@mastra/memory 1.41.0 → 1.41.1-alpha.0

package/dist/docs/references/docs-memory-observational-memory.md

@@ -108,7 +108,7 @@ The agent now has humanlike long-term memory that persists across conversations.
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'deepseek/deepseek-reasoner',
+      model: 'deepseek/deepseek-flash',
     },
   },
 })
@@ -154,7 +154,7 @@ export const agent = new Agent({
   memory: new Memory({
     options: {
       observationalMemory: {
-        model: 'google/gemini-2.5-flash',
+        model: 'auto',
         temporalMarkers: true,
       },
     },
@@ -180,7 +180,7 @@ import { Memory } from '@mastra/memory'
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       activateAfterIdle: 'auto',
       activateOnProviderChange: true,
     },
@@ -196,7 +196,7 @@ import { Memory } from '@mastra/memory'
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       activateAfterIdle: '5m',
       observation: {
         activateAfterIdle: false,
@@ -455,7 +455,7 @@ new Agent({
     options: {
       observationalMemory: {
         observation: {
-          model: 'deepseek/deepseek-reasoner',
+          model: 'deepseek/deepseek-flash',
           observeAttachments: false,
         },
       },
@@ -522,7 +522,7 @@ Set `retrieval: true` to enable the recall tool for browsing raw messages. No ve
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       retrieval: true,
     },
   },
@@ -540,7 +540,7 @@ const memory = new Memory({
   embedder: myEmbedder,
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       retrieval: { vector: true },
     },
   },
@@ -557,7 +557,7 @@ By default, the recall tool scope is `'resource'`, the agent can list threads an
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       retrieval: { vector: true, scope: 'thread' },
     },
   },
@@ -572,7 +572,7 @@ Mastra injects scope-aware instructions that teach the agent when to search, lis
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       retrieval: {
         vector: true,
         instructions: `
@@ -641,13 +641,13 @@ const memory = new Memory({
 
 Mastra recommends using a model that has a large context window (128K+ tokens) and is fast enough to run in the background without slowing down your actions.
 
-To pin a model instead, set a concrete model ID. We've also successfully tested `openai/gpt-6-luna`, `anthropic/claude-haiku-5-5`, `deepseek/deepseek-reasoner`, `deepseek/deepseek-v4-pro`, `deepseek/deepseek-v4-flash`, `xai/grok-4-1-fast`, `qwen3`, and `glm-4.7`.
+To pin a model instead, set a concrete model ID. We've also successfully tested `openai/gpt-6-luna`, `anthropic/claude-haiku-5-5`, and `deepseek/deepseek-flash`.
 
 ```typescript
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'deepseek/deepseek-reasoner',
+      model: 'deepseek/deepseek-flash',
     },
   },
 })
@@ -655,9 +655,9 @@ const memory = new Memory({
 
 See [model configuration](https://mastra.ai/reference/memory/observational-memory) for using different models per agent.
 
-> **Note:** `google/gemini-2.5-flash` is unusually good at preserving detail in long output. As a result, the Reflector can produce reflections that stay above the configured `reflection.observationTokens` threshold even after the maximum compression retry. When this happens, the Reflector returns the smallest non-degenerate candidate produced during retries so the loop terminates instead of running forever.
+> **Note:** Some models preserve detail in long output unusually well. In testing, `google/gemini-2.5-flash` did this. Other models may behave differently. With such a model, the Reflector can produce reflections that stay above the configured `reflection.observationTokens` threshold even after the maximum compression retry. When this happens, the Reflector returns the smallest non-degenerate candidate produced during retries so the loop terminates instead of running forever.
 >
-> If you'd rather have more aggressive compression on the Reflector, swap to a model that condenses more readily, such as `xai/grok-4-1-fast`, `deepseek/deepseek-v4-pro`, or `deepseek/deepseek-v4-flash`. You can keep `google/gemini-2.5-flash` for the Observer and use a different model for the Reflector. See [different models per agent](https://mastra.ai/reference/memory/observational-memory).
+> If you'd rather have more aggressive compression on the Reflector, pick a model that condenses more readily, such as `deepseek/deepseek-flash`. You can use a detail-preserving model for the Observer and a condensing one for the Reflector. See [different models per agent](https://mastra.ai/reference/memory/observational-memory).
 
 ### Token-tiered model selection
 
@@ -678,7 +678,7 @@ const memory = new Memory({
             5_000: 'openrouter/mistralai/ministral-8b-2512',
             20_000: 'openrouter/mistralai/mistral-small-2603',
             40_000: 'openai/gpt-6-luna',
-            1_000_000: 'google/gemini-3.1-flash-lite-preview',
+            1_000_000: 'google/gemini-3.8-flash',
           },
         }),
       },
@@ -686,7 +686,7 @@ const memory = new Memory({
         model: new ModelByInputTokens({
           upTo: {
             20_000: 'openai/gpt-6-luna',
-            100_000: 'google/gemini-2.5-flash',
+            100_000: 'deepseek/deepseek-flash',
           },
         }),
       },
@@ -709,7 +709,7 @@ Each thread has its own observations. This scope is well tested and works well a
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       scope: 'thread',
     },
   },
@@ -728,7 +728,7 @@ In resource scope, observations are shared across all threads for a resource (ty
 - **Agent understanding:** Each thread sees a mix of observations from every thread for the resource. Every new thread starts with context from earlier threads, and the agent may continue work that another thread started but didn't finish.
 - **Performance:** Unobserved messages across _all_ threads are processed together, which is slow for users with many existing threads.
 
-A new knowledge and subconscious memory primitive for cross-thread memory is coming soon and will replace resource scope. Until then, use the alternatives below.
+[Knowledge](https://mastra.ai/docs/knowledge/overview), now in alpha, gives agents shared information across threads and is planned to replace resource scope. Until it's stable, use the alternatives below.
 
 #### Cross-thread continuity without resource scope
 
@@ -743,7 +743,7 @@ Use thread scope with one or both of these features instead:
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       retrieval: true,
     },
     workingMemory: {
@@ -762,7 +762,7 @@ OM uses token thresholds to decide when to observe and reflect. See [token budge
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       observation: {
         // when to run the Observer (default: 30,000)
         messageTokens: 30_000,
@@ -873,7 +873,7 @@ With `"auto"`, Mastra chooses an idle activation TTL from the active model provi
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       activateAfterIdle: 'auto',
       activateOnProviderChange: true,
     },
@@ -893,7 +893,7 @@ Changing models or providers mid-thread will invalidate the prompt cache. If you
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       activateAfterIdle: { default: 'auto', anthropic: '1h' },
     },
   },
@@ -917,7 +917,7 @@ To disable async buffering and use synchronous observation/reflection instead:
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       observation: {
         bufferTokens: false,
       },
@@ -940,7 +940,7 @@ Set `observation.previousObserverTokens` to limit how many tokens of previous ob
 const memory = new Memory({
   options: {
     observationalMemory: {
-      model: 'google/gemini-2.5-flash',
+      model: 'auto',
       observation: {
         previousObserverTokens: 10_000, // keep only ~10k tokens of recent observations
       },

package/dist/docs/references/docs-memory-overview.md

@@ -9,8 +9,7 @@ Memory gives your agent access to earlier messages and tool results. The agent c
 Mastra agents can be configured to store [message history](https://mastra.ai/docs/memory/message-history). Additionally, you can enable:
 
 - [Observational Memory](https://mastra.ai/docs/memory/observational-memory) (Recommended): Uses background agents to maintain a dense observation log that replaces raw message history as it grows. This keeps the context window small while preserving long-term memory.
-- [Working memory](https://mastra.ai/docs/memory/working-memory): Stores persistent, structured user data such as names, preferences, and goals.
-- [Semantic recall](https://mastra.ai/docs/memory/semantic-recall): Retrieves relevant past messages based on semantic meaning rather than exact keywords.
+- [Working memory](https://mastra.ai/docs/memory/working-memory): Stores persistent, structured information about the user, the application, or the task, such as names, preferences, and goals.
 - [Multi-user threads](https://mastra.ai/docs/memory/multi-user-threads): Share one thread between multiple users.
 
 If the combined memory exceeds the model's context limit, [memory processors](https://mastra.ai/docs/memory/memory-processors) can filter, trim, or prioritize content so the most relevant information is preserved.
@@ -161,7 +160,7 @@ const response = await memoryAgent.generate('Remember my favorite color is blue.
 })
 ```
 
-Scopes can also come from the agent and from the request context. Mastra combines all three sources:
+Scopes can also come from the request context and from the agent. Scopes from the call and the request context arrive first. The agent's `scopes` option then adds a list to them, or a function receives them and returns the run's final set:
 
 ```typescript
 import { Agent } from '@mastra/core/agent'
@@ -170,10 +169,10 @@ import { Memory } from '@mastra/memory'
 export const memoryAgent = new Agent({
   id: 'memory-agent',
   name: 'Memory Agent',
-  scopes: ({ requestContext }) => {
+  scopes: ({ scopes, requestContext }) => {
     // Set by your auth middleware. Request bodies can fill unreserved keys.
     const orgId = requestContext.get('org-id')
-    return orgId ? [`org:${orgId}`] : []
+    return orgId ? [...scopes, `org:${orgId}`] : scopes
   },
   memory: new Memory(),
 })
@@ -181,7 +180,7 @@ export const memoryAgent = new Agent({
 
 The resolver's inputs must come from the server, not the client: a request body can set any request context key that isn't reserved. Trusted server middleware can also set scopes directly with the `MASTRA_SCOPES_KEY` request context key. Visit [Request Context](https://mastra.ai/docs/server/request-context) for details. Clients can't send scopes in a request body.
 
-You can mix scopes with `memory.resource`, `memory.thread`, or the resource and thread request context keys, as long as they agree. If a `resource:` or `thread:` scope names a different value than another source, the call throws an error.
+You can mix scopes with `memory.resource`, `memory.thread`, or the resource and thread request context keys, as long as they agree. If a `resource:` or `thread:` scope names a different value than another source, the call throws an error. A `scopes` function must also return every `resource:` and `thread:` scope it receives. Dropping one throws an error.
 
 Tools, processors, and subagents receive the run's other scopes, but not its `resource:` and `thread:` scopes. Each nested agent call picks its own resource and thread.
 
@@ -189,6 +188,8 @@ Tools, processors, and subagents receive the run's other scopes, but not its `re
 
 Long conversations can fill the context window with raw message history and reduce agent performance. [Observational Memory](https://mastra.ai/docs/memory/observational-memory) uses background agents to compress older messages into observations. This reduces the context used by message history while retaining information for later turns.
 
+Compression loses exact wording, so Observational Memory can register a `recall` tool that goes back to the original messages. The tool is off by default: set `retrieval: true` to enable browsing, which lists earlier messages across the resource's threads unless you pass `scope: 'thread'`. With `retrieval: { vector: true }` plus a vector store and embedder on `Memory`, it can also search by meaning. That search matches observation groups rather than raw messages, so a detail the summary dropped may not turn up, and each hit links back to the messages it came from. See [retrieval mode](https://mastra.ai/docs/memory/observational-memory).
+
 **For AI agents:** Using Observational Memory requires a storage provider! You either need to set it on the Mastra instance at `src/mastra/index.ts` or pass it to the Agent constructor.
 
 ```typescript
@@ -302,4 +303,15 @@ Visit [Request Context](https://mastra.ai/docs/server/request-context) for more
 - [`Memory` reference](https://mastra.ai/reference/memory/memory-class)
 - [Tracing](https://mastra.ai/docs/observability/tracing/overview)
 - [Request Context](https://mastra.ai/docs/server/request-context)
-- [Mastra Code](https://code.mastra.ai/): A coding agent using Mastra's memory system
+- [Mastra Code](https://code.mastra.ai/): A coding agent using Mastra's memory system
+
+**For AI agents:** Observational Memory has a few failure modes that cause real problems.
+
+- Thread scope requires a `threadId` when you call the agent. Without one, Observational Memory throws.
+- On an existing thread, the first turn loads every stored message until the first observation completes. A large thread pays a one-time context cost.
+- `lastMessages` does nothing under Observational Memory. It replaces raw history itself, so the MessageHistory processor is skipped.
+- Leave `observation.observeOnContextOverflow` at its default `true`. When the main model's context overflows, it can activate buffered observations, observe pending messages, and retry the request. It skips read-only requests and only retries when it observed or activated something. With it off, the overflowing request fails.
+- Some invalid options are validated on first use, not when you construct `Memory`. For example, `shareTokenBudget: true` with async buffering enabled fails on the first memory operation. Read the error instead of retrying.
+- With async buffering, work continues in the background. Call `await memory.settled()` before closing storage or shutting down, or that work is lost.
+- The observation mutex only works inside one process. Across processes, coordination comes from a shared PubSub backend and storage: the agent loop uses a backend with leases to run one loop per thread, and the Observer runs inside that loop. Configure every process with the same backend with leases and the same storage. See [Running across processes](https://mastra.ai/docs/signals/running-across-processes).
+- Debug with `OM_DEBUG=1`, which writes `om-debug.log`, plus `getStatus()` and the `onDebugEvent` callback.

package/dist/docs/references/reference-core-getMemory.md

@@ -20,11 +20,11 @@ const thread = await memory.createThread({
 
 ## Parameters
 
-**key** (`TMemoryKey extends keyof TMemory`): The registry key of the memory instance to retrieve. Must match a key used when registering memory in the Mastra constructor.
+**name** (`TMemoryName extends keyof TMemory`): The registry key of the memory instance to retrieve. Must match a key used when registering memory in the Mastra constructor.
 
 ## Returns
 
-**memory** (`TMemory[TMemoryKey]`): The memory instance with the specified key. Throws an error if the memory is not found.
+**memory** (`TMemory[TMemoryName]`): The memory instance with the specified key. Throws an error if the memory is not found.
 
 ## Example: Registering and retrieving memory
 

package/dist/docs/references/reference-core-listMemory.md

@@ -9,7 +9,7 @@ The `.listMemory()` method returns all memory instances registered with the Mast
 ## Usage example
 
 ```typescript
-const memoryInstances = mastra.listMemory()
+const memoryInstances = mastra.listMemory() ?? {}
 
 for (const [key, memory] of Object.entries(memoryInstances)) {
   console.log(`Memory "${key}": ${memory.id}`)
@@ -22,7 +22,7 @@ This method takes no parameters.
 
 ## Returns
 
-**memory** (`Record<string, MastraMemory>`): An object containing all registered memory instances, keyed by their registry keys.
+**memory** (`Record<string, MastraMemory> | undefined`): An object containing all registered memory instances, keyed by their registry keys.
 
 ## Example: Checking registered memory
 
@@ -49,7 +49,7 @@ const mastra = new Mastra({
 })
 
 // List all registered memory instances
-const allMemory = mastra.listMemory()
+const allMemory = mastra.listMemory() ?? {}
 console.log(Object.keys(allMemory)) // ["conversationMemory", "analyticsMemory"]
 ```
 

package/dist/docs/references/reference-memory-observational-memory.md

@@ -43,6 +43,8 @@ OM performs thresholding with fast local token estimation. Text uses `tokenx`, a
 
 **resolveModel** (`(modelId: string, context: { requestContext?: RequestContext }) => MastraModelConfig | Promise<MastraModelConfig>`): Creates the model automatic selection picked, so it can use your credentials and routing. Receives only concrete model IDs, never "auto". When omitted, the picked ID goes through Mastra model routing.
 
+**canUseProvider** (`(providerId: string, context: { requestContext?: RequestContext }) => boolean`): Decides whether "auto" treats a provider as usable. Replaces the built-in GOOGLE_GENERATIVE_AI_API_KEY check that decides whether automatic selection prefers Gemini. Return false for any provider the current request cannot authenticate, for example a provider whose key is not available in the request context. The context.requestContext is the request context of the run making the OM model call, so credentials can be resolved per request instead of from the server environment. See the model row for the selection order.
+
 **scope** (`'resource' | 'thread'`): Memory scope for observations. 'thread' keeps observations per-thread. 'resource' is deprecated and will be removed in a future release. It shares observations across all threads for a resource and works much worse than thread scope for prompt caching and agent understanding. For cross-thread continuity, use thread scope with retrieval or resource-scoped working memory. See resource scope (deprecated). (Default: `'thread'`)
 
 **activateAfterIdle** (`number | string | false | "auto" | Record<string, number | string | false | "auto">`): Time before buffered observations are forced to activate after inactivity, even before observation.messageTokens is reached. Accepts a numeric millisecond value such as 300_000, duration strings like "5m" or "1hr", "auto" for a provider-aware prompt cache TTL, false to disable inherited observation idle activation, or an object of per-provider TTLs such as { default: "auto", anthropic: "1h" }. Object keys match the actor model provider before the first ., case-insensitively, so anthropic covers the direct Anthropic provider but not Claude through another provider such as amazon-bedrock; default covers every other provider. Reflections do not inherit this setting. Use reflection.activateAfterIdle to opt reflections into idle activation.
@@ -71,7 +73,7 @@ OM performs thresholding with fast local token estimation. Text uses `tokenx`, a
 
 **observation.manageWorkingMemory** (`boolean`): Let the Observer manage working memory through OM extraction. Adds WorkingMemoryExtractor, defaults workingMemory.agentManaged to false, and defaults workingMemory.useStateSignals to true. See Working memory updates.
 
-**observation.observeAttachments** (`'auto' | boolean | string[]`): Controls which image/file attachments are forwarded to the Observer model alongside their placeholder text lines. true (default) forwards all attachments. false drops all attachments while keeping placeholders visible. 'auto' uses the provider capabilities registry to decide: attachments are forwarded when the Observer model supports multimodal input, dropped otherwise, and forwarded when no capability data is available for the model. An array is a case-insensitive mimeType allowlist supporting exact matches ('application/pdf'), wildcard subtypes ('image/\*'), and bare '\*' for everything. Useful when the Observer model is text-only (e.g. some DeepSeek endpoints) while the main agent uses a multimodal model. Tool-result attachments are filtered using the same rule.
+**observation.observeAttachments** (`'auto' | boolean | string[]`): Controls which image/file attachments are forwarded to the Observer model alongside their placeholder text lines. By default only images and PDFs are forwarded. true forwards all attachments. false drops all attachments while keeping placeholders visible. 'auto' uses the provider capabilities registry to decide: attachments are forwarded when the Observer model supports multimodal input, dropped otherwise, and forwarded when no capability data is available for the model. An array is a case-insensitive mimeType allowlist supporting exact matches ('application/pdf'), wildcard subtypes ('image/\*'), and bare '\*' for everything. Useful when the Observer model is text-only (e.g. some DeepSeek endpoints) while the main agent uses a multimodal model, or when you only want specific file types forwarded. Tool-result attachments are filtered using the same rule.
 
 **observation.maxRetries** (`number`): Retries after the initial Observer model call on transient provider errors. Governs OM's own retry ladder; the Observer model call itself is configured with no provider-level retries.
 
@@ -244,7 +246,7 @@ export const agent = new Agent({
   memory: new Memory({
     options: {
       observationalMemory: {
-        model: 'google/gemini-2.5-flash',
+        model: 'auto',
         observation: {
           messageTokens: 20_000,
         },
@@ -325,7 +327,7 @@ export const agent = new Agent({
     options: {
       observationalMemory: {
         observation: {
-          model: 'google/gemini-2.5-flash',
+          model: 'auto',
         },
         reflection: {
           model: 'openai/gpt-6-luna',
@@ -352,7 +354,7 @@ export const agent = new Agent({
   memory: new Memory({
     options: {
       observationalMemory: {
-        model: 'google/gemini-2.5-flash',
+        model: 'auto',
         observation: {
           // Focus observations on health-related preferences and goals
           instruction:
@@ -397,7 +399,7 @@ export const agent = new Agent({
   memory: new Memory({
     options: {
       observationalMemory: {
-        model: 'google/gemini-2.5-flash',
+        model: 'auto',
         observation: {
           messageTokens: 30_000,
           // Buffer every 5k tokens (runs in background)
@@ -424,7 +426,7 @@ To disable async buffering entirely:
 
 ```typescript
 observationalMemory: {
-  model: "google/gemini-2.5-flash",
+  model: "auto",
   observation: {
     bufferTokens: false,
   },
@@ -910,7 +912,7 @@ Group IDs in search results include the record that holds them (`groupId@recordI
 new ModelByInputTokens(config)
 ```
 
-Where `config` is an object with `upTo` keys that map token thresholds (numbers) to model targets.
+Where `config` is an object with `upTo` keys that map token thresholds (numbers) to model targets. Each tier needs a concrete model. The example below names specific models instead of `"auto"`.
 
 #### Example
 

package/dist/docs/references/reference-processors-token-limiter-processor.md

@@ -42,7 +42,7 @@ const processor = new TokenLimiterProcessor({
 
 **name** (`string`): Processor display name set to 'Token Limiter'
 
-**processInput** (`(args: { messages: MastraDBMessage[]; abort: (reason?: string) => never }) => Promise<MastraDBMessage[]>`): Trims stored history to fit within the token limit before the agentic loop starts in 'memory-only' trim mode, prioritizing recent messages while preserving system messages and the current turn
+**processInput** (`(args: ProcessInputArgs) => Promise<MessageList>`): Trims stored history to fit within the token limit before the agentic loop starts in 'memory-only' trim mode, prioritizing recent messages while preserving system messages and the current turn
 
 **processInputStep** (`(args: ProcessInputStepArgs) => Promise<void>`): In 'memory-only' trim mode, applies stored-history trimming at each step. In the 'best-fit' and 'contiguous' trim modes, it leaves trimming to processLLMRequest when the agent runs that method for this processor. Otherwise it trims stored history, for example in generateLegacy() and streamLegacy() or when the limiter is inside a processor workflow.
 

package/dist/index.cjs

@@ -1,5 +1,5 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
-const require_src = require("./src-gNoNdZv8.cjs");
+const require_src = require("./src-BbbBzTaj.cjs");
 let _mastra_core_processors = require("@mastra/core/processors");
 exports.AUTO_MODEL_BY_PROVIDER = require_src.AUTO_MODEL_BY_PROVIDER;
 exports.Extractor = require_src.Extractor;

package/dist/index.js

@@ -1 +1 @@
-export { d as AUTO_MODEL_BY_PROVIDER, U as Extractor, w as KnowledgeSemanticIndexCoordinator, t as Memory, n as MessageHistory, g as ModelByInputTokens, k as SUMMARIZE_THREAD_DEFAULTS, r as SemanticRecall, T as StaleKnowledgeSemanticIndexError, E as Subconscious, p as WORKING_MEMORY_STATE_ID, m as WORKING_MEMORY_STATE_PROCESSOR_ID, i as WorkingMemory, D as WorkingMemoryExtractor, h as WorkingMemoryStateProcessor, O as deepMergeWorkingMemory, a as extractWorkingMemoryContent, o as extractWorkingMemoryTags, c as getObservationsAsOf, s as removeWorkingMemoryTags, f as resolveAutoModelId, A as summarizeConversation } from "./src-BncFunCW.js";
+export { d as AUTO_MODEL_BY_PROVIDER, U as Extractor, w as KnowledgeSemanticIndexCoordinator, t as Memory, n as MessageHistory, g as ModelByInputTokens, k as SUMMARIZE_THREAD_DEFAULTS, r as SemanticRecall, T as StaleKnowledgeSemanticIndexError, E as Subconscious, p as WORKING_MEMORY_STATE_ID, m as WORKING_MEMORY_STATE_PROCESSOR_ID, i as WorkingMemory, D as WorkingMemoryExtractor, h as WorkingMemoryStateProcessor, O as deepMergeWorkingMemory, a as extractWorkingMemoryContent, o as extractWorkingMemoryTags, c as getObservationsAsOf, s as removeWorkingMemoryTags, f as resolveAutoModelId, A as summarizeConversation } from "./src-BIae5T_v.js";

package/dist/processors/index.cjs

@@ -1,5 +1,5 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
-const require_src = require("../src-gNoNdZv8.cjs");
+const require_src = require("../src-BbbBzTaj.cjs");
 exports.AUTO_MODEL_BY_PROVIDER = require_src.AUTO_MODEL_BY_PROVIDER;
 exports.Extractor = require_src.Extractor;
 exports.ModelByInputTokens = require_src.ModelByInputTokens;

package/dist/processors/index.js

@@ -1 +1 @@
-export { d as AUTO_MODEL_BY_PROVIDER, U as Extractor, g as ModelByInputTokens, W as OBSERVATIONAL_MEMORY_DEFAULTS, G as OBSERVATION_CONTEXT_INSTRUCTIONS, K as OBSERVATION_CONTEXT_PROMPT, q as OBSERVATION_CONTEXT_PROMPT_THREAD, J as OBSERVATION_CONTINUATION_HINT, M as OBSERVER_SYSTEM_PROMPT, u as ObservationalMemory, l as ObservationalMemoryProcessor, j as TokenCounter, D as WorkingMemoryExtractor, N as buildObserverPrompt, P as buildObserverSystemPrompt, _ as combineObservationGroupRanges, v as deriveObservationGroupProvenance, F as extractCurrentTask, I as formatMessagesForObserver, Y as getObservationContextPrompt, c as getObservationsAsOf, L as hasCurrentTaskSection, B as injectAnchorIds, R as optimizeObservationsForContext, V as parseAnchorId, y as parseObservationGroups, z as parseObserverOutput, b as reconcileObservationGroupsFromReflection, x as renderObservationGroupsForReflection, f as resolveAutoModelId, H as stripEphemeralAnchorIds, S as stripObservationGroups, A as summarizeConversation, C as wrapInObservationGroup } from "../src-BncFunCW.js";
+export { d as AUTO_MODEL_BY_PROVIDER, U as Extractor, g as ModelByInputTokens, W as OBSERVATIONAL_MEMORY_DEFAULTS, G as OBSERVATION_CONTEXT_INSTRUCTIONS, K as OBSERVATION_CONTEXT_PROMPT, q as OBSERVATION_CONTEXT_PROMPT_THREAD, J as OBSERVATION_CONTINUATION_HINT, M as OBSERVER_SYSTEM_PROMPT, u as ObservationalMemory, l as ObservationalMemoryProcessor, j as TokenCounter, D as WorkingMemoryExtractor, N as buildObserverPrompt, P as buildObserverSystemPrompt, _ as combineObservationGroupRanges, v as deriveObservationGroupProvenance, F as extractCurrentTask, I as formatMessagesForObserver, Y as getObservationContextPrompt, c as getObservationsAsOf, L as hasCurrentTaskSection, B as injectAnchorIds, R as optimizeObservationsForContext, V as parseAnchorId, y as parseObservationGroups, z as parseObserverOutput, b as reconcileObservationGroupsFromReflection, x as renderObservationGroupsForReflection, f as resolveAutoModelId, H as stripEphemeralAnchorIds, S as stripObservationGroups, A as summarizeConversation, C as wrapInObservationGroup } from "../src-BIae5T_v.js";