
+  credential_id: string | null;
   context_window_tokens: number;
 }
 
@@ -202,6 +202,8 @@ export function buildTaskPrompt(input: {
   knowledgeLines: string[];
   /** Completed runs this task already spent without reaching done. */
   priorCompletedRuns: number;
+  /** The model this session calls (pinned at its start). */
+  runModel: string | null;
   /** What earlier sessions on this task already did, and where the last one failed. */
   journalLines: string[];
   /** Names of the tools the model can call this run. */
@@ -221,7 +223,7 @@ export function buildTaskPrompt(input: {
     `Requires filesystem: ${input.task.requires_filesystem === 1 ? "yes" : "no"}`,
     `Assignee: ${input.agent.name} (${input.agent.role})`,
     ...(input.managerName ? [`Reports to: ${input.managerName}`] : []),
-    `Provider: ${input.provider.label} ${effectiveModel(input.agent.model_id, input.provider.model) ?? ""}`.trim(),
+    `Provider: ${input.provider.label} ${input.runModel ?? ""}`.trim(),
     `Execution context: ${input.isolation.executionContextId}`,
     `Service identity: ${input.isolation.serviceIdentityId}`,
   ];

src/main/services/task-run-preparation.ts

@@ -67,6 +67,13 @@ export function sameWorkspaceAccess(mountJson: string, current: Array<{ workspac
   return JSON.stringify(recorded) === JSON.stringify(current);
 }
 
+/** Azure calls the deployment stored with its credential, so an agent there cannot pick another model. */
+export function assertModelSelectable(providerKey: string, agentModelId: string | null, providerModel: string | null): void {
+  if (providerKey === "azure-openai" && agentModelId && agentModelId !== providerModel) {
+    throw new Error(`Azure OpenAI calls the deployment configured on the provider; this agent asks for "${agentModelId}". Clear the agent's model or set it to "${providerModel}".`);
+  }
+}
+
 /** The model a run calls: the agent's own choice, or else its provider's default. */
 export function effectiveModel(agentModelId: string | null, providerModel: string | null): string | null {
   return agentModelId?.trim() || providerModel;
@@ -75,7 +82,7 @@ export function effectiveModel(agentModelId: string | null, providerModel: strin
 export function requireAgentProvider(db: DatabaseSync, orgId: string, agent: TaskExecutionAgentRow): TaskExecutionProviderRow {
   if (!agent.model_provider_id) throw new Error("Task assignee requires a model provider.");
   const provider = db.prepare(`
-    select id, label, provider_key, model, context_window_tokens
+    select id, label, provider_key, model, credential_id, context_window_tokens
     from providers
     where id = ? and org_id = ?
   `).get(agent.model_provider_id, orgId) as TaskExecutionProviderRow | undefined;

tests/integration/agent-model.test.ts

@@ -67,4 +67,51 @@ describe("agent model", () => {
     await expect(execution.runTask({ orgId, taskId })).rejects.toThrow('Azure OpenAI calls the deployment configured on the provider; this agent asks for "gpt-5.4-mini".');
     connection.close();
   });
+
+  test("a session keeps the model and account it started with", async () => {
+    const models: string[] = [];
+    let onTurn = (_turn: number) => {};
+    let lastPrompt = "";
+    const turn = async (input: ProviderToolTurnInput): Promise<AgentTurnResult> => {
+      models.push(input.model);
+      const first = input.messages[0];
+      if (first?.role === "user") lastPrompt = first.text;
+      onTurn(models.length);
+      return models.length < 3
+        ? { text: "Working.", toolCalls: [{ id: `c${models.length}`, name: "read_messages", arguments: {} }], rawResponseJson: "{}" }
+        : { text: "done", toolCalls: [], rawResponseJson: "{}" };
+    };
+    const connection = new DatabaseConnection(join(tmpdir(), `ac-model-pin-${randomUUID()}.db`));
+    const organization = new OrganizationService(connection.db);
+    const providers = new ProviderService(connection.db, providerAdapters, testVault, undefined, turn);
+    const execution = new TaskExecutionService(connection.db, providers);
+    let orgId = "";
+    let taskId = "";
+    let providerId = "";
+    connection.transaction(() => {
+      orgId = organization.createOrganization({ organizationName: "Acme", employeeName: "Dana", mission: "m", goal: "g" });
+      providerId = (connection.db.prepare("select id from providers where org_id = ? and provider_key = 'openai'").get(orgId) as { id: string }).id;
+      providers.saveCredential({ orgId, providerId, secret: "sk-test", model: "gpt-5.4" });
+      const agentId = organization.createAgent({ orgId, name: "Ada", role: "Engineer", department: "Engineering", modelProviderId: providerId, modelId: null, budgetCents: 1000, managerAgentId: null });
+      taskId = organization.createTask({ orgId, title: "Work", description: "d", status: TASK_STATUS.IN_PROGRESS, assigneeAgentId: agentId, risk: RISK_LEVEL.LOW, projectId: null, goalId: null, workspaceId: null, priority: TASK_PRIORITY.NORMAL, requiresFilesystem: false });
+    });
+    onTurn = (count) => { if (count === 1) connection.db.prepare("update providers set model = 'gpt-6' where id = ?").run(providerId); };
+    await execution.runTask({ orgId, taskId });
+    expect(models).toEqual(["gpt-5.4", "gpt-5.4", "gpt-5.4"]);
+
+    // A resumed run builds its prompt and calls with the model it was pinned to, not today's configuration.
+    models.length = 0;
+    onTurn = () => {};
+    const resumedId = (connection.db.prepare("select id from task_runs where task_id = ?").get(taskId) as { id: string }).id;
+    connection.db.prepare("update task_runs set state = 'running', completed_at = null where id = ?").run(resumedId);
+    await execution.executeRun(orgId, resumedId);
+    expect(models).toEqual(["gpt-5.4", "gpt-5.4", "gpt-5.4"]);
+    expect(lastPrompt).toContain("Provider: OpenAI gpt-5.4");
+
+    models.length = 0;
+    onTurn = (count) => { if (count === 1) connection.db.prepare("update providers set credential_id = 'cred_reconnected' where id = ?").run(providerId); };
+    await expect(execution.runTask({ orgId, taskId })).rejects.toThrow("The provider's account was reconnected during this session");
+    expect(models).toEqual(["gpt-6"]);
+    connection.close();
+  });
 });

tests/unit/domain-services.test.ts

@@ -24,7 +24,7 @@ describe("domain services", () => {
     const connection = createConnection();
 
     const rows = connection.db.prepare("select version from schema_migrations order by version asc").all() as Array<{ version: number }>;
-    expect(rows.map((row) => row.version)).toEqual([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75]);
+    expect(rows.map((row) => row.version)).toEqual([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76]);
     connection.close();
   });