@mastra/memory 1.9.0-alpha.1 → 1.9.0-alpha.2

package/dist/docs/references/docs-agents-networks.md

@@ -1,32 +1,16 @@
 # Agent networks
 
-> **Agent Network Deprecated — Supervisor Pattern Recommended:** Agent networks are deprecated and will be removed in a future release. The [supervisor pattern](https://mastra.ai/docs/agents/supervisor-agents) using `agent.stream()` or `agent.generate()` is now the recommended approach for coordinating multiple agents. It provides the same multi-agent coordination capabilities as `.network()` with significant improvements:
+> **Deprecated — Use supervisor agents:** Agent networks are deprecated and will be removed in a future major release. [Supervisor agents](https://mastra.ai/docs/agents/supervisor-agents) using `agent.stream()` or `agent.generate()` are now the recommended approach. It provides the same multi-agent coordination with better control, a simpler API, and easier debugging.
 >
-> - **Better control**: Iteration hooks, delegation hooks, and task completion scoring give you fine-grained control over execution
-> - **Simpler API**: Uses familiar `stream()` and `generate()` methods instead of a separate `.network()` API
-> - **More flexible**: Stop execution early, modify delegations, filter context, and provide feedback to guide the agent
-> - **Type-safe**: Full TypeScript support for all hooks and callbacks
-> - **Easier debugging**: Monitor progress with `onIterationComplete`, track delegations with `onDelegationStart`/`onDelegationComplete`
->
-> See the [migration guide](https://mastra.ai/guides/migrations/network-to-supervisor) to upgrade from `.network()`.
-
-Agent networks in Mastra coordinate multiple agents, workflows, and tools to handle tasks that aren't clearly defined upfront but can be inferred from the user's message or context. A top-level **routing agent** (a Mastra agent with other agents, workflows, and tools configured) uses an LLM to interpret the request and decide which primitives (subagents, workflows, or tools) to call, in what order, and with what data.
-
-## When to use networks
+> See the [migration guide](https://mastra.ai/guides/migrations/network-to-supervisor) to upgrade.
 
-Use networks for complex tasks that require coordination across multiple primitives. Unlike workflows, which follow a predefined sequence, networks rely on LLM reasoning to interpret the request and decide what to run.
+A **routing agent** uses an LLM to interpret a request and decide which primitives (subagents, workflows, or tools) to call, in what order, and with what data.
 
-## Core principles
+## Create an agent network
 
-Mastra agent networks operate using these principles:
+Configure a routing agent with `agents`, `workflows`, and `tools`. Memory is required as `.network()` uses it to store task history and determine when a task is complete.
 
-- Memory is required when using `.network()` and is used to store task history and determine when a task is complete.
-- Primitives are selected based on their descriptions. Clear, specific descriptions improve routing. For workflows and tools, the input schema helps determine the right inputs at runtime.
-- If multiple primitives have overlapping functionality, the agent favors the more specific one, using a combination of schema and descriptions to decide which to run.
-
-## Creating an agent network
-
-An agent network is built around a top-level routing agent that delegates tasks to subagents, workflows, and tools defined in its configuration. Memory is configured on the routing agent using the `memory` option, and `instructions` define the agent's routing behavior.
+Each primitive needs a clear `description` so the routing agent can decide which to use. For workflows and tools, `inputSchema` and `outputSchema` also help the router determine the right inputs.
 
 ```typescript
 import { Agent } from '@mastra/core/agent'
@@ -35,7 +19,6 @@ import { LibSQLStore } from '@mastra/libsql'
 
 import { researchAgent } from './research-agent'
 import { writingAgent } from './writing-agent'
-
 import { cityWorkflow } from '../workflows/city-workflow'
 import { weatherTool } from '../tools/weather-tool'
 
@@ -43,11 +26,7 @@ export const routingAgent = new Agent({
   id: 'routing-agent',
   name: 'Routing Agent',
   instructions: `
-      You are a network of writers and researchers.
-      The user will ask you to research a topic.
-      Always respond with a complete report—no bullet points.
-      Write in full paragraphs, like a blog post.
-      Do not answer with incomplete or uncertain information.`,
+    You are a network of writers and researchers. The user will ask you to research a topic. Always respond with a complete report—no bullet points. Write in full paragraphs, like a blog post. Do not answer with incomplete or uncertain information.`,
   model: 'openai/gpt-5.4',
   agents: {
     researchAgent,
@@ -68,81 +47,11 @@ export const routingAgent = new Agent({
 })
 ```
 
-### Writing descriptions for network primitives
-
-When configuring a Mastra agent network, each primitive (agent, workflow, or tool) needs a clear description to help the routing agent decide which to use. The routing agent uses each primitive's description and schema to determine what it does and how to use it. Clear descriptions and well-defined input and output schemas improve routing accuracy.
-
-#### Agent descriptions
-
-Each subagent in a network should include a clear `description` that explains what the agent does.
-
-```typescript
-export const researchAgent = new Agent({
-  id: 'research-agent',
-  name: 'Research Agent',
-  description: `This agent gathers concise research insights in bullet-point form.
-    It's designed to extract key facts without generating full
-    responses or narrative content.`,
-})
-```
-
-```typescript
-export const writingAgent = new Agent({
-  id: 'writing-agent',
-  name: 'Writing Agent',
-  description: `This agent turns researched material into well-structured
-    written content. It produces full-paragraph reports with no bullet points,
-    suitable for use in articles, summaries, or blog posts.`,
-})
-```
-
-#### Workflow descriptions
-
-Workflows in a network should include a `description` to explain their purpose, along with `inputSchema` and `outputSchema` to describe the expected data.
-
-```typescript
-export const cityWorkflow = createWorkflow({
-  id: 'city-workflow',
-  description: `This workflow handles city-specific research tasks.
-    It first gathers factual information about the city, then synthesizes
-    that research into a full written report. Use it when the user input
-    includes a city to be researched.`,
-  inputSchema: z.object({
-    city: z.string(),
-  }),
-  outputSchema: z.object({
-    text: z.string(),
-  }),
-})
-```
-
-#### Tool descriptions
-
-Tools in a network should include a `description` to explain their purpose, along with `inputSchema` and `outputSchema` to describe the expected data.
-
-```typescript
-export const weatherTool = createTool({
-  id: 'weather-tool',
-  description: ` Retrieves current weather information using the wttr.in API.
-    Accepts a city or location name as input and returns a short weather summary.
-    Use this tool whenever up-to-date weather data is requested.
-  `,
-  inputSchema: z.object({
-    location: z.string(),
-  }),
-  outputSchema: z.object({
-    weather: z.string(),
-  }),
-})
-```
-
-## Calling agent networks
-
-Call a Mastra agent network using `.network()` with a user message. The method returns a stream of events that you can iterate over to track execution progress and retrieve the final result.
+> **Note:** Subagents need a `description` on the `Agent` instance. Workflows and tools need a `description` plus `inputSchema` and `outputSchema` on `createWorkflow()` or `createTool()`.
 
-### Agent example
+## Call the network
 
-In this example, the network interprets the message and would route the request to both the `researchAgent` and `writingAgent` to generate a complete response.
+Call `.network()` with a user message. The method returns a stream of events you can iterate over.
 
 ```typescript
 const result = await routingAgent.network('Tell me three cool ways to use Mastra')
@@ -155,145 +64,119 @@ for await (const chunk of result) {
 }
 ```
 
-#### Agent output
-
-The following `chunk.type` events are emitted during this request:
-
-```text
-routing-agent-start
-routing-agent-end
-agent-execution-start
-agent-execution-event-start
-agent-execution-event-step-start
-agent-execution-event-text-start
-agent-execution-event-text-delta
-agent-execution-event-text-end
-agent-execution-event-step-finish
-agent-execution-event-finish
-agent-execution-end
-network-execution-event-step-finish
-```
-
-## Workflow example
+## Structured output
 
-In this example, the routing agent recognizes the city name in the message and runs the `cityWorkflow`. The workflow defines steps that call the `researchAgent` to gather facts, then the `writingAgent` to generate the final text.
+Pass `structuredOutput` to get typed, validated results. Use `objectStream` for partial objects as they generate.
 
 ```typescript
-const result = await routingAgent.network('Tell me some historical facts about London')
+import { z } from 'zod'
 
-for await (const chunk of result) {
-  console.log(chunk.type)
-  if (chunk.type === 'network-execution-event-step-finish') {
-    console.log(chunk.payload.result)
-  }
-}
-```
+const resultSchema = z.object({
+  summary: z.string().describe('A brief summary of the findings'),
+  recommendations: z.array(z.string()).describe('List of recommendations'),
+  confidence: z.number().min(0).max(1).describe('Confidence score'),
+})
 
-### Workflow output
+const stream = await routingAgent.network('Research AI trends', {
+  structuredOutput: { schema: resultSchema },
+})
 
-The following `chunk.type` events are emitted during this request:
+for await (const partial of stream.objectStream) {
+  console.log('Building result:', partial)
+}
 
-```text
-routing-agent-end
-workflow-execution-start
-workflow-execution-event-workflow-start
-workflow-execution-event-workflow-step-start
-workflow-execution-event-workflow-step-result
-workflow-execution-event-workflow-finish
-workflow-execution-end
-routing-agent-start
-network-execution-event-step-finish
+const final = await stream.object
+console.log(final?.summary)
 ```
 
-### Tool example
+## Approve and decline tool calls
+
+When a primitive requires approval, the stream emits an `agent-execution-approval` or `tool-execution-approval` chunk. Use `approveNetworkToolCall()` or `declineNetworkToolCall()` to respond.
 
-In this example, the routing agent skips the `researchAgent`, `writingAgent`, and `cityWorkflow`, and calls the `weatherTool` directly to complete the task.
+Network approval uses snapshots to capture execution state. Ensure a [storage provider](https://mastra.ai/docs/memory/storage) is enabled in your Mastra instance.
 
 ```typescript
-const result = await routingAgent.network("What's the weather in London?")
+const stream = await routingAgent.network('Perform some sensitive action', {
+  memory: {
+    thread: 'user-123',
+    resource: 'my-app',
+  },
+})
 
-for await (const chunk of result) {
-  console.log(chunk.type)
-  if (chunk.type === 'network-execution-event-step-finish') {
-    console.log(chunk.payload.result)
+for await (const chunk of stream) {
+  if (chunk.type === 'agent-execution-approval' || chunk.type === 'tool-execution-approval') {
+    // Approve
+    const approvedStream = await routingAgent.approveNetworkToolCall(chunk.payload.toolCallId, {
+      runId: stream.runId,
+      memory: { thread: 'user-123', resource: 'my-app' },
+    })
+
+    for await (const c of approvedStream) {
+      if (c.type === 'network-execution-event-step-finish') {
+        console.log(c.payload.result)
+      }
+    }
   }
 }
 ```
 
-#### Tool output
+To decline instead, call `declineNetworkToolCall()` with the same arguments.
 
-The following `chunk.type` events are emitted during this request:
+## Suspend and resume
 
-```text
-routing-agent-start
-routing-agent-end
-tool-execution-start
-tool-execution-end
-network-execution-event-step-finish
-```
-
-## Structured output
-
-When you need typed, validated results from a network, use the `structuredOutput` option. After the network completes its task, it generates a structured response matching your schema.
+When a primitive calls `suspend()`, the stream emits a suspension chunk (e.g., `tool-execution-suspended`). Use `resumeNetwork()` to provide the requested data and continue execution.
 
 ```typescript
-import { z } from 'zod'
-
-const resultSchema = z.object({
-  summary: z.string().describe('A brief summary of the findings'),
-  recommendations: z.array(z.string()).describe('List of recommendations'),
-  confidence: z.number().min(0).max(1).describe('Confidence score'),
-})
-
-const stream = await routingAgent.network('Research AI trends', {
-  structuredOutput: {
-    schema: resultSchema,
-  },
+const stream = await routingAgent.network('Delete the old records', {
+  memory: { thread: 'user-123', resource: 'my-app' },
 })
 
-// Consume the stream
 for await (const chunk of stream) {
-  if (chunk.type === 'network-object') {
-    // Partial object during generation
-    console.log('Partial:', chunk.payload.object)
-  }
-  if (chunk.type === 'network-object-result') {
-    // Final structured object
-    console.log('Final:', chunk.payload.object)
+  if (chunk.type === 'workflow-execution-suspended') {
+    console.log(chunk.payload.suspendPayload)
   }
 }
 
-// Get the typed result
-const result = await stream.object
-console.log(result?.summary)
-console.log(result?.recommendations)
-console.log(result?.confidence)
+// Resume with user confirmation
+const resumedStream = await routingAgent.resumeNetwork(
+  { confirmed: true },
+  {
+    runId: stream.runId,
+    memory: { thread: 'user-123', resource: 'my-app' },
+  },
+)
+
+for await (const chunk of resumedStream) {
+  if (chunk.type === 'network-execution-event-step-finish') {
+    console.log(chunk.payload.result)
+  }
+}
 ```
 
-### Streaming partial objects
+### Automatic resumption
 
-For real-time updates during structured output generation, use `objectStream`:
+Set `autoResumeSuspendedTools` to `true` so the network resumes suspended primitives based on the user's next message. This creates a conversational flow where users provide the required information naturally.
 
 ```typescript
-const stream = await routingAgent.network('Analyze market data', {
-  structuredOutput: { schema: resultSchema },
+const stream = await routingAgent.network('Delete the old records', {
+  autoResumeSuspendedTools: true,
+  memory: { thread: 'user-123', resource: 'my-app' },
 })
+```
 
-// Stream partial objects as they're generated
-for await (const partial of stream.objectStream) {
-  console.log('Building result:', partial)
-}
+Requirements for automatic resumption:
 
-// Get the final typed result
-const final = await stream.object
-```
+- **Memory configured**: The agent needs memory to track suspended tools across messages.
+- **Same thread**: The follow-up message must use the same `thread` and `resource` identifiers.
+- **`resumeSchema` defined**: The tool must define a `resumeSchema` so the network can extract data from the user's message.
+
+|          | Manual (`resumeNetwork`)                       | Automatic (`autoResumeSuspendedTools`)    |
+| -------- | ---------------------------------------------- | ----------------------------------------- |
+| Best for | Custom UIs with approval buttons               | Chat-style interfaces                     |
+| Control  | Full control over resume timing and data       | Network extracts data from user's message |
+| Setup    | Handle suspension chunks, call `resumeNetwork` | Set flag, define `resumeSchema` on tools  |
 
 ## Related
 
-- [Supervisor Agents](https://mastra.ai/docs/agents/supervisor-agents)
-- [Migration: .network() to Supervisor Pattern](https://mastra.ai/guides/migrations/network-to-supervisor)
-- [Guide: Research Coordinator](https://mastra.ai/guides/guide/research-coordinator)
-- [Agent Memory](https://mastra.ai/docs/agents/agent-memory)
-- [Agent Approval](https://mastra.ai/docs/agents/agent-approval)
-- [Workflows Overview](https://mastra.ai/docs/workflows/overview)
-- [Request Context](https://mastra.ai/docs/server/request-context)
+- [Supervisor agents](https://mastra.ai/docs/agents/supervisor-agents)
+- [Migration: `.network()` to supervisor agents](https://mastra.ai/guides/migrations/network-to-supervisor)

package/dist/docs/references/docs-agents-supervisor-agents.md

@@ -1,6 +1,8 @@
 # Supervisor agents
 
-A supervisor agent coordinates multiple subagents using `agent.stream()` or `agent.generate()`. You configure subagents on the supervisor's `agents` property, and the supervisor uses its instructions and each subagent's `description` to decide when and how to delegate tasks.
+**Added in:** `@mastra/core@1.8.0`
+
+A supervisor agent coordinates multiple subagents using [`Agent.stream()`](https://mastra.ai/reference/streaming/agents/stream) or [`Agent.generate()`](https://mastra.ai/reference/agents/generate). You configure subagents on the supervisor's `agents` property, and the supervisor uses its instructions and each subagent's `description` to decide when and how to delegate tasks.
 
 ## When to use supervisor agents
 
@@ -12,7 +14,9 @@ Common use cases:
 - Multi-step tasks that need different expertise at each stage
 - Tasks where you need fine-grained control over delegation behavior
 
-## Quick start
+> **Note:** Supervisor agents are one approach to building multi-agent systems in Mastra. For other patterns, read the [conceptual overview](https://mastra.ai/guides/concepts/multi-agent-systems).
+
+## Quickstart
 
 Define subagents with clear descriptions, then create a supervisor agent that references them:
 
@@ -61,10 +65,10 @@ Delegation hooks let you intercept, modify, or reject delegations as they happen
 
 Called before the supervisor delegates to a subagent. Return an object to control the delegation:
 
-- `proceed: true` — allow the delegation (default behavior)
-- `proceed: false` — reject the delegation with a `rejectionReason`
-- `modifiedPrompt` — rewrite the prompt sent to the subagent
-- `modifiedMaxSteps` — limit the subagent's iteration count
+- `proceed: true`: Allow the delegation (default behavior)
+- `proceed: false`: Reject the delegation with a `rejectionReason`
+- `modifiedPrompt`: Rewrite the prompt sent to the subagent
+- `modifiedMaxSteps`: Limit the subagent's iteration count
 
 ```typescript
 const stream = await supervisor.stream('Research AI trends', {
@@ -108,8 +112,8 @@ The `context` object includes:
 
 Called after a delegation finishes. Use it to inspect results, provide feedback, or stop execution:
 
-- `context.bail()` — stop the supervisor loop immediately
-- Return `{ feedback: '...' }` — add feedback that gets saved to the supervisor's memory and is visible to subsequent iterations
+- `context.bail()`: Stop the supervisor loop immediately
+- Return `{ feedback: '...' }`: Add feedback that gets saved to the supervisor's memory and is visible to subsequent iterations
 
 ```typescript
 const stream = await supervisor.stream('Research AI trends', {
@@ -196,16 +200,18 @@ Return `{ continue: true }` to keep iterating, or `{ continue: false }` to stop.
 
 ## Memory isolation
 
-The supervisor pattern implements memory isolation — subagents receive the full conversation context for better decision-making, but only their specific delegation prompt and response are saved to their memory.
+Supervisor agents implement memory isolation. Subagents receive the full conversation context for better decision-making, but only their specific delegation prompt and response are saved to their memory.
 
 How it works:
 
-1. **Full context forwarded** — When the supervisor delegates, the subagent receives all messages from the supervisor's conversation
-2. **Scoped memory saves** — Only the delegation prompt and the subagent's response are saved to the subagent's memory
-3. **Fresh thread per invocation** — Each delegation uses a unique thread ID, ensuring clean separation
+1. **Full context forwarded**: When the supervisor delegates, the subagent receives all messages from the supervisor's conversation
+2. **Scoped memory saves**: Only the delegation prompt and the subagent's response are saved to the subagent's memory
+3. **Fresh thread per invocation**: Each delegation uses a unique thread ID, ensuring clean separation
 
 This ensures subagents have the context they need without cluttering their memory with the entire supervisor conversation.
 
+> **Note:** Visit [memory in multi-agent systems](https://mastra.ai/docs/memory/overview) for more details.
+
 ## Tool approval propagation
 
 Tool approvals propagate through the delegation chain. When a subagent uses a tool with `requireApproval: true` or calls `suspend()`, the approval request surfaces to the supervisor level.
@@ -296,9 +302,9 @@ Success criteria:
 
 ## Related
 
-- [Agent Networks](https://mastra.ai/docs/agents/networks)
-- [Migration: .network() to Supervisor Pattern](https://mastra.ai/guides/migrations/network-to-supervisor)
-- [Guide: Research Coordinator](https://mastra.ai/guides/guide/research-coordinator)
-- [Agent.stream() Reference](https://mastra.ai/reference/streaming/agents/stream)
-- [Agent.generate() Reference](https://mastra.ai/reference/agents/generate)
-- [Agent Approval](https://mastra.ai/docs/agents/agent-approval)
+- [Guide: Research coordinator](https://mastra.ai/guides/guide/research-coordinator)
+- [Agent.stream() reference](https://mastra.ai/reference/streaming/agents/stream)
+- [Agent.generate() reference](https://mastra.ai/reference/agents/generate)
+- [Agent approval](https://mastra.ai/docs/agents/agent-approval)
+- [Memory in multi-agent systems](https://mastra.ai/docs/memory/overview)
+- [Concept: Multi-agent systems](https://mastra.ai/guides/concepts/multi-agent-systems)

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

@@ -1,10 +1,10 @@
-# Observational memory
+# Observational Memory
 
 **Added in:** `@mastra/memory@1.1.0`
 
 Observational Memory (OM) is Mastra's memory system for long-context agentic memory. Two background agents — an **Observer** and a **Reflector** — watch your agent's conversations and maintain a dense observation log that replaces raw message history as it grows.
 
-## Quick start
+## Quickstart
 
 Enable `observationalMemory` in the memory options when creating your agent:
 
@@ -89,6 +89,34 @@ The result is a three-tier system:
 2. **Observations**: A log of what the Observer has seen
 3. **Reflections**: Condensed observations when memory becomes too long
 
+### Retrieval mode (experimental)
+
+> **Note:** Retrieval mode is experimental. The API may change in future releases.
+
+Normal OM compresses messages into observations, which is great for staying on task — but the original wording is gone. Retrieval mode fixes this by keeping each observation group linked to the raw messages that produced it. When the agent needs exact wording, tool output, or chronology that the summary compressed away, it can call a `recall` tool to page through the source messages.
+
+```typescript
+const memory = new Memory({
+  options: {
+    observationalMemory: {
+      model: 'google/gemini-2.5-flash',
+      scope: 'thread',
+      retrieval: true,
+    },
+  },
+})
+```
+
+With retrieval mode enabled, OM:
+
+- Stores a `range` (e.g. `startId:endId`) on each observation group pointing to the messages it was derived from
+- Keeps range metadata visible in the agent's context so the agent knows which observations map to which messages
+- Registers a `recall` tool the agent can call to page through the raw messages behind any range
+
+Retrieval mode is only active for thread-scoped OM. Setting `retrieval: true` with `scope: 'resource'` has no effect — OM keeps resource-scoped behavior but skips retrieval-mode context and does not register the `recall` tool.
+
+See the [recall tool reference](https://mastra.ai/reference/memory/observational-memory) for the full API (detail levels, part indexing, pagination, and token limiting).
+
 ## Models
 
 The Observer and Reflector run in the background. Any model that works with Mastra's [model routing](https://mastra.ai/models) (`provider/model`) can be used. When using `observationalMemory: true`, the default model is `google/gemini-2.5-flash`. When passing a config object, a `model` must be explicitly set.