@soat/sdk 0.77.0 → 0.78.0

package/dist/index.cjs

@@ -994,7 +994,7 @@ var AiProviders = class {
 	*
 	* Deletes an AI provider configuration.
 	*
-	* Live references — chats, agents, and model routes whose targets name this provider — always block deletion with `409 AI_PROVIDER_HAS_DEPENDENTS`; `force` does not override them, so delete or repoint those resources first. Soft dependents — price overrides and usage/generation records — also block with `409` unless `force=true`, which deletes the provider's price overrides and unlinks (nulls) its usage history, preserving those rows. The `409` body's `error.meta` reports the counts, a sample of offending IDs, and a `forcible` flag that is `true` when a `force=true` retry would succeed.
+	* Live references — chats, agents, and model routes whose targets name this provider — always block deletion with `409 AI_PROVIDER_HAS_DEPENDENTS`; `force` does not override them, so delete or repoint those resources first. Soft dependents — price overrides and usage/generation records — also block with `409` unless `force=true`, which deletes the provider's price overrides and unlinks its usage history, preserving those rows: each usage event keeps the provider's id as its `ai_provider_id`. The `409` body's `error.meta` reports the counts, a sample of offending IDs, and a `forcible` flag that is `true` when a `force=true` retry would succeed.
 	*
 	*/
 	static deleteAiProvider(options) {
@@ -4753,7 +4753,7 @@ var Usage = class {
 	*
 	* Returns the raw usage events the caller can access, most recent first, narrowed by any combination of the same filters the aggregate takes — agent, generation, trace, actor, session, AI provider, orchestration run, orchestration, tool, trigger, action, outcome, meter type, model or `source` — so a rollup and the events behind it are addressed the same way. Each event is the per-generation token usage as reported by the provider, for audit and reconciliation.
 	*
-	* An id naming nothing in scope yields an empty page rather than dropping the filter, so a mistyped narrowing can never widen the list past what was asked for.
+	* An id filter matches the id the event recorded, so it still selects the spend of an agent, provider, session or tool since deleted. An id no event in scope recorded yields an empty page rather than dropping the filter, so a mistyped narrowing can never widen the list past what was asked for.
 	*
 	*/
 	static listUsageEvents(options) {
@@ -4767,7 +4767,7 @@ var Usage = class {
 	*
 	* Returns a project's usage rolled up over an optional `[from, to]` time window, optionally bucketed by a single dimension and narrowed by any combination of sixteen filters — session, actor, agent, AI provider, orchestration run, orchestration, generation, trace, tool, publisher project, meter type, model, source, trigger, action and outcome. That is how "what did this conversation cost", "what has this end user cost" and "what did this agent cost" are answered. Each group and the grand total carry an `event_count`, summed token counts, a measured `quantity` per component (so infra meters report their real amount, not zeros), and `cost_usd` (null when no event in the bucket was priced). This is the per-project cost-by-range/by-category query — a monthly figure without scanning raw usage events client-side.
 	*
-	* Narrowings intersect and apply to the whole rollup — every bucket, `totals` and `totals.distinct` alike — so each composes with any `group_by`, and all sixteen are echoed back under `filters`. The ten naming a resource are resolved against the project (a tool also through an accepted share), and one naming nothing there empties the rollup rather than being dropped: a mistyped id must never read back as the project's whole spend. The other six are matched against the value the event recorded.
+	* Narrowings intersect and apply to the whole rollup — every bucket, `totals` and `totals.distinct` alike — so each composes with any `group_by`, and all sixteen are echoed back under `filters`. The ten naming a resource match the id the event recorded, so a deleted resource's spend stays selectable; an id no event recorded empties the rollup rather than being dropped: a mistyped id must never read back as the project's whole spend. `orchestration_id` is the exception — the event records no orchestration, so a deleted one empties the rollup. The other six are matched against the value the event recorded.
 	*
 	* `groups` is paginated. Its `total` is the number of distinct buckets in the window — bucket cardinality, including a null bucket for the events the dimension does not apply to, and never a count of entities. To count entities, send `include=distinct` and read `totals.distinct`. `totals` always describes the whole window — never the page above it, and never the bucketing: omit `group_by` and `totals` is unchanged while `groups` is empty.
 	*

package/dist/index.d.cts

@@ -526,6 +526,16 @@ type Agent = {
    * Maximum number of recent messages to include in the context window sent to the model. When null, all messages are included.
    */
   max_context_messages?: number | null;
+  /**
+   * Maximum number of tokens the model may generate per step of a turn. A response that reaches it stops with `stop_reason` `length`. Null leaves the model's default.
+   */
+  max_output_tokens?: number | null;
+  /**
+   * How much the model reasons before answering. Translated per provider (Anthropic thinking, Gemini thinking config, OpenAI reasoning effort) against what the model supports; a model that cannot reason ignores it. `none` turns reasoning off where the model allows it. Reasoning tokens are metered as output. Null leaves the model's default.
+   */
+  reasoning?: {
+    effort: 'none' | 'low' | 'medium' | 'high';
+  } | null;
   /**
    * When true, only one open session per actor_id is allowed for this agent. Creating a second open session for the same actor returns 409.
    */
@@ -587,7 +597,7 @@ type AgentVersion = {
    */
   version?: number;
   /**
-   * The agent's configuration as it stood at this version: every mutable field of the `Agent` schema (`instructions`, `model`, `tool_bindings`, `max_steps`, `tool_choice`, `stop_conditions`, `active_tool_ids`, `step_rules`, `boundary_policy`, `temperature`, `knowledge_config`, `output_schema`, `max_context_messages`, `single_session_per_actor`, `on_approval_expiry`, `guardrail_ids`, `ai_provider_id`, `model_route_id`, `name`), and none of its identity or bookkeeping fields (`id`, `project_id`, `version`, `active_release`, timestamps).
+   * The agent's configuration as it stood at this version: every mutable field of the `Agent` schema (`instructions`, `model`, `tool_bindings`, `max_steps`, `tool_choice`, `stop_conditions`, `active_tool_ids`, `step_rules`, `boundary_policy`, `temperature`, `knowledge_config`, `output_schema`, `max_context_messages`, `max_output_tokens`, `reasoning`, `single_session_per_actor`, `on_approval_expiry`, `guardrail_ids`, `ai_provider_id`, `model_route_id`, `name`), and none of its identity or bookkeeping fields (`id`, `project_id`, `version`, `active_release`, timestamps).
    *
    * Deliberately open rather than a fixed schema: an archive written by an earlier release of SOAT reflects the agent surface **of its own time**, so it may carry fields the current schema no longer defines, or lack ones it has since gained. Knowledge retrieval is not part of the snapshot — a version records which `knowledge_config` applied, while the documents and memory stores it resolves keep their own histories and are pinned at generation time.
    */
@@ -764,6 +774,16 @@ type CreateAgentRequest = {
    * Maximum number of recent messages included in the context window. Null means no limit.
    */
   max_context_messages?: number;
+  /**
+   * Maximum number of tokens the model may generate per step of a turn. A response that reaches it stops with `stop_reason` `length`. Null leaves the model's default.
+   */
+  max_output_tokens?: number | null;
+  /**
+   * How much the model reasons before answering. Translated per provider (Anthropic thinking, Gemini thinking config, OpenAI reasoning effort) against what the model supports; a model that cannot reason ignores it. `none` turns reasoning off where the model allows it. Reasoning tokens are metered as output. Null leaves the model's default.
+   */
+  reasoning?: {
+    effort: 'none' | 'low' | 'medium' | 'high';
+  } | null;
   /**
    * When true, only one open session per actor_id is allowed for this agent.
    */
@@ -899,6 +919,16 @@ type UpdateAgentRequest = {
    * Maximum number of recent messages included in the context window. Null means no limit.
    */
   max_context_messages?: number | null;
+  /**
+   * Maximum number of tokens the model may generate per step of a turn. A response that reaches it stops with `stop_reason` `length`. Null leaves the model's default.
+   */
+  max_output_tokens?: number | null;
+  /**
+   * How much the model reasons before answering. Translated per provider (Anthropic thinking, Gemini thinking config, OpenAI reasoning effort) against what the model supports; a model that cannot reason ignores it. `none` turns reasoning off where the model allows it. Reasoning tokens are metered as output. Null leaves the model's default.
+   */
+  reasoning?: {
+    effort: 'none' | 'low' | 'medium' | 'high';
+  } | null;
   /**
    * When true, only one open session per actor_id is allowed for this agent.
    */
@@ -2942,6 +2972,16 @@ type AgentResourceProperties = {
    * Maximum number of recent messages to include in the context window sent to the model. When null, all messages are included.
    */
   max_context_messages?: number | null;
+  /**
+   * Maximum number of tokens the model may generate per step of a turn. A response that reaches it stops with `stop_reason` `length`. Null leaves the model's default.
+   */
+  max_output_tokens?: number | null;
+  /**
+   * How much the model reasons before answering. Translated per provider (Anthropic thinking, Gemini thinking config, OpenAI reasoning effort) against what the model supports; a model that cannot reason ignores it. `none` turns reasoning off where the model allows it. Reasoning tokens are metered as output. Null leaves the model's default.
+   */
+  reasoning?: {
+    effort: 'none' | 'low' | 'medium' | 'high';
+  } | null;
   /**
    * When true, only one open session per actor_id is allowed for this agent.
    */
@@ -3671,6 +3711,10 @@ type SessionResourceProperties = {
    * Number of seconds of inactivity after which the session expires. 0 means never expires.
    */
   inactivity_ttl_seconds?: number;
+  /**
+   * Maximum number of recent conversation messages sent to the model on each generation, taking precedence over the agent's. null defers to the agent.
+   */
+  max_context_messages?: number | null;
   /**
    * Optional context object passed to tool calls. Write-only: no read of a session returns it.
    */
@@ -6251,6 +6295,11 @@ type SessionRecord = {
    *
    */
   message_delay_seconds?: number | null;
+  /**
+   * Maximum number of recent conversation messages sent to the model on each generation. Takes precedence over the agent's `max_context_messages`; null defers to it. A fork keeps its parent's value.
+   *
+   */
+  max_context_messages?: number | null;
   /**
    * Timestamp of the last activity on the session (message added or response generated).
    */
@@ -6331,6 +6380,11 @@ type CreateSessionRequest = {
    *
    */
   message_delay_seconds?: number | null;
+  /**
+   * Maximum number of recent conversation messages sent to the model on each generation, taking precedence over the agent's `max_context_messages`. null or absent defers to the agent.
+   *
+   */
+  max_context_messages?: number | null;
 };
 type UpdateSessionRequest = {
   /**
@@ -6361,6 +6415,11 @@ type UpdateSessionRequest = {
    *
    */
   message_delay_seconds?: number | null;
+  /**
+   * Maximum number of recent conversation messages sent to the model on each generation, taking precedence over the agent's `max_context_messages`. Set to null to defer to the agent.
+   *
+   */
+  max_context_messages?: number | null;
 };
 type AddSessionMessageRequest = {
   /**
@@ -7415,7 +7474,7 @@ type UsageEvent = {
    */
   session_id?: string | null;
   /**
-   * Public ID of the AI provider instance billed — the target a model route picked for the turn, or the agent's pinned provider. Null if the provider was later deleted; the `provider`/`model` snapshot still records what was billed.
+   * Public ID of the AI provider instance billed — the target a model route picked for the turn, or the agent's pinned provider. Kept after the provider is deleted, like every attribution id on the event.
    *
    */
   ai_provider_id?: string | null;
@@ -7430,7 +7489,7 @@ type UsageEvent = {
    */
   action_id?: string | null;
   /**
-   * Public ID of the tool a `tool_execution` event metered. Null on every other meter, for an inline tool definition, and once the tool is deleted.
+   * Public ID of the tool a `tool_execution` event metered. Null on every other meter and for an inline tool definition.
    *
    */
   tool_id?: string | null;
@@ -7440,12 +7499,12 @@ type UsageEvent = {
    */
   publisher_project_id?: string | null;
   /**
-   * Public ID of the document an `embedding` event embedded a chunk of — ingestion, a re-chunk, a conversation's turns. Null on a query embedding, on every other meter, and once the document is deleted.
+   * Public ID of the document an `embedding` event embedded a chunk of — ingestion, a re-chunk, a conversation's turns. Null on a query embedding and on every other meter.
    *
    */
   document_id?: string | null;
   /**
-   * Public ID of the memory store an `embedding` event embedded written content for. The store rather than the memory: stored content is shared by every memory in the store that states it. Null on a query embedding, on every other meter, and once the store is deleted.
+   * Public ID of the memory store an `embedding` event embedded written content for. The store rather than the memory: stored content is shared by every memory in the store that states it. Null on a query embedding and on every other meter.
    *
    */
   memory_store_id?: string | null;
@@ -20752,22 +20811,22 @@ type ListUsageEventsData = {
      */
     trace_id?: string;
     /**
-     * Filter by the actor (end user) the usage is attributed to. An actor that does not exist in scope yields an empty page.
+     * Filter by the actor (end user) the usage is attributed to.
      *
      */
     actor_id?: string;
     /**
-     * Filter by the session the usage was produced in. A session that does not exist in scope yields an empty page.
+     * Filter by the session the usage was produced in.
      *
      */
     session_id?: string;
     /**
-     * Filter by the provider record the spend was billed against. A provider that does not exist in scope yields an empty page.
+     * Filter by the provider record the spend was billed against.
      *
      */
     ai_provider_id?: string;
     /**
-     * Filter by the orchestration run that metered the event. A run that does not exist in scope yields an empty page.
+     * Filter by the orchestration run that metered the event.
      *
      */
     orchestration_run_id?: string;
@@ -20790,12 +20849,12 @@ type ListUsageEventsData = {
      */
     action_id?: string;
     /**
-     * Filter by the tool a `tool_execution` event metered, including a tool another project shares with one in scope. A tool that does not exist in scope yields an empty page.
+     * Filter by the tool a `tool_execution` event metered, including a tool another project shares with one in scope.
      *
      */
     tool_id?: string;
     /**
-     * Filter to the calls made through a share of a resource the named project owns. A project that does not exist yields an empty page.
+     * Filter to the calls made through a share of a resource the named project owns.
      *
      */
     publisher_project_id?: string;
@@ -20886,7 +20945,7 @@ type GetUsageAggregateData = {
      */
     meter_type?: string;
     /**
-     * Narrow the whole rollup — every bucket, `totals` and `totals.distinct` alike — to one session's traffic. Combines with `group_by`: `group_by=day&session_id=…` is a conversation's spend per day, `group_by=model` is the same spend split by model. An id naming no session in this project yields an empty rollup, never the project total.
+     * Narrow the whole rollup — every bucket, `totals` and `totals.distinct` alike — to one session's traffic. Combines with `group_by`: `group_by=day&session_id=…` is a conversation's spend per day, `group_by=model` is the same spend split by model. Like every id narrowing it matches the id the event recorded, so a deleted session's spend stays selectable; an id no event in this project recorded yields an empty rollup, never the project total.
      *
      */
     session_id?: string;
@@ -20944,12 +21003,12 @@ type GetUsageAggregateData = {
      */
     action_id?: string;
     /**
-     * Narrow to the executions of one tool, including one another project shares with this one. An id naming no tool in reach yields an empty rollup.
+     * Narrow to the executions of one tool, including one another project shares with this one.
      *
      */
     tool_id?: string;
     /**
-     * Narrow to the calls made through a share of a resource the named project owns. A project that does not exist yields an empty rollup.
+     * Narrow to the calls made through a share of a resource the named project owns.
      *
      */
     publisher_project_id?: string;
@@ -22237,7 +22296,7 @@ export declare class AiProviders {
    *
    * Deletes an AI provider configuration.
    *
-   * Live references — chats, agents, and model routes whose targets name this provider — always block deletion with `409 AI_PROVIDER_HAS_DEPENDENTS`; `force` does not override them, so delete or repoint those resources first. Soft dependents — price overrides and usage/generation records — also block with `409` unless `force=true`, which deletes the provider's price overrides and unlinks (nulls) its usage history, preserving those rows. The `409` body's `error.meta` reports the counts, a sample of offending IDs, and a `forcible` flag that is `true` when a `force=true` retry would succeed.
+   * Live references — chats, agents, and model routes whose targets name this provider — always block deletion with `409 AI_PROVIDER_HAS_DEPENDENTS`; `force` does not override them, so delete or repoint those resources first. Soft dependents — price overrides and usage/generation records — also block with `409` unless `force=true`, which deletes the provider's price overrides and unlinks its usage history, preserving those rows: each usage event keeps the provider's id as its `ai_provider_id`. The `409` body's `error.meta` reports the counts, a sample of offending IDs, and a `forcible` flag that is `true` when a `force=true` retry would succeed.
    *
    */
   static deleteAiProvider<ThrowOnError extends boolean = false>(options: Options<DeleteAiProviderData, ThrowOnError>): RequestResult<DeleteAiProviderResponses, DeleteAiProviderErrors, ThrowOnError>;
@@ -24163,7 +24222,7 @@ export declare class Usage {
    *
    * Returns the raw usage events the caller can access, most recent first, narrowed by any combination of the same filters the aggregate takes — agent, generation, trace, actor, session, AI provider, orchestration run, orchestration, tool, trigger, action, outcome, meter type, model or `source` — so a rollup and the events behind it are addressed the same way. Each event is the per-generation token usage as reported by the provider, for audit and reconciliation.
    *
-   * An id naming nothing in scope yields an empty page rather than dropping the filter, so a mistyped narrowing can never widen the list past what was asked for.
+   * An id filter matches the id the event recorded, so it still selects the spend of an agent, provider, session or tool since deleted. An id no event in scope recorded yields an empty page rather than dropping the filter, so a mistyped narrowing can never widen the list past what was asked for.
    *
    */
   static listUsageEvents<ThrowOnError extends boolean = false>(options?: Options<ListUsageEventsData, ThrowOnError>): RequestResult<ListUsageEventsResponses, ListUsageEventsErrors, ThrowOnError>;
@@ -24172,7 +24231,7 @@ export declare class Usage {
    *
    * Returns a project's usage rolled up over an optional `[from, to]` time window, optionally bucketed by a single dimension and narrowed by any combination of sixteen filters — session, actor, agent, AI provider, orchestration run, orchestration, generation, trace, tool, publisher project, meter type, model, source, trigger, action and outcome. That is how "what did this conversation cost", "what has this end user cost" and "what did this agent cost" are answered. Each group and the grand total carry an `event_count`, summed token counts, a measured `quantity` per component (so infra meters report their real amount, not zeros), and `cost_usd` (null when no event in the bucket was priced). This is the per-project cost-by-range/by-category query — a monthly figure without scanning raw usage events client-side.
    *
-   * Narrowings intersect and apply to the whole rollup — every bucket, `totals` and `totals.distinct` alike — so each composes with any `group_by`, and all sixteen are echoed back under `filters`. The ten naming a resource are resolved against the project (a tool also through an accepted share), and one naming nothing there empties the rollup rather than being dropped: a mistyped id must never read back as the project's whole spend. The other six are matched against the value the event recorded.
+   * Narrowings intersect and apply to the whole rollup — every bucket, `totals` and `totals.distinct` alike — so each composes with any `group_by`, and all sixteen are echoed back under `filters`. The ten naming a resource match the id the event recorded, so a deleted resource's spend stays selectable; an id no event recorded empties the rollup rather than being dropped: a mistyped id must never read back as the project's whole spend. `orchestration_id` is the exception — the event records no orchestration, so a deleted one empties the rollup. The other six are matched against the value the event recorded.
    *
    * `groups` is paginated. Its `total` is the number of distinct buckets in the window — bucket cardinality, including a null bucket for the events the dimension does not apply to, and never a count of entities. To count entities, send `include=distinct` and read `totals.distinct`. `totals` always describes the whole window — never the page above it, and never the bucketing: omit `group_by` and `totals` is unchanged while `groups` is empty.
    *

package/dist/index.d.mts

@@ -526,6 +526,16 @@ type Agent = {
    * Maximum number of recent messages to include in the context window sent to the model. When null, all messages are included.
    */
   max_context_messages?: number | null;
+  /**
+   * Maximum number of tokens the model may generate per step of a turn. A response that reaches it stops with `stop_reason` `length`. Null leaves the model's default.
+   */
+  max_output_tokens?: number | null;
+  /**
+   * How much the model reasons before answering. Translated per provider (Anthropic thinking, Gemini thinking config, OpenAI reasoning effort) against what the model supports; a model that cannot reason ignores it. `none` turns reasoning off where the model allows it. Reasoning tokens are metered as output. Null leaves the model's default.
+   */
+  reasoning?: {
+    effort: 'none' | 'low' | 'medium' | 'high';
+  } | null;
   /**
    * When true, only one open session per actor_id is allowed for this agent. Creating a second open session for the same actor returns 409.
    */
@@ -587,7 +597,7 @@ type AgentVersion = {
    */
   version?: number;
   /**
-   * The agent's configuration as it stood at this version: every mutable field of the `Agent` schema (`instructions`, `model`, `tool_bindings`, `max_steps`, `tool_choice`, `stop_conditions`, `active_tool_ids`, `step_rules`, `boundary_policy`, `temperature`, `knowledge_config`, `output_schema`, `max_context_messages`, `single_session_per_actor`, `on_approval_expiry`, `guardrail_ids`, `ai_provider_id`, `model_route_id`, `name`), and none of its identity or bookkeeping fields (`id`, `project_id`, `version`, `active_release`, timestamps).
+   * The agent's configuration as it stood at this version: every mutable field of the `Agent` schema (`instructions`, `model`, `tool_bindings`, `max_steps`, `tool_choice`, `stop_conditions`, `active_tool_ids`, `step_rules`, `boundary_policy`, `temperature`, `knowledge_config`, `output_schema`, `max_context_messages`, `max_output_tokens`, `reasoning`, `single_session_per_actor`, `on_approval_expiry`, `guardrail_ids`, `ai_provider_id`, `model_route_id`, `name`), and none of its identity or bookkeeping fields (`id`, `project_id`, `version`, `active_release`, timestamps).
    *
    * Deliberately open rather than a fixed schema: an archive written by an earlier release of SOAT reflects the agent surface **of its own time**, so it may carry fields the current schema no longer defines, or lack ones it has since gained. Knowledge retrieval is not part of the snapshot — a version records which `knowledge_config` applied, while the documents and memory stores it resolves keep their own histories and are pinned at generation time.
    */
@@ -764,6 +774,16 @@ type CreateAgentRequest = {
    * Maximum number of recent messages included in the context window. Null means no limit.
    */
   max_context_messages?: number;
+  /**
+   * Maximum number of tokens the model may generate per step of a turn. A response that reaches it stops with `stop_reason` `length`. Null leaves the model's default.
+   */
+  max_output_tokens?: number | null;
+  /**
+   * How much the model reasons before answering. Translated per provider (Anthropic thinking, Gemini thinking config, OpenAI reasoning effort) against what the model supports; a model that cannot reason ignores it. `none` turns reasoning off where the model allows it. Reasoning tokens are metered as output. Null leaves the model's default.
+   */
+  reasoning?: {
+    effort: 'none' | 'low' | 'medium' | 'high';
+  } | null;
   /**
    * When true, only one open session per actor_id is allowed for this agent.
    */
@@ -899,6 +919,16 @@ type UpdateAgentRequest = {
    * Maximum number of recent messages included in the context window. Null means no limit.
    */
   max_context_messages?: number | null;
+  /**
+   * Maximum number of tokens the model may generate per step of a turn. A response that reaches it stops with `stop_reason` `length`. Null leaves the model's default.
+   */
+  max_output_tokens?: number | null;
+  /**
+   * How much the model reasons before answering. Translated per provider (Anthropic thinking, Gemini thinking config, OpenAI reasoning effort) against what the model supports; a model that cannot reason ignores it. `none` turns reasoning off where the model allows it. Reasoning tokens are metered as output. Null leaves the model's default.
+   */
+  reasoning?: {
+    effort: 'none' | 'low' | 'medium' | 'high';
+  } | null;
   /**
    * When true, only one open session per actor_id is allowed for this agent.
    */
@@ -2942,6 +2972,16 @@ type AgentResourceProperties = {
    * Maximum number of recent messages to include in the context window sent to the model. When null, all messages are included.
    */
   max_context_messages?: number | null;
+  /**
+   * Maximum number of tokens the model may generate per step of a turn. A response that reaches it stops with `stop_reason` `length`. Null leaves the model's default.
+   */
+  max_output_tokens?: number | null;
+  /**
+   * How much the model reasons before answering. Translated per provider (Anthropic thinking, Gemini thinking config, OpenAI reasoning effort) against what the model supports; a model that cannot reason ignores it. `none` turns reasoning off where the model allows it. Reasoning tokens are metered as output. Null leaves the model's default.
+   */
+  reasoning?: {
+    effort: 'none' | 'low' | 'medium' | 'high';
+  } | null;
   /**
    * When true, only one open session per actor_id is allowed for this agent.
    */
@@ -3671,6 +3711,10 @@ type SessionResourceProperties = {
    * Number of seconds of inactivity after which the session expires. 0 means never expires.
    */
   inactivity_ttl_seconds?: number;
+  /**
+   * Maximum number of recent conversation messages sent to the model on each generation, taking precedence over the agent's. null defers to the agent.
+   */
+  max_context_messages?: number | null;
   /**
    * Optional context object passed to tool calls. Write-only: no read of a session returns it.
    */
@@ -6251,6 +6295,11 @@ type SessionRecord = {
    *
    */
   message_delay_seconds?: number | null;
+  /**
+   * Maximum number of recent conversation messages sent to the model on each generation. Takes precedence over the agent's `max_context_messages`; null defers to it. A fork keeps its parent's value.
+   *
+   */
+  max_context_messages?: number | null;
   /**
    * Timestamp of the last activity on the session (message added or response generated).
    */
@@ -6331,6 +6380,11 @@ type CreateSessionRequest = {
    *
    */
   message_delay_seconds?: number | null;
+  /**
+   * Maximum number of recent conversation messages sent to the model on each generation, taking precedence over the agent's `max_context_messages`. null or absent defers to the agent.
+   *
+   */
+  max_context_messages?: number | null;
 };
 type UpdateSessionRequest = {
   /**
@@ -6361,6 +6415,11 @@ type UpdateSessionRequest = {
    *
    */
   message_delay_seconds?: number | null;
+  /**
+   * Maximum number of recent conversation messages sent to the model on each generation, taking precedence over the agent's `max_context_messages`. Set to null to defer to the agent.
+   *
+   */
+  max_context_messages?: number | null;
 };
 type AddSessionMessageRequest = {
   /**
@@ -7415,7 +7474,7 @@ type UsageEvent = {
    */
   session_id?: string | null;
   /**
-   * Public ID of the AI provider instance billed — the target a model route picked for the turn, or the agent's pinned provider. Null if the provider was later deleted; the `provider`/`model` snapshot still records what was billed.
+   * Public ID of the AI provider instance billed — the target a model route picked for the turn, or the agent's pinned provider. Kept after the provider is deleted, like every attribution id on the event.
    *
    */
   ai_provider_id?: string | null;
@@ -7430,7 +7489,7 @@ type UsageEvent = {
    */
   action_id?: string | null;
   /**
-   * Public ID of the tool a `tool_execution` event metered. Null on every other meter, for an inline tool definition, and once the tool is deleted.
+   * Public ID of the tool a `tool_execution` event metered. Null on every other meter and for an inline tool definition.
    *
    */
   tool_id?: string | null;
@@ -7440,12 +7499,12 @@ type UsageEvent = {
    */
   publisher_project_id?: string | null;
   /**
-   * Public ID of the document an `embedding` event embedded a chunk of — ingestion, a re-chunk, a conversation's turns. Null on a query embedding, on every other meter, and once the document is deleted.
+   * Public ID of the document an `embedding` event embedded a chunk of — ingestion, a re-chunk, a conversation's turns. Null on a query embedding and on every other meter.
    *
    */
   document_id?: string | null;
   /**
-   * Public ID of the memory store an `embedding` event embedded written content for. The store rather than the memory: stored content is shared by every memory in the store that states it. Null on a query embedding, on every other meter, and once the store is deleted.
+   * Public ID of the memory store an `embedding` event embedded written content for. The store rather than the memory: stored content is shared by every memory in the store that states it. Null on a query embedding and on every other meter.
    *
    */
   memory_store_id?: string | null;
@@ -20752,22 +20811,22 @@ type ListUsageEventsData = {
      */
     trace_id?: string;
     /**
-     * Filter by the actor (end user) the usage is attributed to. An actor that does not exist in scope yields an empty page.
+     * Filter by the actor (end user) the usage is attributed to.
      *
      */
     actor_id?: string;
     /**
-     * Filter by the session the usage was produced in. A session that does not exist in scope yields an empty page.
+     * Filter by the session the usage was produced in.
      *
      */
     session_id?: string;
     /**
-     * Filter by the provider record the spend was billed against. A provider that does not exist in scope yields an empty page.
+     * Filter by the provider record the spend was billed against.
      *
      */
     ai_provider_id?: string;
     /**
-     * Filter by the orchestration run that metered the event. A run that does not exist in scope yields an empty page.
+     * Filter by the orchestration run that metered the event.
      *
      */
     orchestration_run_id?: string;
@@ -20790,12 +20849,12 @@ type ListUsageEventsData = {
      */
     action_id?: string;
     /**
-     * Filter by the tool a `tool_execution` event metered, including a tool another project shares with one in scope. A tool that does not exist in scope yields an empty page.
+     * Filter by the tool a `tool_execution` event metered, including a tool another project shares with one in scope.
      *
      */
     tool_id?: string;
     /**
-     * Filter to the calls made through a share of a resource the named project owns. A project that does not exist yields an empty page.
+     * Filter to the calls made through a share of a resource the named project owns.
      *
      */
     publisher_project_id?: string;
@@ -20886,7 +20945,7 @@ type GetUsageAggregateData = {
      */
     meter_type?: string;
     /**
-     * Narrow the whole rollup — every bucket, `totals` and `totals.distinct` alike — to one session's traffic. Combines with `group_by`: `group_by=day&session_id=…` is a conversation's spend per day, `group_by=model` is the same spend split by model. An id naming no session in this project yields an empty rollup, never the project total.
+     * Narrow the whole rollup — every bucket, `totals` and `totals.distinct` alike — to one session's traffic. Combines with `group_by`: `group_by=day&session_id=…` is a conversation's spend per day, `group_by=model` is the same spend split by model. Like every id narrowing it matches the id the event recorded, so a deleted session's spend stays selectable; an id no event in this project recorded yields an empty rollup, never the project total.
      *
      */
     session_id?: string;
@@ -20944,12 +21003,12 @@ type GetUsageAggregateData = {
      */
     action_id?: string;
     /**
-     * Narrow to the executions of one tool, including one another project shares with this one. An id naming no tool in reach yields an empty rollup.
+     * Narrow to the executions of one tool, including one another project shares with this one.
      *
      */
     tool_id?: string;
     /**
-     * Narrow to the calls made through a share of a resource the named project owns. A project that does not exist yields an empty rollup.
+     * Narrow to the calls made through a share of a resource the named project owns.
      *
      */
     publisher_project_id?: string;
@@ -22237,7 +22296,7 @@ export declare class AiProviders {
    *
    * Deletes an AI provider configuration.
    *
-   * Live references — chats, agents, and model routes whose targets name this provider — always block deletion with `409 AI_PROVIDER_HAS_DEPENDENTS`; `force` does not override them, so delete or repoint those resources first. Soft dependents — price overrides and usage/generation records — also block with `409` unless `force=true`, which deletes the provider's price overrides and unlinks (nulls) its usage history, preserving those rows. The `409` body's `error.meta` reports the counts, a sample of offending IDs, and a `forcible` flag that is `true` when a `force=true` retry would succeed.
+   * Live references — chats, agents, and model routes whose targets name this provider — always block deletion with `409 AI_PROVIDER_HAS_DEPENDENTS`; `force` does not override them, so delete or repoint those resources first. Soft dependents — price overrides and usage/generation records — also block with `409` unless `force=true`, which deletes the provider's price overrides and unlinks its usage history, preserving those rows: each usage event keeps the provider's id as its `ai_provider_id`. The `409` body's `error.meta` reports the counts, a sample of offending IDs, and a `forcible` flag that is `true` when a `force=true` retry would succeed.
    *
    */
   static deleteAiProvider<ThrowOnError extends boolean = false>(options: Options<DeleteAiProviderData, ThrowOnError>): RequestResult<DeleteAiProviderResponses, DeleteAiProviderErrors, ThrowOnError>;
@@ -24163,7 +24222,7 @@ export declare class Usage {
    *
    * Returns the raw usage events the caller can access, most recent first, narrowed by any combination of the same filters the aggregate takes — agent, generation, trace, actor, session, AI provider, orchestration run, orchestration, tool, trigger, action, outcome, meter type, model or `source` — so a rollup and the events behind it are addressed the same way. Each event is the per-generation token usage as reported by the provider, for audit and reconciliation.
    *
-   * An id naming nothing in scope yields an empty page rather than dropping the filter, so a mistyped narrowing can never widen the list past what was asked for.
+   * An id filter matches the id the event recorded, so it still selects the spend of an agent, provider, session or tool since deleted. An id no event in scope recorded yields an empty page rather than dropping the filter, so a mistyped narrowing can never widen the list past what was asked for.
    *
    */
   static listUsageEvents<ThrowOnError extends boolean = false>(options?: Options<ListUsageEventsData, ThrowOnError>): RequestResult<ListUsageEventsResponses, ListUsageEventsErrors, ThrowOnError>;
@@ -24172,7 +24231,7 @@ export declare class Usage {
    *
    * Returns a project's usage rolled up over an optional `[from, to]` time window, optionally bucketed by a single dimension and narrowed by any combination of sixteen filters — session, actor, agent, AI provider, orchestration run, orchestration, generation, trace, tool, publisher project, meter type, model, source, trigger, action and outcome. That is how "what did this conversation cost", "what has this end user cost" and "what did this agent cost" are answered. Each group and the grand total carry an `event_count`, summed token counts, a measured `quantity` per component (so infra meters report their real amount, not zeros), and `cost_usd` (null when no event in the bucket was priced). This is the per-project cost-by-range/by-category query — a monthly figure without scanning raw usage events client-side.
    *
-   * Narrowings intersect and apply to the whole rollup — every bucket, `totals` and `totals.distinct` alike — so each composes with any `group_by`, and all sixteen are echoed back under `filters`. The ten naming a resource are resolved against the project (a tool also through an accepted share), and one naming nothing there empties the rollup rather than being dropped: a mistyped id must never read back as the project's whole spend. The other six are matched against the value the event recorded.
+   * Narrowings intersect and apply to the whole rollup — every bucket, `totals` and `totals.distinct` alike — so each composes with any `group_by`, and all sixteen are echoed back under `filters`. The ten naming a resource match the id the event recorded, so a deleted resource's spend stays selectable; an id no event recorded empties the rollup rather than being dropped: a mistyped id must never read back as the project's whole spend. `orchestration_id` is the exception — the event records no orchestration, so a deleted one empties the rollup. The other six are matched against the value the event recorded.
    *
    * `groups` is paginated. Its `total` is the number of distinct buckets in the window — bucket cardinality, including a null bucket for the events the dimension does not apply to, and never a count of entities. To count entities, send `include=distinct` and read `totals.distinct`. `totals` always describes the whole window — never the page above it, and never the bucketing: omit `group_by` and `totals` is unchanged while `groups` is empty.
    *

package/dist/index.mjs

@@ -993,7 +993,7 @@ var AiProviders = class {
 	*
 	* Deletes an AI provider configuration.
 	*
-	* Live references — chats, agents, and model routes whose targets name this provider — always block deletion with `409 AI_PROVIDER_HAS_DEPENDENTS`; `force` does not override them, so delete or repoint those resources first. Soft dependents — price overrides and usage/generation records — also block with `409` unless `force=true`, which deletes the provider's price overrides and unlinks (nulls) its usage history, preserving those rows. The `409` body's `error.meta` reports the counts, a sample of offending IDs, and a `forcible` flag that is `true` when a `force=true` retry would succeed.
+	* Live references — chats, agents, and model routes whose targets name this provider — always block deletion with `409 AI_PROVIDER_HAS_DEPENDENTS`; `force` does not override them, so delete or repoint those resources first. Soft dependents — price overrides and usage/generation records — also block with `409` unless `force=true`, which deletes the provider's price overrides and unlinks its usage history, preserving those rows: each usage event keeps the provider's id as its `ai_provider_id`. The `409` body's `error.meta` reports the counts, a sample of offending IDs, and a `forcible` flag that is `true` when a `force=true` retry would succeed.
 	*
 	*/
 	static deleteAiProvider(options) {
@@ -4752,7 +4752,7 @@ var Usage = class {
 	*
 	* Returns the raw usage events the caller can access, most recent first, narrowed by any combination of the same filters the aggregate takes — agent, generation, trace, actor, session, AI provider, orchestration run, orchestration, tool, trigger, action, outcome, meter type, model or `source` — so a rollup and the events behind it are addressed the same way. Each event is the per-generation token usage as reported by the provider, for audit and reconciliation.
 	*
-	* An id naming nothing in scope yields an empty page rather than dropping the filter, so a mistyped narrowing can never widen the list past what was asked for.
+	* An id filter matches the id the event recorded, so it still selects the spend of an agent, provider, session or tool since deleted. An id no event in scope recorded yields an empty page rather than dropping the filter, so a mistyped narrowing can never widen the list past what was asked for.
 	*
 	*/
 	static listUsageEvents(options) {
@@ -4766,7 +4766,7 @@ var Usage = class {
 	*
 	* Returns a project's usage rolled up over an optional `[from, to]` time window, optionally bucketed by a single dimension and narrowed by any combination of sixteen filters — session, actor, agent, AI provider, orchestration run, orchestration, generation, trace, tool, publisher project, meter type, model, source, trigger, action and outcome. That is how "what did this conversation cost", "what has this end user cost" and "what did this agent cost" are answered. Each group and the grand total carry an `event_count`, summed token counts, a measured `quantity` per component (so infra meters report their real amount, not zeros), and `cost_usd` (null when no event in the bucket was priced). This is the per-project cost-by-range/by-category query — a monthly figure without scanning raw usage events client-side.
 	*
-	* Narrowings intersect and apply to the whole rollup — every bucket, `totals` and `totals.distinct` alike — so each composes with any `group_by`, and all sixteen are echoed back under `filters`. The ten naming a resource are resolved against the project (a tool also through an accepted share), and one naming nothing there empties the rollup rather than being dropped: a mistyped id must never read back as the project's whole spend. The other six are matched against the value the event recorded.
+	* Narrowings intersect and apply to the whole rollup — every bucket, `totals` and `totals.distinct` alike — so each composes with any `group_by`, and all sixteen are echoed back under `filters`. The ten naming a resource match the id the event recorded, so a deleted resource's spend stays selectable; an id no event recorded empties the rollup rather than being dropped: a mistyped id must never read back as the project's whole spend. `orchestration_id` is the exception — the event records no orchestration, so a deleted one empties the rollup. The other six are matched against the value the event recorded.
 	*
 	* `groups` is paginated. Its `total` is the number of distinct buckets in the window — bucket cardinality, including a null bucket for the events the dimension does not apply to, and never a count of entities. To count entities, send `include=distinct` and read `totals.distinct`. `totals` always describes the whole window — never the page above it, and never the bucketing: omit `group_by` and `totals` is unchanged while `groups` is empty.
 	*

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@soat/sdk",
-  "version": "0.77.0",
+  "version": "0.78.0",
   "description": "TypeScript SDK for SOAT — open-source infrastructure for production-ready AI agents.",
   "keywords": [
     "ai-agents",