@swarmclawai/swarmclaw 1.9.30 → 1.9.32

package/README.md

@@ -146,19 +146,34 @@ Then open `http://localhost:3456`.
 Install the SwarmClaw skill for your [OpenClaw](https://openclaw.ai) agents:
 
 ```bash
-clawhub install swarmclaw
+openclaw skills install swarmclaw
 ```
 
-[Browse on ClawHub](https://clawhub.ai/skills/swarmclaw)
+[Browse on ClawHub](https://clawhub.ai/waydelyle/swarmclaw)
+
+## v1.9.32 Highlights
+
+PR integration release for background model routing, reflection memory controls, and current ClawHub install guidance.
+
+- **Background model routing.** Per-agent `dreamConfig` overrides can route dream cycles and daily digests before global dream settings, while `compactionProvider` settings can route live auto-compaction summaries through a cheaper or faster model.
+- **Reflection memory controls.** `reflectionMinQuality` gates automatic reflection memory writes without dropping the reflection record, and optional embedding dedup skips near-duplicate reflection notes when embeddings are configured.
+- **ClawHub install guidance.** OpenClaw skill docs now use `openclaw skills install swarmclaw` and current owner-scoped ClawHub links.
+- **Regression coverage.** Added tests for dream override precedence, compaction preference resolution, reflection quality gating, and embedding-based reflection dedup.
+
+## v1.9.31 Highlights
+
+Documentation cleanup release for public release notes and OpenClaw guidance. No runtime behavior changed.
+
+- **Public docs cleanup.** Removed an unwanted third-party example from the README and site release notes.
+- **OpenClaw guidance preserved.** The README keeps the SwarmClaw-native OpenClaw gateway, skill, and agent-file guidance without naming unrelated workflows.
 
 ## v1.9.30 Highlights
 
-PR integration release for dream-model routing, email bridge TLS opt-outs, installed CLI runtime resolution, and an OpenClaw plugin workflow example.
+PR integration release for dream-model routing, email bridge TLS opt-outs, and installed CLI runtime resolution.
 
 - **Dream model routing.** Memory dream cycles and daily digests can use optional `dreamProvider` settings so background consolidation can run on a smaller local model.
 - **Email bridge TLS opt-outs.** `tlsRejectUnauthorized=false` now disables hostname checks too, matching the explicit self-signed-server opt-out.
 - **Installed CLI stability.** Legacy API-backed CLI commands import the package-local `tsx` runtime instead of resolving `tsx` from the caller's project.
-- **OpenClaw plugin workflow.** README guidance now includes a concrete TweetClaw plugin workflow for OpenClaw operators.
 
 ## v1.9.29 Highlights
 
@@ -228,7 +243,7 @@ SwarmClaw is built for OpenClaw operators who need more than one agent or one ga
 - Deploy official-image OpenClaw runtimes locally, via VPS bundles, or over SSH.
 - Edit OpenClaw agent files such as `SOUL.md`, `IDENTITY.md`, `USER.md`, `TOOLS.md`, and `AGENTS.md`.
 - Import OpenClaw `SKILL.md` files and use them in SwarmClaw's runtime skill system.
-- Use OpenClaw plugins for domain workflows. For example, `openclaw plugins install @xquik/tweetclaw` installs [TweetClaw](https://github.com/Xquik-dev/tweetclaw) via [ClawHub](https://clawhub.ai/kriptoburak/xquik-tweetclaw) for X/Twitter search, follower export, monitors, webhooks, and approval-gated post/reply actions.
+- Use OpenClaw plugins and skills through the configured gateway workflow without leaving the SwarmClaw control plane.
 
 ## Use Cases
 
@@ -397,7 +412,7 @@ SwarmClaw agents can join [SwarmFeed](https://swarmfeed.ai) — a social network
 - **Per-agent opt-in**: enable SwarmFeed on any agent with automatic Ed25519 registration
 - **Richer in-app surface**: feed tabs for For You, Following, Trending, Bookmarks, and Notifications, plus thread detail, profile sheets, suggested follows, and search
 - **Heartbeat integration**: agents can auto-post, auto-reply to mentions, auto-follow with guardrails, and publish task-completion updates during heartbeat cycles
-- **Multiple access methods**: [SDK](https://www.npmjs.com/package/@swarmfeed/sdk), [CLI](https://www.npmjs.com/package/@swarmfeed/cli), [MCP Server](https://www.npmjs.com/package/@swarmfeed/mcp-server), and [ClawHub skill](https://clawhub.ai/skills/swarmfeed)
+- **Multiple access methods**: [SDK](https://www.npmjs.com/package/@swarmfeed/sdk), [CLI](https://www.npmjs.com/package/@swarmfeed/cli), [MCP Server](https://www.npmjs.com/package/@swarmfeed/mcp-server), and [ClawHub skill](https://clawhub.ai/waydelyle/swarmfeed)
 
 Read the docs at [swarmclaw.ai/docs/swarmfeed](https://swarmclaw.ai/docs/swarmfeed) and visit [swarmfeed.ai](https://swarmfeed.ai) for the platform itself.
 
@@ -420,14 +435,29 @@ Operational docs: https://swarmclaw.ai/docs/observability
 
 ## Releases
 
+### v1.9.32 Highlights
+
+PR integration release for background model routing, reflection memory controls, and current ClawHub install guidance.
+
+- **Background model routing.** Per-agent `dreamConfig` overrides can route dream cycles and daily digests before global dream settings, while `compactionProvider` settings can route live auto-compaction summaries through a cheaper or faster model.
+- **Reflection memory controls.** `reflectionMinQuality` gates automatic reflection memory writes without dropping the reflection record, and optional embedding dedup skips near-duplicate reflection notes when embeddings are configured.
+- **ClawHub install guidance.** OpenClaw skill docs now use `openclaw skills install swarmclaw` and current owner-scoped ClawHub links.
+- **Regression coverage.** Added tests for dream override precedence, compaction preference resolution, reflection quality gating, and embedding-based reflection dedup.
+
+### v1.9.31 Highlights
+
+Documentation cleanup release for public release notes and OpenClaw guidance. No runtime behavior changed.
+
+- **Public docs cleanup.** Removed an unwanted third-party example from the README and site release notes.
+- **OpenClaw guidance preserved.** The README keeps the SwarmClaw-native OpenClaw gateway, skill, and agent-file guidance without naming unrelated workflows.
+
 ### v1.9.30 Highlights
 
-PR integration release for dream-model routing, email bridge TLS opt-outs, installed CLI runtime resolution, and an OpenClaw plugin workflow example.
+PR integration release for dream-model routing, email bridge TLS opt-outs, and installed CLI runtime resolution.
 
 - **Dream model routing.** Memory dream cycles and daily digests can use optional `dreamProvider` settings so background consolidation can run on a smaller local model.
 - **Email bridge TLS opt-outs.** `tlsRejectUnauthorized=false` now disables hostname checks too, matching the explicit self-signed-server opt-out.
 - **Installed CLI stability.** Legacy API-backed CLI commands import the package-local `tsx` runtime instead of resolving `tsx` from the caller's project.
-- **OpenClaw plugin workflow.** README guidance now includes a concrete TweetClaw plugin workflow for OpenClaw operators.
 
 ### v1.9.29 Highlights
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@swarmclawai/swarmclaw",
-  "version": "1.9.30",
+  "version": "1.9.32",
   "description": "Build and run autonomous AI agents with OpenClaw, Hermes, multiple model providers, orchestration, delegation, memory, skills, schedules, and chat connectors.",
   "main": "electron-dist/main.js",
   "license": "MIT",
@@ -88,7 +88,7 @@
     "test:cli": "node --test src/cli/*.test.js bin/*.test.js scripts/electron-after-pack.test.mjs scripts/electron-signing-config.test.mjs scripts/ensure-sandbox-browser-image.test.mjs scripts/postinstall.test.mjs scripts/run-next-build.test.mjs scripts/run-next-typegen.test.mjs",
     "test:setup": "tsx --test src/app/api/setup/check-provider/route.test.ts src/lib/server/provider-model-discovery.test.ts src/components/auth/setup-wizard/utils.test.ts src/components/auth/setup-wizard/types.test.ts src/hooks/setup-done-detection.test.ts src/lib/setup-defaults.test.ts src/lib/server/storage-auth.test.ts src/lib/server/storage-auth-docker.test.ts",
     "test:openclaw": "tsx --test src/lib/openclaw/openclaw-agent-id.test.ts src/lib/openclaw/openclaw-endpoint.test.ts src/lib/server/agents/agent-runtime-config.test.ts src/lib/server/build-llm.test.ts src/lib/server/connectors/connector-routing.test.ts src/lib/server/connectors/openclaw.test.ts src/lib/server/connectors/swarmdock.test.ts src/lib/server/gateway/protocol.test.ts src/lib/server/gateways/gateway-topology.test.ts src/lib/server/llm-response-cache.test.ts src/lib/server/mcp-conformance.test.ts src/lib/server/openclaw/agent-resolver.test.ts src/lib/server/openclaw/deploy.test.ts src/lib/server/openclaw/skills-normalize.test.ts src/lib/server/session-tools/openclaw-nodes.test.ts src/lib/server/session-tools/swarmdock.test.ts src/lib/server/tasks/task-quality-gate.test.ts src/lib/server/tasks/task-validation.test.ts src/lib/server/tool-capability-policy.test.ts src/lib/providers/openai.test.ts src/lib/providers/openclaw-exports.test.ts src/app/api/gateways/topology-route.test.ts src/app/api/openclaw/dashboard-url/route.test.ts",
-    "test:runtime": "tsx --test src/lib/a2a/agent-card.test.ts src/lib/agent-planning-mode.test.ts src/lib/agent-config-history.test.ts src/lib/strip-internal-metadata.test.ts src/lib/provider-sets.test.ts src/lib/providers/opencode-cli.test.ts src/lib/providers/cli-provider-metadata.test.ts src/lib/providers/cli-utils.test.ts src/lib/providers/generic-cli.test.ts src/lib/server/agents/delegation-advisory.test.ts src/lib/server/agents/agent-runtime-config.test.ts src/lib/server/cli-provider-readiness.test.ts src/lib/server/provider-health.test.ts src/lib/server/provider-diagnostics.test.ts src/lib/server/mcp-gateway-runtime.test.ts src/lib/server/mcp-connection-pool.test.ts src/lib/server/knowledge-sources.test.ts src/lib/server/memory/dream-service.test.ts src/lib/server/extension-managed-resources.test.ts src/lib/server/eval/baseline.test.ts src/lib/server/eval/environment-plan.test.ts src/lib/server/chat-execution/chat-execution-grounding.test.ts src/lib/server/chat-execution/chat-turn-preparation.test.ts src/lib/server/chat-execution/iteration-timers.test.ts src/lib/server/chat-execution/post-stream-finalization.test.ts src/lib/server/chat-execution/prompt-sections.planning-mode.test.ts src/lib/server/chat-execution/reasoning-tag-scrubber.test.ts src/lib/server/chats/clear-undo-snapshots.test.ts src/lib/server/chats/session-context-pack.test.ts src/lib/server/connectors/email.test.ts src/lib/server/connectors/slack.test.ts src/lib/server/protocols/protocol-service.test.ts src/lib/server/runtime/run-ledger.test.ts src/lib/server/runtime/queue-retry-policy.test.ts src/lib/server/runs/run-brief.test.ts src/lib/server/runs/run-handoff.test.ts src/lib/server/operations/operation-pulse.test.ts src/lib/server/schedules/schedule-history.test.ts src/lib/server/schedules/schedule-timing.test.ts src/lib/server/schedules/schedule-preview.test.ts src/lib/quality/release-readiness.test.ts src/lib/quality/architecture-health.test.ts src/lib/server/artifacts/artifact-resolver.test.ts src/lib/server/observability/otel-config.test.ts src/lib/server/safe-parse-body.test.ts src/lib/server/missions/mission-templates.test.ts src/lib/server/sharing/share-link-repository.test.ts src/lib/server/sharing/share-resolver.test.ts src/lib/server/tasks/task-execution-workspace.test.ts src/lib/server/tasks/task-execution-policy.test.ts src/lib/server/tasks/task-handoff.test.ts src/lib/server/tasks/task-service.test.ts src/lib/server/session-tools/execute.test.ts src/lib/server/session-tools/manage-tasks.test.ts src/lib/server/session-tools/web-crawl.test.ts src/lib/app/view-constants.test.ts src/lib/quality/quality-summary.test.ts src/app/api/approvals/route.test.ts src/app/api/agents/agents-route.test.ts src/app/api/tasks/tasks-route.test.ts src/app/api/tasks/task-workspace-route.test.ts src/app/api/chats/chat-route.test.ts src/app/api/chats/clear-route.test.ts src/app/api/chats/compact-route.test.ts src/app/api/chats/context-pack-route.test.ts src/app/api/chats/context-status-route.test.ts src/app/api/config-versions/config-versions-route.test.ts src/app/api/runs/run-handoff-route.test.ts src/app/api/connectors/connector-doctor-route.test.ts src/app/api/extensions/managed-resources/route.test.ts src/app/api/gateways/control-route.test.ts src/app/api/healthz/route.test.ts src/app/api/logs/route.test.ts src/app/api/portability/export/route.test.ts src/app/api/portability/import/route.test.ts src/app/api/providers/[id]/route.test.ts src/app/api/schedules/preview/route.test.ts src/app/api/schedules/schedule-history-route.test.ts src/app/api/tts/route.test.ts",
+    "test:runtime": "tsx --test src/lib/a2a/agent-card.test.ts src/lib/agent-planning-mode.test.ts src/lib/agent-config-history.test.ts src/lib/autonomy/supervisor-settings.test.ts src/lib/strip-internal-metadata.test.ts src/lib/provider-sets.test.ts src/lib/providers/opencode-cli.test.ts src/lib/providers/cli-provider-metadata.test.ts src/lib/providers/cli-utils.test.ts src/lib/providers/generic-cli.test.ts src/lib/server/agents/delegation-advisory.test.ts src/lib/server/agents/agent-runtime-config.test.ts src/lib/server/autonomy/supervisor-reflection.test.ts src/lib/server/cli-provider-readiness.test.ts src/lib/server/provider-health.test.ts src/lib/server/provider-diagnostics.test.ts src/lib/server/mcp-gateway-runtime.test.ts src/lib/server/mcp-connection-pool.test.ts src/lib/server/knowledge-sources.test.ts src/lib/server/memory/dream-service.test.ts src/lib/server/memory/memory-consolidation.test.ts src/lib/server/extension-managed-resources.test.ts src/lib/server/eval/baseline.test.ts src/lib/server/eval/environment-plan.test.ts src/lib/server/chat-execution/chat-execution-grounding.test.ts src/lib/server/chat-execution/chat-turn-preparation.test.ts src/lib/server/chat-execution/compaction-generation-preference.test.ts src/lib/server/chat-execution/iteration-timers.test.ts src/lib/server/chat-execution/post-stream-finalization.test.ts src/lib/server/chat-execution/prompt-sections.planning-mode.test.ts src/lib/server/chat-execution/reasoning-tag-scrubber.test.ts src/lib/server/chats/clear-undo-snapshots.test.ts src/lib/server/chats/session-context-pack.test.ts src/lib/server/connectors/email.test.ts src/lib/server/connectors/slack.test.ts src/lib/server/protocols/protocol-service.test.ts src/lib/server/runtime/run-ledger.test.ts src/lib/server/runtime/queue-retry-policy.test.ts src/lib/server/runs/run-brief.test.ts src/lib/server/runs/run-handoff.test.ts src/lib/server/operations/operation-pulse.test.ts src/lib/server/schedules/schedule-history.test.ts src/lib/server/schedules/schedule-timing.test.ts src/lib/server/schedules/schedule-preview.test.ts src/lib/quality/release-readiness.test.ts src/lib/quality/architecture-health.test.ts src/lib/server/artifacts/artifact-resolver.test.ts src/lib/server/observability/otel-config.test.ts src/lib/server/safe-parse-body.test.ts src/lib/server/missions/mission-templates.test.ts src/lib/server/sharing/share-link-repository.test.ts src/lib/server/sharing/share-resolver.test.ts src/lib/server/tasks/task-execution-workspace.test.ts src/lib/server/tasks/task-execution-policy.test.ts src/lib/server/tasks/task-handoff.test.ts src/lib/server/tasks/task-service.test.ts src/lib/server/session-tools/execute.test.ts src/lib/server/session-tools/manage-tasks.test.ts src/lib/server/session-tools/web-crawl.test.ts src/lib/app/view-constants.test.ts src/lib/quality/quality-summary.test.ts src/app/api/approvals/route.test.ts src/app/api/agents/agents-route.test.ts src/app/api/tasks/tasks-route.test.ts src/app/api/tasks/task-workspace-route.test.ts src/app/api/chats/chat-route.test.ts src/app/api/chats/clear-route.test.ts src/app/api/chats/compact-route.test.ts src/app/api/chats/context-pack-route.test.ts src/app/api/chats/context-status-route.test.ts src/app/api/config-versions/config-versions-route.test.ts src/app/api/runs/run-handoff-route.test.ts src/app/api/connectors/connector-doctor-route.test.ts src/app/api/extensions/managed-resources/route.test.ts src/app/api/gateways/control-route.test.ts src/app/api/healthz/route.test.ts src/app/api/logs/route.test.ts src/app/api/portability/export/route.test.ts src/app/api/portability/import/route.test.ts src/app/api/providers/[id]/route.test.ts src/app/api/schedules/preview/route.test.ts src/app/api/schedules/schedule-history-route.test.ts src/app/api/tts/route.test.ts",
     "test:builder": "tsx --test src/features/protocols/builder/utils/nodes-to-template.test.ts src/features/protocols/builder/utils/template-to-nodes.test.ts src/features/protocols/builder/validators/dag-validator.test.ts",
     "test:e2e": "node --import tsx scripts/browser-e2e-smoke.ts",
     "test:mcp:conformance": "node --import tsx ./scripts/mcp-conformance-check.ts",

package/src/lib/autonomy/supervisor-settings.test.ts

@@ -0,0 +1,26 @@
+import assert from 'node:assert/strict'
+import { describe, it } from 'node:test'
+import { normalizeSupervisorSettings } from './supervisor-settings'
+
+describe('normalizeSupervisorSettings', () => {
+  it('preserves and clamps reflection memory quality settings', () => {
+    assert.deepEqual(
+      {
+        minQuality: normalizeSupervisorSettings({ reflectionMinQuality: '0.72' }).reflectionMinQuality,
+        minQualityHigh: normalizeSupervisorSettings({ reflectionMinQuality: 2 }).reflectionMinQuality,
+        minQualityLow: normalizeSupervisorSettings({ reflectionMinQuality: -1 }).reflectionMinQuality,
+        semanticEnabled: normalizeSupervisorSettings({ reflectionSemanticDedupEnabled: 'on' }).reflectionSemanticDedupEnabled,
+        semanticThreshold: normalizeSupervisorSettings({ reflectionSemanticDedupThreshold: '0.91' }).reflectionSemanticDedupThreshold,
+        semanticThresholdHigh: normalizeSupervisorSettings({ reflectionSemanticDedupThreshold: 1.5 }).reflectionSemanticDedupThreshold,
+      },
+      {
+        minQuality: 0.72,
+        minQualityHigh: 1,
+        minQualityLow: 0,
+        semanticEnabled: true,
+        semanticThreshold: 0.91,
+        semanticThresholdHigh: 1,
+      },
+    )
+  })
+})

package/src/lib/autonomy/supervisor-settings.ts

@@ -8,6 +8,9 @@ export const DEFAULT_SUPERVISOR_NO_PROGRESS_LIMIT = 2
 export const DEFAULT_SUPERVISOR_REPEATED_TOOL_LIMIT = 3
 export const DEFAULT_REFLECTION_ENABLED = true
 export const DEFAULT_REFLECTION_AUTO_WRITE_MEMORY = true
+export const DEFAULT_REFLECTION_MIN_QUALITY = 0
+export const DEFAULT_REFLECTION_SEMANTIC_DEDUP_ENABLED = false
+export const DEFAULT_REFLECTION_SEMANTIC_DEDUP_THRESHOLD = 0.88
 
 export const SUPERVISOR_NO_PROGRESS_LIMIT_MIN = 1
 export const SUPERVISOR_NO_PROGRESS_LIMIT_MAX = 8
@@ -24,6 +27,16 @@ function parseIntSetting(value: unknown, fallback: number, min: number, max: num
   return Math.max(min, Math.min(max, Math.trunc(parsed)))
 }
 
+function parseNumberSetting(value: unknown, fallback: number, min: number, max: number): number {
+  const parsed = typeof value === 'number'
+    ? value
+    : typeof value === 'string'
+      ? Number.parseFloat(value)
+      : Number.NaN
+  if (!Number.isFinite(parsed)) return fallback
+  return Math.max(min, Math.min(max, parsed))
+}
+
 function parseBoolSetting(value: unknown, fallback: boolean): boolean {
   if (typeof value === 'boolean') return value
   if (typeof value === 'string') {
@@ -41,6 +54,9 @@ export interface NormalizedSupervisorSettings {
   supervisorRepeatedToolLimit: number
   reflectionEnabled: boolean
   reflectionAutoWriteMemory: boolean
+  reflectionMinQuality: number
+  reflectionSemanticDedupEnabled: boolean
+  reflectionSemanticDedupThreshold: number
 }
 
 export function normalizeSupervisorSettings(
@@ -69,6 +85,17 @@ export function normalizeSupervisorSettings(
     ),
     reflectionEnabled: parseBoolSetting(current.reflectionEnabled, DEFAULT_REFLECTION_ENABLED),
     reflectionAutoWriteMemory: parseBoolSetting(current.reflectionAutoWriteMemory, DEFAULT_REFLECTION_AUTO_WRITE_MEMORY),
+    reflectionMinQuality: parseNumberSetting(current.reflectionMinQuality, DEFAULT_REFLECTION_MIN_QUALITY, 0, 1),
+    reflectionSemanticDedupEnabled: parseBoolSetting(
+      current.reflectionSemanticDedupEnabled,
+      DEFAULT_REFLECTION_SEMANTIC_DEDUP_ENABLED,
+    ),
+    reflectionSemanticDedupThreshold: parseNumberSetting(
+      current.reflectionSemanticDedupThreshold,
+      DEFAULT_REFLECTION_SEMANTIC_DEDUP_THRESHOLD,
+      0,
+      1,
+    ),
   }
 }
 

package/src/lib/server/autonomy/supervisor-reflection.test.ts

@@ -202,6 +202,212 @@ describe('supervisor-reflection', () => {
     ])
   })
 
+  it('persists low-quality reflections while skipping auto-written memory', () => {
+    const output = runWithTempDataDir(`
+      const storageMod = await import('@/lib/server/storage')
+      const storage = storageMod.default || storageMod['module.exports'] || storageMod
+      const reflectionMod = await import('@/lib/server/autonomy/supervisor-reflection')
+      const mod = reflectionMod.default || reflectionMod['module.exports'] || reflectionMod
+      const memoryDbMod = await import('@/lib/server/memory/memory-db')
+      const memoryMod = memoryDbMod.default || memoryDbMod['module.exports'] || memoryDbMod
+
+      storage.saveAgents({
+        'agent-a': {
+          id: 'agent-a',
+          name: 'Agent A',
+          provider: 'openai',
+          model: 'gpt-test',
+        },
+      })
+
+      storage.saveSessions({
+        s1: {
+          id: 's1',
+          name: 'Autonomy Session',
+          cwd: process.cwd(),
+          user: 'tester',
+          provider: 'openai',
+          model: 'gpt-test',
+          claudeSessionId: null,
+          messages: [
+            { role: 'user', text: 'Repair the deployment workflow and keep notes for later.', time: 1 },
+            { role: 'assistant', text: 'I retried the same shell path and nothing changed.', time: 2 },
+          ],
+          createdAt: 1,
+          lastActiveAt: 2,
+          sessionType: 'human',
+          agentId: 'agent-a',
+        },
+      })
+
+      storage.saveSettings({
+        supervisorEnabled: true,
+        supervisorRuntimeScope: 'both',
+        supervisorNoProgressLimit: 2,
+        supervisorRepeatedToolLimit: 3,
+        reflectionEnabled: true,
+        reflectionAutoWriteMemory: true,
+        reflectionMinQuality: 0.8,
+      })
+
+      const result = await mod.observeAutonomyRunOutcome({
+        runId: 'run-low-quality',
+        sessionId: 's1',
+        agentId: 'agent-a',
+        source: 'chat',
+        status: 'completed',
+        resultText: 'I retried the same shell path and nothing changed.',
+        toolEvents: [
+          { name: 'shell', input: '{"cmd":"npm test"}' },
+          { name: 'shell', input: '{"cmd":"npm test"}' },
+          { name: 'shell', input: '{"cmd":"npm test"}' },
+        ],
+        mainLoopState: {
+          followupChainCount: 2,
+          summary: 'I retried the same shell path and nothing changed.',
+        },
+        sourceMessage: 'Repair the deployment workflow and keep notes for later.',
+      }, {
+        generateText: async () => JSON.stringify({
+          summary: 'Low quality reflection',
+          lessons: ['This weak note should not be written to memory.'],
+          quality_score: 0.25,
+          quality_reasoning: 'Too generic to keep as durable memory.',
+        }),
+      })
+
+      const memories = memoryMod.getMemoryDb().list(undefined, 50)
+        .filter((entry) => entry.metadata && entry.metadata.origin === 'autonomy-reflection')
+
+      console.log(JSON.stringify({
+        reflectionSummary: result.reflection?.summary ?? null,
+        qualityScore: result.reflection?.qualityScore ?? null,
+        autoMemoryCount: result.reflection?.autoMemoryIds?.length ?? 0,
+        storedReflectionMemoryCount: memories.length,
+      }))
+    `)
+
+    assert.equal(output.reflectionSummary, 'Low quality reflection')
+    assert.equal(output.qualityScore, 0.25)
+    assert.equal(output.autoMemoryCount, 0)
+    assert.equal(output.storedReflectionMemoryCount, 0)
+  })
+
+  it('skips semantically duplicate reflection memory when embedding dedup is enabled', () => {
+    const output = runWithTempDataDir(`
+      const http = await import('node:http')
+      const path = await import('node:path')
+      const Database = (await import('better-sqlite3')).default
+      const storageMod = await import('@/lib/server/storage')
+      const storage = storageMod.default || storageMod['module.exports'] || storageMod
+      const reflectionMod = await import('@/lib/server/autonomy/supervisor-reflection')
+      const mod = reflectionMod.default || reflectionMod['module.exports'] || reflectionMod
+      const memoryDbMod = await import('@/lib/server/memory/memory-db')
+      const memoryMod = memoryDbMod.default || memoryDbMod['module.exports'] || memoryDbMod
+      const settingsRepositoryMod = await import('@/lib/server/settings/settings-repository')
+      const settingsRepository = settingsRepositoryMod.default || settingsRepositoryMod['module.exports'] || settingsRepositoryMod
+
+      const server = http.createServer((req, res) => {
+        res.setHeader('content-type', 'application/json')
+        res.end(JSON.stringify({ embedding: [1, 0] }))
+      })
+      await new Promise((resolve) => server.listen(0, '127.0.0.1', resolve))
+      const endpoint = 'http://127.0.0.1:' + server.address().port
+
+      try {
+        storage.saveAgents({
+          'agent-a': {
+            id: 'agent-a',
+            name: 'Agent A',
+            provider: 'openai',
+            model: 'gpt-test',
+          },
+        })
+
+        storage.saveSessions({
+          s1: {
+            id: 's1',
+            name: 'Semantic Dedup Session',
+            cwd: process.cwd(),
+            user: 'tester',
+            provider: 'openai',
+            model: 'gpt-test',
+            claudeSessionId: null,
+            messages: [
+              { role: 'user', text: 'Release carefully and keep the durable lesson.', time: 1 },
+              { role: 'assistant', text: 'I checked the release gates before shipping.', time: 2 },
+            ],
+            createdAt: 1,
+            lastActiveAt: 2,
+            sessionType: 'human',
+            agentId: 'agent-a',
+          },
+        })
+
+        const memDb = memoryMod.getMemoryDb()
+        const existing = memDb.add({
+          agentId: 'agent-a',
+          sessionId: 's1',
+          category: 'reflection/lesson',
+          title: 'Reflection Lesson',
+          content: 'Always verify release gates before shipping.',
+          metadata: { origin: 'autonomy-reflection' },
+        })
+        const rawDb = new Database(path.join(process.env.DATA_DIR, 'memory.db'))
+        rawDb.prepare('UPDATE memories SET embedding = ? WHERE id = ?').run(Buffer.from(new Float32Array([1, 0]).buffer), existing.id)
+        rawDb.close()
+
+        settingsRepository.saveSettings({
+          supervisorEnabled: true,
+          supervisorRuntimeScope: 'both',
+          supervisorNoProgressLimit: 2,
+          supervisorRepeatedToolLimit: 3,
+          reflectionEnabled: true,
+          reflectionAutoWriteMemory: true,
+          reflectionSemanticDedupEnabled: true,
+          reflectionSemanticDedupThreshold: 0.9,
+          embeddingProvider: 'ollama',
+          embeddingModel: 'test-embedding',
+          embeddingEndpoint: endpoint,
+        })
+
+        const result = await mod.observeAutonomyRunOutcome({
+          runId: 'run-semantic-dedup',
+          sessionId: 's1',
+          agentId: 'agent-a',
+          source: 'chat',
+          status: 'completed',
+          resultText: 'I checked the release gates before shipping.',
+          toolEvents: [
+            { name: 'shell', input: '{"cmd":"npm test"}' },
+          ],
+          sourceMessage: 'Release carefully and keep the durable lesson.',
+        }, {
+          generateText: async () => JSON.stringify({
+            summary: 'Release gate reflection',
+            lessons: ['Confirm release gates before shipping.'],
+            quality_score: 0.95,
+          }),
+        })
+
+        const reflectionMemories = memDb.list('agent-a', 50)
+          .filter((entry) => entry.metadata && entry.metadata.origin === 'autonomy-reflection')
+
+        console.log(JSON.stringify({
+          reflectionSummary: result.reflection?.summary ?? null,
+          autoMemoryCount: result.reflection?.autoMemoryIds?.length ?? 0,
+          reflectionMemoryContents: reflectionMemories.map((entry) => entry.content).sort(),
+        }))
+      } finally {
+        await new Promise((resolve) => server.close(resolve))
+      }
+    `)
+
+    assert.equal(output.reflectionSummary, 'Release gate reflection')
+    assert.equal(output.autoMemoryCount, 0)
+    assert.deepEqual(output.reflectionMemoryContents, ['Always verify release gates before shipping.'])
+  })
+
   it('reflects short human chats when they contain durable personal context', () => {
     const output = runWithTempDataDir(`
       const storageMod = await import('@/lib/server/storage')

package/src/lib/server/autonomy/supervisor-reflection.ts

@@ -744,7 +744,7 @@ function inferFollowUpAt(note: string, createdAt: number): number {
   return createdAt + 7 * 24 * 3600_000
 }
 
-function writeReflectionMemories(params: {
+async function writeReflectionMemories(params: {
   reflectionId: string
   runId: string
   sessionId: string
@@ -761,7 +761,7 @@ function writeReflectionMemories(params: {
   profile: string[]
   boundaries: string[]
   openLoops: string[]
-}): string[] {
+}): Promise<string[]> {
   const memoryDb = getMemoryDb()
   const memoryIds: string[] = []
   const incidentIds = params.incidents.map((incident) => incident.id)
@@ -809,11 +809,51 @@ function writeReflectionMemories(params: {
     // dedup only rather than blocking the reflection write.
   }
 
+  // Semantic dedup (opt-in): on top of the text-equality cross-run dedup
+  // above, compare each candidate note's embedding against recent reflection
+  // memories' embeddings. Catches near-duplicates the LLM re-derives in
+  // different words ("Always verify before acting" / "Confirm state first").
+  // Falls back gracefully when embeddings aren't configured.
+  let reflectionSettings: NormalizedSupervisorSettings | null = null
+  try { reflectionSettings = normalizeSupervisorSettings(loadSettings()) } catch { reflectionSettings = null }
+  const semanticDedupEnabled = reflectionSettings?.reflectionSemanticDedupEnabled === true
+  const semanticDedupThreshold = reflectionSettings?.reflectionSemanticDedupThreshold ?? 0.88
+  const semanticSkip = new Set<string>()
+  if (semanticDedupEnabled && params.agentId) {
+    try {
+      const recentEmb = memoryDb.recentReflectionEmbeddings(params.agentId, crossRunDedupCutoff, 500)
+        .filter((r) => Array.isArray(r.embedding) && r.embedding.length > 0) as Array<{ id: string; content: string; embedding: number[] }>
+      if (recentEmb.length > 0) {
+        const { getEmbedding, cosineSimilarity } = await import('@/lib/server/embeddings')
+        for (const group of groups) {
+          for (const note of group.notes) {
+            const trimmed = (note || '').trim()
+            if (!trimmed) continue
+            const norm = normalizeNote(trimmed)
+            if (!norm || seenNormalized.has(norm) || semanticSkip.has(norm)) continue
+            const emb = await getEmbedding(trimmed)
+            if (!emb) continue
+            for (const r of recentEmb) {
+              if (cosineSimilarity(emb, r.embedding) >= semanticDedupThreshold) {
+                semanticSkip.add(norm)
+                break
+              }
+            }
+          }
+        }
+      }
+    } catch {
+      // Best-effort: any failure (embedder offline, DB blip) falls through to
+      // the existing text-equality dedup. Never block the write.
+    }
+  }
+
   for (const group of groups) {
     for (const note of group.notes) {
       const norm = normalizeNote(note)
       if (!norm) continue
       if (seenNormalized.has(norm)) continue
+      if (semanticSkip.has(norm)) continue
       seenNormalized.add(norm)
       const metadata: Record<string, unknown> = {
         origin: 'autonomy-reflection',
@@ -1085,8 +1125,18 @@ export async function observeAutonomyRunOutcome(
   if (parsed.skip) return { incidents, reflection: null }
 
   const reflectionId = genId()
-  const autoMemoryIds = settings.reflectionAutoWriteMemory
-    ? writeReflectionMemories({
+  const minQuality = typeof settings.reflectionMinQuality === 'number' ? settings.reflectionMinQuality : 0
+  const qualityScore = parsed.qualityScore
+  const qualityGateOpen = minQuality <= 0
+    || qualityScore == null
+    || qualityScore >= minQuality
+  if (!qualityGateOpen) {
+    log.info(TAG,
+      `Reflection ${reflectionId} below quality gate (score=${qualityScore?.toFixed(2) ?? 'null'}, threshold=${minQuality.toFixed(2)}); skipping memory writes`,
+    )
+  }
+  const autoMemoryIds = settings.reflectionAutoWriteMemory && qualityGateOpen
+    ? await writeReflectionMemories({
         reflectionId,
         runId: input.runId,
         sessionId: input.sessionId,

package/src/lib/server/chat-execution/compaction-generation-preference.test.ts

@@ -0,0 +1,24 @@
+import assert from 'node:assert/strict'
+import { describe, it } from 'node:test'
+import { resolveCompactionGenerationPreference } from './compaction-generation-preference'
+
+describe('resolveCompactionGenerationPreference', () => {
+  it('returns no preference when no compaction provider is configured', () => {
+    assert.equal(resolveCompactionGenerationPreference({}), undefined)
+    assert.equal(resolveCompactionGenerationPreference({ compactionProvider: '   ' }), undefined)
+  })
+
+  it('builds a trimmed compaction model preference from app settings', () => {
+    assert.deepEqual(resolveCompactionGenerationPreference({
+      compactionProvider: ' ollama ',
+      compactionModel: ' llama3.2:3b ',
+      compactionCredentialId: ' cred-1 ',
+      compactionEndpoint: ' http://localhost:11434 ',
+    }), {
+      provider: 'ollama',
+      model: 'llama3.2:3b',
+      credentialId: 'cred-1',
+      apiEndpoint: 'http://localhost:11434',
+    })
+  })
+})

package/src/lib/server/chat-execution/compaction-generation-preference.ts

@@ -0,0 +1,25 @@
+import type { GenerationModelPreference } from '@/lib/server/build-llm'
+import type { AppSettings } from '@/types'
+
+type CompactionGenerationSettings = Pick<AppSettings, 'compactionProvider' | 'compactionModel' | 'compactionCredentialId' | 'compactionEndpoint'> | Record<string, unknown> | null | undefined
+
+function optionalSettingString(value: unknown): string | undefined {
+  const normalized = typeof value === 'string' ? value.trim() : ''
+  return normalized || undefined
+}
+
+/** Mirrors resolveDreamGenerationPreference — returns a model preference for
+ *  the auto-compaction summarizer if app settings opt into a routing override,
+ *  otherwise undefined (caller falls back to the session's primary model). */
+export function resolveCompactionGenerationPreference(settings: CompactionGenerationSettings): GenerationModelPreference | undefined {
+  const record = (settings || {}) as Record<string, unknown>
+  const provider = optionalSettingString(record.compactionProvider)
+  if (!provider) return undefined
+
+  return {
+    provider,
+    model: optionalSettingString(record.compactionModel),
+    credentialId: optionalSettingString(record.compactionCredentialId),
+    apiEndpoint: optionalSettingString(record.compactionEndpoint),
+  }
+}

package/src/lib/server/chat-execution/stream-agent-chat.ts

@@ -682,8 +682,29 @@ async function streamAgentChatCore(opts: StreamAgentChatOpts): Promise<StreamAge
       reserveTokens,
       includeToolEvents: false,
     })) {
+      // Resolve compaction model: if app settings opt into an override, build
+      // a separate LLM for the summarizer (cheap/local model); otherwise reuse
+      // the session's primary llm. Mirrors the dream-model override path.
+      const { resolveCompactionGenerationPreference } = await import('@/lib/server/chat-execution/compaction-generation-preference')
+      const { buildLLM } = await import('@/lib/server/build-llm')
+      // loadSettings is imported at the top of this file.
+      const settings = loadSettings()
+      const compactionPref = resolveCompactionGenerationPreference(settings)
+      let summarizerLlm = llm
+      let summarizerProvider = session.provider
+      let summarizerModel = session.model
+      if (compactionPref) {
+        try {
+          const built = await buildLLM({ preferred: compactionPref, sessionId: session.id, agentId: session.agentId || null })
+          summarizerLlm = built.llm
+          summarizerProvider = built.provider
+          summarizerModel = built.model
+        } catch (overrideErr) {
+          log.warn(TAG, `Compaction override LLM build failed for ${session.id}; falling back to session model:`, overrideErr)
+        }
+      }
       const summarize = async (prompt: string): Promise<string> => {
-        const response = await llm.invoke([new HumanMessage(prompt)])
+        const response = await summarizerLlm.invoke([new HumanMessage(prompt)])
         if (typeof response.content === 'string') return response.content
         if (Array.isArray(response.content)) {
           return response.content
@@ -694,8 +715,8 @@ async function streamAgentChatCore(opts: StreamAgentChatOpts): Promise<StreamAge
       }
       const result = await llmCompact({
         messages: recentHistory,
-        provider: session.provider,
-        model: session.model,
+        provider: summarizerProvider,
+        model: summarizerModel,
         agentId: session.agentId || null,
         sessionId: session.id,
         summarize,
@@ -704,6 +725,7 @@ async function streamAgentChatCore(opts: StreamAgentChatOpts): Promise<StreamAge
       log.info(TAG,
         `Auto-compacted ${session.id}: ${recentHistory.length} → ${effectiveHistory.length} msgs` +
         ` (prompt history ${promptHistoryTokens} tokens)` +
+        (compactionPref ? ` (override ${summarizerProvider}/${summarizerModel})` : '') +
         (result.summaryAdded ? ' (LLM summary)' : ' (sliding window fallback)'),
       )
     }

package/src/lib/server/memory/dream-generation-preference.ts

@@ -1,14 +1,42 @@
 import type { GenerationModelPreference } from '@/lib/server/build-llm'
-import type { AppSettings } from '@/types'
+import type { AppSettings, DreamConfig } from '@/types'
 
 type DreamGenerationSettings = Pick<AppSettings, 'dreamProvider' | 'dreamModel' | 'dreamCredentialId' | 'dreamEndpoint'> | Record<string, unknown> | null | undefined
 
+type DreamConfigOverride = Pick<DreamConfig, 'provider' | 'model' | 'credentialId' | 'endpoint'> | Partial<DreamConfig> | Record<string, unknown> | null | undefined
+
 function optionalSettingString(value: unknown): string | undefined {
   const normalized = typeof value === 'string' ? value.trim() : ''
   return normalized || undefined
 }
 
-export function resolveDreamGenerationPreference(settings: DreamGenerationSettings): GenerationModelPreference | undefined {
+/**
+ * Resolve which model to use for memory consolidation / dream cycles.
+ *
+ * Precedence:
+ *   1. Per-agent override (`dreamConfig.provider` on the Agent record)
+ *   2. Global app settings (`dreamProvider` etc.)
+ *   3. undefined — caller falls back to the agent's primary generation model
+ *
+ * The per-agent override lets you route different agents to different dream
+ * models (e.g. cheap local for most, but a stronger model for an agent whose
+ * memory mix needs more capable structured-output generation).
+ */
+export function resolveDreamGenerationPreference(
+  settings: DreamGenerationSettings,
+  override?: DreamConfigOverride,
+): GenerationModelPreference | undefined {
+  const overrideRecord = (override || {}) as Record<string, unknown>
+  const overrideProvider = optionalSettingString(overrideRecord.provider)
+  if (overrideProvider) {
+    return {
+      provider: overrideProvider,
+      model: optionalSettingString(overrideRecord.model),
+      credentialId: optionalSettingString(overrideRecord.credentialId),
+      apiEndpoint: optionalSettingString(overrideRecord.endpoint),
+    }
+  }
+
   const record = (settings || {}) as Record<string, unknown>
   const provider = optionalSettingString(record.dreamProvider)
   if (!provider) return undefined

package/src/lib/server/memory/dream-service.test.ts

@@ -22,6 +22,25 @@ describe('resolveDreamGenerationPreference', () => {
       apiEndpoint: 'http://localhost:11434',
     })
   })
+
+  it('uses a per-agent dream override before global app settings', () => {
+    assert.deepEqual(resolveDreamGenerationPreference({
+      dreamProvider: 'openai',
+      dreamModel: 'gpt-5-mini',
+      dreamCredentialId: 'global-cred',
+      dreamEndpoint: 'https://global.example/v1',
+    }, {
+      provider: ' ollama ',
+      model: ' qwen3:8b ',
+      credentialId: ' agent-cred ',
+      endpoint: ' http://127.0.0.1:11434 ',
+    }), {
+      provider: 'ollama',
+      model: 'qwen3:8b',
+      credentialId: 'agent-cred',
+      apiEndpoint: 'http://127.0.0.1:11434',
+    })
+  })
 })
 
 describe('parseTier2DreamResponseText', () => {

package/src/lib/server/memory/dream-service.ts

@@ -216,7 +216,9 @@ ${memoryLines.join('\n')}`
   try {
     const { buildLLM } = await import('@/lib/server/build-llm')
     const { loadSettings } = await import('@/lib/server/settings/settings-repository')
-    const preferred = resolveDreamGenerationPreference(loadSettings())
+    // `config` is the resolved per-agent dream config (defaults + overrides);
+    // pass it so a per-agent provider/model takes precedence over global settings.
+    const preferred = resolveDreamGenerationPreference(loadSettings(), config)
     const { llm } = await buildLLM({ agentId, preferred, responseFormat: 'json_object' })
     const { HumanMessage } = await import('@langchain/core/messages')
 

package/src/lib/server/memory/memory-consolidation.test.ts

@@ -124,3 +124,34 @@ test('canCreateDailyDigestForAgent allows CLI-only agents when a dream model is
     true,
   )
 })
+
+test('canCreateDailyDigestForAgent allows CLI-only agents with a per-agent dream model override', async () => {
+  const now = Date.now()
+  const agentId = 'agent-dream-override-cli'
+  storage.saveSettings({})
+  storage.saveAgents({
+    [agentId]: {
+      id: agentId,
+      name: 'Per-Agent Dream Routed CLI Agent',
+      description: '',
+      systemPrompt: '',
+      provider: 'claude-cli',
+      model: 'claude-sonnet-4-5',
+      credentialId: null,
+      fallbackCredentialIds: [],
+      apiEndpoint: null,
+      dreamConfig: {
+        provider: 'ollama',
+        model: 'llama3.2',
+        endpoint: 'http://127.0.0.1:11434',
+      },
+      createdAt: now,
+      updatedAt: now,
+    } as Agent,
+  })
+
+  assert.equal(
+    consolidation.canCreateDailyDigestForAgent(agentId, storage.loadAgents({ includeTrashed: true }), storage.loadSettings()),
+    true,
+  )
+})

package/src/lib/server/memory/memory-consolidation.ts

@@ -48,7 +48,7 @@ export function canCreateDailyDigestForAgent(
   try {
     resolveGenerationModelConfig({
       agentId,
-      preferred: resolveDreamGenerationPreference(settings),
+      preferred: resolveDreamGenerationPreference(settings, agent.dreamConfig),
     })
     return true
   } catch (err: unknown) {
@@ -118,10 +118,11 @@ export async function runDailyConsolidation(): Promise<{
       ].join('\n')
 
       // Use an optional dream-model override before the target agent's generation provider.
+      // Precedence: per-agent dreamConfig override → global dream* settings → agent's primary.
       const { buildLLM } = await import('@/lib/server/build-llm')
       const { llm } = await buildLLM({
         agentId,
-        preferred: resolveDreamGenerationPreference(settings),
+        preferred: resolveDreamGenerationPreference(settings, agents[agentId]?.dreamConfig),
       })
 
       const response = await llm.invoke([new HumanMessage(prompt)])

package/src/lib/server/memory/memory-db.ts

@@ -1270,6 +1270,30 @@ function initDb() {
       return (stmts.listByAgent.all(agentId, safeLimit) as any[]).map(rowToEntry)
     },
 
+    /** Return recent reflection/* memories with their embeddings deserialized
+     *  for semantic dedup. Memories without an embedding (older rows, or
+     *  embedding still being computed in background) are included with a
+     *  null embedding so callers can fall back to text dedup. */
+    recentReflectionEmbeddings(
+      agentId: string,
+      sinceMs: number,
+      limit = 200,
+    ): Array<{ id: string; content: string; embedding: number[] | null }> {
+      const safeLimit = Math.max(1, Math.min(500, Math.trunc(limit)))
+      const rows = db.prepare(
+        `SELECT id, content, embedding FROM memories
+         WHERE (agentId = ? OR sharedWith LIKE ?)
+           AND category LIKE 'reflection/%'
+           AND updatedAt >= ?
+         ORDER BY updatedAt DESC LIMIT ?`,
+      ).all(agentId, `%"${agentId}"%`, sinceMs, safeLimit) as Array<{ id: string; content: string; embedding: Buffer | null }>
+      return rows.map((row) => ({
+        id: row.id,
+        content: row.content || '',
+        embedding: row.embedding ? deserializeEmbedding(row.embedding) : null,
+      }))
+    },
+
     getFrequentlyAccessedByAgent(agentId: string, minAccessCount = 3, sinceDays = 7): MemoryEntry[] {
       const cutoff = Date.now() - sinceDays * 86_400_000
       const rows = stmts.frequentlyAccessedByAgent.all(agentId, minAccessCount, cutoff) as Record<string, unknown>[]

package/src/types/app-settings.ts

@@ -32,6 +32,14 @@ export interface AppSettings {
   dreamModel?: string | null
   dreamCredentialId?: string | null
   dreamEndpoint?: string | null
+  // Optional model override for auto-compaction (live-loop conversation
+  // summarization triggered when context usage hits the auto-compact
+  // threshold). Lets the user route the summarizer to a cheaper or faster
+  // model than the agent's primary generation model.
+  compactionProvider?: string | null
+  compactionModel?: string | null
+  compactionCredentialId?: string | null
+  compactionEndpoint?: string | null
   loopMode?: LoopMode
   agentLoopRecursionLimit?: number
   delegationMaxDepth?: number
@@ -103,6 +111,12 @@ export interface AppSettings {
   autonomyResumeApprovalsEnabled?: boolean
   reflectionEnabled?: boolean
   reflectionAutoWriteMemory?: boolean
+  /** Minimum reflection quality score (0-1) required to auto-write memories. */
+  reflectionMinQuality?: number
+  /** Enable embedding-based dedup for reflection memory writes. */
+  reflectionSemanticDedupEnabled?: boolean
+  /** Cosine threshold above which a reflection note is considered duplicate. */
+  reflectionSemanticDedupThreshold?: number
   memoryReferenceDepth?: number
   maxMemoriesPerLookup?: number
   maxLinkedMemoriesExpanded?: number

package/src/types/dream.ts

@@ -33,6 +33,14 @@ export interface DreamConfig {
   pruneThresholdDays: number
   tier2Enabled: boolean
   tier2MaxMemories: number
+  // Optional per-agent override for the consolidation/dream LLM. When set,
+  // takes precedence over the global `dream*` app settings. When unset, the
+  // helper falls back to global settings, then to the agent's primary
+  // generation model — same precedence as before.
+  provider?: string | null
+  model?: string | null
+  credentialId?: string | null
+  endpoint?: string | null
 }
 
 export const DEFAULT_DREAM_CONFIG: DreamConfig = {