chore(llm): upgrade pi-ai to 0.85.1

.agents/notes/implemented/bug-fix/2026-09-05-pi-ai-upgrade-compatibility.i18n.yaml

@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/bug-fix/2026-09-05-pi-ai-upgrade-compatibility.md
+2026-09-05-pi-ai-upgrade-compatibility.md: 511bb5a9e1c805832a90d0c3364995cd98b2bf7d
+2026-09-05-pi-ai-upgrade-compatibility.zh.md: 779931adfa8c02d6a660cafb35df6cbb1bc14c2b

.agents/notes/implemented/bug-fix/2026-09-05-pi-ai-upgrade-compatibility.md

@@ -0,0 +1,25 @@
+# Agent Note: pi-ai upgrade compatibility
+
+Status: implemented
+
+English | [中文](2026-09-05-pi-ai-upgrade-compatibility.zh.md)
+
+## Problem
+
+The pi-ai adapter classifies upstream compatibility fields explicitly and persists only replay metadata needed by later requests. An SDK upgrade can add fields to either set without changing the Harness provider-neutral API. Unclassified configuration fields fail compilation; omitted replay metadata can silently change subsequent provider requests.
+
+## Decision
+
+The adapter follows [pi-ai 0.85.1](https://github.com/earendil-works/pi/blob/v0.85.1/packages/ai/CHANGELOG.md). `thinkingTokenBudgetField`, `vllmPriority`, and `supportsMaxOutputTokens` are opt-in gateway controls; `thinking.budget` joins the existing template placeholders. The SDK owns budget resolution and serialization. `supportsMidConvoEffort` and `allowedFallbackModels` remain catalog-owned because their correctness depends on exact Anthropic transports, model capabilities, and fallback pricing.
+
+Optional `providerThinkingLevel` remains in the adapter replay-v2 response metadata so Anthropic history retains its provider-native effort. Absence remains valid; neither the replay version nor the released Session format changes. The provider-neutral LLM API stays unchanged.
+
+## Alternatives considered
+
+**Withhold every new field.** This would misclassify deployment-owned gateway controls as catalog facts: upstream explicitly leaves budget-field selection and vLLM priority out of its generated catalog.
+
+**Expose every new field.** This would let arbitrary gateways claim model-specific Anthropic effort and fallback support without the catalog evidence that makes those features valid.
+
+## Consequences
+
+Compile-time coverage retains explicit field classification. [Compatibility tests](../../../../packages/llm/llm-pi-ai/tests/compat-upgrade.spec.ts) cover schema acceptance, invalid values, protocol applicability, and materialization without changing defaults. [Replay conversion tests](../../../../packages/llm/llm-pi-ai/tests/convert.spec.ts) cover optional effort preservation. Mixed-protocol catalog tests use the installed OpenCode catalog. Provider behavior remains upstream-owned; live provider verification is separate from keyless adapter tests.

.agents/notes/implemented/bug-fix/2026-09-05-pi-ai-upgrade-compatibility.zh.md

@@ -0,0 +1,25 @@
+# Agent Note: pi-ai 升级兼容性
+
+Status: implemented
+
+[English](2026-09-05-pi-ai-upgrade-compatibility.md) | 中文
+
+## Problem
+
+pi-ai 适配器显式分类上游兼容字段，并且只持久化后续请求需要的回放元数据。SDK 升级可能在不改变 Harness 提供方无关 API 的情况下为任一集合新增字段。未分类的配置字段会导致编译失败；遗漏回放元数据则可能静默改变后续提供方请求。
+
+## Decision
+
+适配器遵循 [pi-ai 0.85.1](https://github.com/earendil-works/pi/blob/v0.85.1/packages/ai/CHANGELOG.md)。`thinkingTokenBudgetField`、`vllmPriority` 和 `supportsMaxOutputTokens` 是显式启用的网关控制；`thinking.budget` 加入现有模板占位符。SDK 拥有预算解析和序列化。`supportsMidConvoEffort` 和 `allowedFallbackModels` 仍由目录拥有，因为其正确性依赖确切的 Anthropic 传输、模型能力和回退定价。
+
+可选的 `providerThinkingLevel` 保存在适配器 replay-v2 响应元数据中，让 Anthropic 历史保留提供方原生 effort。缺失仍然有效；回放版本与已发布 Session 格式均不改变。提供方无关的 LLM API 保持不变。
+
+## Alternatives considered
+
+**保留所有新增字段为不可配置。** 这会把部署拥有的网关控制误分类为目录事实：上游明确不在生成目录中设置预算字段选择和 vLLM 优先级。
+
+**开放所有新增字段。** 这会允许任意网关在缺少目录证据时声明支持模型专属的 Anthropic effort 与回退能力，而这些证据正是相关功能有效的依据。
+
+## Consequences
+
+编译期覆盖保留显式字段分类。[兼容性测试](../../../../packages/llm/llm-pi-ai/tests/compat-upgrade.spec.ts) 覆盖 schema 接受、无效值、协议适用范围及物化，且不改变默认值。[回放转换测试](../../../../packages/llm/llm-pi-ai/tests/convert.spec.ts) 覆盖可选 effort 的保留。混合协议目录测试使用已安装的 OpenCode 目录。提供方行为仍由上游拥有；真实提供方验证与无需密钥的适配器测试分开。

docs/config-catalog.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/config-catalog.md
-config-catalog.md: 66657ff6c3f5874a4922f26fd61c65ebb7f37826
-config-catalog.zh.md: 6ae44ca258720dd1a7aa567f0ed71e996410b9ea
+config-catalog.md: e386886329a0aa71f809ed47c4a730c40c3e0419
+config-catalog.zh.md: ef4281a2ddd51010893af482999314a8915b8f19

docs/config-catalog.md

@@ -1248,8 +1248,14 @@ export interface PiAiCompatProfile {
   chatTemplateKwargs?: NonNullable<OpenAICompletionsCompat['chatTemplateKwargs']>
   /** Arguments sent as `chat_template_args` under the `baseten` thinking format; `openai-completions`. */
   chatTemplateArgs?: NonNullable<OpenAICompletionsCompat['chatTemplateArgs']>
-  /** Whether the endpoint accepts `thinking_token_budget` to cap vLLM reasoning; `openai-completions`. */
+  /** Alias for `thinkingTokenBudgetField: "thinking_token_budget"`; an explicit field wins. `openai-completions`. */
   supportsThinkingTokenBudget?: boolean
+  /** Request field carrying the reasoning budget from `thinkingBudgets`; omitted unless configured. `openai-completions`. */
+  thinkingTokenBudgetField?: PiAiThinkingTokenBudgetField
+  /** vLLM scheduler `priority`; lower runs earlier, and the server must enable priority scheduling. Omitted unless configured. */
+  vllmPriority?: number
+  /** Whether the endpoint accepts `max_output_tokens`; `false` omits it. The three Responses protocols. */
+  supportsMaxOutputTokens?: boolean
   /**
    * Whether the endpoint accepts `strict` in tool definitions;
    * `openai-completions`, the three Responses protocols, `bedrock-converse-stream`.
@@ -1291,11 +1297,14 @@ export type PiAiReasoningEfforts = Partial<Record<ModelThinkingLevel, string | n
 
 /** One reasoning-dispatch wire format a profile may name. */
 export type PiAiThinkingFormat = NonNullable<OpenAICompletionsCompat['thinkingFormat']>
+
+/** The reasoning-budget field spellings pi-ai accepts. */
+export type PiAiThinkingTokenBudgetField = NonNullable<OpenAICompletionsCompat['thinkingTokenBudgetField']>
 ```
 
 Depends on: `Api` (`@earendil-works/pi-ai`) · `CacheRetention` (`@earendil-works/pi-ai`) · `Model` (`@earendil-works/pi-ai`) · `ModelThinkingLevel` (`@earendil-works/pi-ai`) · `OpenAICompletionsCompat` (`@earendil-works/pi-ai`) · [`RetryPolicyConfig`](../packages/llm/llm/src/index.ts) · `ThinkingBudgets` (`@earendil-works/pi-ai`) · `Transport` (`@earendil-works/pi-ai`)
 
-Source: [`packages/llm/llm-pi-ai/src/config.ts:216`](../packages/llm/llm-pi-ai/src/config.ts)
+Source: [`packages/llm/llm-pi-ai/src/config.ts:217`](../packages/llm/llm-pi-ai/src/config.ts)
 
 <a id="deepseek-aidsh-llm-replay"></a>
 

docs/config-catalog.zh.md

@@ -1250,8 +1250,14 @@ export interface PiAiCompatProfile {
   chatTemplateKwargs?: NonNullable<OpenAICompletionsCompat['chatTemplateKwargs']>
   /** Arguments sent as `chat_template_args` under the `baseten` thinking format; `openai-completions`. */
   chatTemplateArgs?: NonNullable<OpenAICompletionsCompat['chatTemplateArgs']>
-  /** Whether the endpoint accepts `thinking_token_budget` to cap vLLM reasoning; `openai-completions`. */
+  /** Alias for `thinkingTokenBudgetField: "thinking_token_budget"`; an explicit field wins. `openai-completions`. */
   supportsThinkingTokenBudget?: boolean
+  /** Request field carrying the reasoning budget from `thinkingBudgets`; omitted unless configured. `openai-completions`. */
+  thinkingTokenBudgetField?: PiAiThinkingTokenBudgetField
+  /** vLLM scheduler `priority`; lower runs earlier, and the server must enable priority scheduling. Omitted unless configured. */
+  vllmPriority?: number
+  /** Whether the endpoint accepts `max_output_tokens`; `false` omits it. The three Responses protocols. */
+  supportsMaxOutputTokens?: boolean
   /**
    * Whether the endpoint accepts `strict` in tool definitions;
    * `openai-completions`, the three Responses protocols, `bedrock-converse-stream`.
@@ -1293,11 +1299,14 @@ export type PiAiReasoningEfforts = Partial<Record<ModelThinkingLevel, string | n
 
 /** One reasoning-dispatch wire format a profile may name. */
 export type PiAiThinkingFormat = NonNullable<OpenAICompletionsCompat['thinkingFormat']>
+
+/** The reasoning-budget field spellings pi-ai accepts. */
+export type PiAiThinkingTokenBudgetField = NonNullable<OpenAICompletionsCompat['thinkingTokenBudgetField']>
 ```
 
 依赖：`Api`（`@earendil-works/pi-ai`）· `CacheRetention`（`@earendil-works/pi-ai`）· `Model`（`@earendil-works/pi-ai`）· `ModelThinkingLevel`（`@earendil-works/pi-ai`）· `OpenAICompletionsCompat`（`@earendil-works/pi-ai`）· [`RetryPolicyConfig`](../packages/llm/llm/src/index.ts) · `ThinkingBudgets`（`@earendil-works/pi-ai`）· `Transport`（`@earendil-works/pi-ai`)
 
-来源：[`packages/llm/llm-pi-ai/src/config.ts:213`](../packages/llm/llm-pi-ai/src/config.ts)
+来源：[`packages/llm/llm-pi-ai/src/config.ts:217`](../packages/llm/llm-pi-ai/src/config.ts)
 
 <a id="deepseek-aidsh-llm-replay"></a>
 

packages/llm/llm-pi-ai/README.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write packages/llm/llm-pi-ai/README.md
-README.md: c0aacbe3e0b727cd7b6b163814dc96859b954b9c
-README.zh.md: 7b8cef7db2ba5cebe3e42345b308716ae0099c03
+README.md: 6f10f6998e5213aaa44ac51970d473bebb904cba
+README.zh.md: 492bb53f3f9e0afd427d10414dcff99feb62560b

packages/llm/llm-pi-ai/README.md

@@ -100,6 +100,8 @@ A profile's `models` list replaces the route's installed catalog rather than ext
 
 `reasoningEfforts` declares a model's selectable thinking levels: each key is a level selectors offer, its value the spelling dispatch sends on the wire, so `max: ultra` renames a level for a gateway with its own vocabulary. Omitting the field keeps the installed catalog entry's capability; `false` declares a non-reasoning model. `compat` switches reshape the request for endpoints pi-ai cannot recognize — which role carries the system prompt, which field caps output, how a thinking level travels — configurable per route and per model. A model neither the entry nor the installed catalog sizes takes the route's `defaultContextWindow` and `defaultMaxTokens` fallbacks.
 
+For self-hosted Chat Completions endpoints, `thinkingTokenBudgetField` selects the reasoning-budget parameter, and `vllmPriority` sets an integer scheduler priority when the server enables priority scheduling. Template arguments accept `$var: thinking.budget`. Responses gateways can set `supportsMaxOutputTokens: false` to omit `max_output_tokens`. These controls are opt-in; catalog-owned Anthropic effort and fallback capabilities are not configurable switches.
+
 ### Change configuration at runtime
 
 Profiles are re-read once per operation through the optional settings seam: the base and the user's `llm-pi-ai:` settings section merge per provider, so a user can add a route, override one field of a composition route, or point a route at another proxy, all effective on the next request with no restart. A section the adapter could not serve is refused where it is written — `settings.mutate` answers `settings-rejected` — and a stored section that later fails keeps the namespace's last good value. When the route set or a route's retry policy changes, the plugin re-registers atomically: a conflicting route leaves the previous routes serving.
@@ -147,7 +149,7 @@ The plugin declares every installed catalog provider it can authenticate in the
 
 ### Replay and vocabulary
 
-Successful assistant responses store a versioned, lossless-JSON replay state beside the provider and model that produced them — response-level facts plus one per-block entry per streamed block. At request time, `LlmRuntime` passes replay state only when the same adapter instance owns both routes; the adapter validates it and restores native response ids and provider signatures, degrading an unusable state to provider-neutral content instead of failing the request. pi-ai tool-call arguments are parsed objects, so the adapter parses input and re-stringifies output to the harness raw-JSON convention; pi-ai in-stream error events map to terminal `finish` chunks.
+Successful assistant responses store a versioned, lossless-JSON replay state beside the provider and model that produced them — response-level facts plus one per-block entry per streamed block. At request time, `LlmRuntime` passes replay state only when the same adapter instance owns both routes; the adapter validates it and restores native response ids, provider signatures, and optional `providerThinkingLevel` effort metadata, keeping absent effort metadata absent, degrading an unusable state to provider-neutral content instead of failing the request. pi-ai tool-call arguments are parsed objects, so the adapter parses input and re-stringifies output to the harness raw-JSON convention; pi-ai in-stream error events map to terminal `finish` chunks.
 
 </details>
 

packages/llm/llm-pi-ai/README.zh.md

@@ -100,6 +100,8 @@ profile 的 `models` 列表会替换而非扩展路由的已安装目录；每
 
 `reasoningEfforts` 声明模型可选择的 thinking 等级：每个键都是选择器提供的等级，其值是该等级过线的拼写，因此 `max: ultra` 可以为拥有自有词汇的网关重命名等级。省略该字段时保留已安装目录条目的能力；`false` 声明非推理模型。对于 pi-ai 无法识别的端点，`compat` 开关重塑请求——哪个角色携带系统提示词、哪个字段限制输出、thinking 等级如何传递——可逐路由、逐模型配置。条目与已安装目录都没有尺寸的模型，会采用路由的 `defaultContextWindow` 与 `defaultMaxTokens` 回退值。
 
+对于自托管 Chat Completions 端点，`thinkingTokenBudgetField` 选择推理预算参数，`vllmPriority` 在服务端启用优先级调度时设置整数调度优先级。模板参数接受 `$var: thinking.budget`。Responses 网关可设置 `supportsMaxOutputTokens: false` 来省略 `max_output_tokens`。这些控制均需显式启用；目录拥有的 Anthropic effort 和回退能力不是可配置开关。
+
 ### 运行时更改配置
 
 profile 通过可选 settings seam 每次操作重新读取：base 与用户的 `llm-pi-ai:` 设置分节按提供方合并，因此用户可以新增路由、覆盖组合路由的一个字段或把路由指向另一个代理，全部在下一个请求生效、无需重启。适配器无法服务的分节会在写入处被拒绝——`settings.mutate` 回答 `settings-rejected`——之后失效的已存储分节会保留 namespace 最后有效值。当路由集合或某路由的重试策略变化时，插件会原子地重新注册：冲突路由会让此前路由继续服务。
@@ -147,7 +149,7 @@ pi-ai 不提供的路由需要 `api`、`baseURL` 与非空 `models` 列表；无
 
 ### 回放与词汇
 
-成功 assistant 响应会存储带版本的、无损 JSON 回放状态，与产生它们的提供方和模型放在一起——响应级事实加每个流式块一条逐块条目。请求时，`LlmRuntime` 仅当同一适配器实例拥有两条路由时才传递回放状态；适配器校验它并恢复原生响应 id 与提供方签名，无法使用的状态会降级为提供方无关内容而不是让请求失败。pi-ai 工具调用参数是解析后的对象，因此适配器解析输入并重新字符串化输出，以符合 harness 原始 JSON 约定；pi-ai 流内错误事件映射为终止 `finish` 分片。
+成功 assistant 响应会存储带版本的、无损 JSON 回放状态，与产生它们的提供方和模型放在一起——响应级事实加每个流式块一条逐块条目。请求时，`LlmRuntime` 仅当同一适配器实例拥有两条路由时才传递回放状态；适配器校验它并恢复原生响应 id、提供方签名与可选的 `providerThinkingLevel` effort 元数据，缺失的 effort 元数据仍保持缺失，无法使用的状态会降级为提供方无关内容而不是让请求失败。pi-ai 工具调用参数是解析后的对象，因此适配器解析输入并重新字符串化输出，以符合 harness 原始 JSON 约定；pi-ai 流内错误事件映射为终止 `finish` 分片。
 
 </details>
 

packages/llm/llm-pi-ai/package.json

@@ -41,7 +41,7 @@
     "@deepseek-ai/dsh-brand": "workspace:^",
     "@deepseek-ai/dsh-util-values": "workspace:^",
     "@deepseek-ai/schemastery": "workspace:^",
-    "@earendil-works/pi-ai": "^0.84.2"
+    "@earendil-works/pi-ai": "^0.85.1"
   },
   "devDependencies": {
     "@deepseek-ai/cordis": "workspace:^",

packages/llm/llm-pi-ai/src/catalog.ts

@@ -125,6 +125,19 @@ const MAX_TOKENS_FIELD_GATE: Record<PiAiMaxTokensField, true> = {
 /** The output-cap field spellings a profile may name. */
 export const MAX_TOKENS_FIELDS = Object.keys(MAX_TOKENS_FIELD_GATE) as readonly PiAiMaxTokensField[]
 
+/** The reasoning-budget field spellings pi-ai accepts. */
+export type PiAiThinkingTokenBudgetField = NonNullable<OpenAICompletionsCompat['thinkingTokenBudgetField']>
+
+/** An upstream reasoning-budget spelling must be classified before it can be configured. */
+const THINKING_TOKEN_BUDGET_FIELD_GATE: Record<PiAiThinkingTokenBudgetField, true> = {
+  thinking_token_budget: true,
+  thinking_budget: true,
+  thinking_budget_tokens: true,
+}
+
+/** The reasoning-budget field spellings a profile may name. */
+export const THINKING_TOKEN_BUDGET_FIELDS = Object.keys(THINKING_TOKEN_BUDGET_FIELD_GATE) as readonly PiAiThinkingTokenBudgetField[]
+
 /** The prompt-cache marker conventions pi-ai accepts. */
 export type PiAiCacheControlFormat = NonNullable<OpenAICompletionsCompat['cacheControlFormat']>
 
@@ -143,6 +156,7 @@ export type PiAiChatTemplateVar = Extract<ChatTemplateKwargValue, { $var: string
 const CHAT_TEMPLATE_VAR_GATE: Record<PiAiChatTemplateVar, true> = {
   'thinking.enabled': true,
   'thinking.effort': true,
+  'thinking.budget': true,
 }
 
 /** The request-state placeholders a profile may name. */
@@ -229,6 +243,8 @@ const COMPLETIONS_COMPAT_GATE = {
   chatTemplateKwargs: 'offer',
   chatTemplateArgs: 'offer',
   supportsThinkingTokenBudget: 'offer',
+  thinkingTokenBudgetField: 'offer',
+  vllmPriority: 'offer',
   supportsStrictMode: 'offer',
   cacheControlFormat: 'offer',
   supportsLongCacheRetention: 'offer',
@@ -244,6 +260,7 @@ const COMPLETIONS_COMPAT_GATE = {
 /** Disposition of every `OpenAIResponsesCompat` field; a drift gate like the one above. */
 const RESPONSES_COMPAT_GATE = {
   supportsDeveloperRole: 'offer',
+  supportsMaxOutputTokens: 'offer',
   supportsStrictMode: 'offer',
   supportsLongCacheRetention: 'offer',
   sessionAffinityFormat: 'withhold',
@@ -264,6 +281,8 @@ const ANTHROPIC_COMPAT_GATE = {
   supportsStrictTools: 'offer',
   sendSessionAffinityHeaders: 'withhold',
   supportsToolReferences: 'withhold',
+  supportsMidConvoEffort: 'withhold',
+  allowedFallbackModels: 'withhold',
 } as const satisfies Record<keyof AnthropicMessagesCompat, CompatDisposition>
 
 /** Disposition of every `BedrockCompat` field; a drift gate like the one above. */
@@ -377,8 +396,14 @@ export interface PiAiCompatProfile {
   chatTemplateKwargs?: NonNullable<OpenAICompletionsCompat['chatTemplateKwargs']>
   /** Arguments sent as `chat_template_args` under the `baseten` thinking format; `openai-completions`. */
   chatTemplateArgs?: NonNullable<OpenAICompletionsCompat['chatTemplateArgs']>
-  /** Whether the endpoint accepts `thinking_token_budget` to cap vLLM reasoning; `openai-completions`. */
+  /** Alias for `thinkingTokenBudgetField: "thinking_token_budget"`; an explicit field wins. `openai-completions`. */
   supportsThinkingTokenBudget?: boolean
+  /** Request field carrying the reasoning budget from `thinkingBudgets`; omitted unless configured. `openai-completions`. */
+  thinkingTokenBudgetField?: PiAiThinkingTokenBudgetField
+  /** vLLM scheduler `priority`; lower runs earlier, and the server must enable priority scheduling. Omitted unless configured. */
+  vllmPriority?: number
+  /** Whether the endpoint accepts `max_output_tokens`; `false` omits it. The three Responses protocols. */
+  supportsMaxOutputTokens?: boolean
   /**
    * Whether the endpoint accepts `strict` in tool definitions;
    * `openai-completions`, the three Responses protocols, `bedrock-converse-stream`.

packages/llm/llm-pi-ai/src/config.ts

@@ -29,6 +29,7 @@ import {
   resolveRouteModels,
   SUPPORTED_THINKING_FORMATS,
   THINKING_LEVELS,
+  THINKING_TOKEN_BUDGET_FIELDS,
 } from './catalog.ts'
 import type {
   PiAiCompatProfile,
@@ -261,6 +262,9 @@ const compatProfile: z<PiAiCompatProfile> = z.object({
   chatTemplateKwargs: z.dict(chatTemplateKwarg),
   chatTemplateArgs: z.dict(chatTemplateKwarg),
   supportsThinkingTokenBudget: z.boolean(),
+  thinkingTokenBudgetField: z.union(THINKING_TOKEN_BUDGET_FIELDS),
+  vllmPriority: z.number().step(1),
+  supportsMaxOutputTokens: z.boolean(),
   supportsStrictMode: z.boolean(),
   cacheControlFormat: z.union(CACHE_CONTROL_FORMATS),
   supportsLongCacheRetention: z.boolean(),

packages/llm/llm-pi-ai/src/replay.ts

@@ -27,6 +27,7 @@ export interface PiAiReplayResponse {
   model: string
   responseModel?: string
   responseId?: string
+  providerThinkingLevel?: string
   stopReason: AssistantMessage['stopReason']
 }
 
@@ -78,6 +79,7 @@ export function toPiReplayState(message: AssistantMessage): ReplayEnvelope {
     model: message.model,
     ...message.responseModel === undefined ? {} : { responseModel: message.responseModel },
     ...message.responseId === undefined ? {} : { responseId: message.responseId },
+    ...message.providerThinkingLevel === undefined ? {} : { providerThinkingLevel: message.providerThinkingLevel },
     stopReason: message.stopReason,
   }
   return {
@@ -123,6 +125,7 @@ function readReplayState(value: unknown): PiAiReplayState {
   }
   if (response['responseModel'] !== undefined && typeof response['responseModel'] !== 'string') return invalidReplay('responseModel must be a string')
   if (response['responseId'] !== undefined && typeof response['responseId'] !== 'string') return invalidReplay('responseId must be a string')
+  if (response['providerThinkingLevel'] !== undefined && typeof response['providerThinkingLevel'] !== 'string') return invalidReplay('providerThinkingLevel must be a string')
   const blocks = envelope['blocks']
   if (!Array.isArray(blocks)) return invalidReplay('blocks must be an array')
   for (const [index, value] of blocks.entries()) {
@@ -215,6 +218,7 @@ function replayedAssistant(message: Message, source: ModelMessageSource, rawStat
     model: state.response.model,
     ...state.response.responseModel === undefined ? {} : { responseModel: state.response.responseModel },
     ...state.response.responseId === undefined ? {} : { responseId: state.response.responseId },
+    ...state.response.providerThinkingLevel === undefined ? {} : { providerThinkingLevel: state.response.providerThinkingLevel },
     usage: emptyPiUsage(),
     stopReason: state.response.stopReason,
     timestamp: 0,

packages/llm/llm-pi-ai/tests/catalog.spec.ts

@@ -795,19 +795,17 @@ describe('compat switches', () => {
   })
 
   it('skips models of other protocols on a mixed route instead of failing them', () => {
-    // xai ships both completions and responses models, so a route-level switch
-    // must land on the former without invalidating the latter.
-    const catalog = getBuiltinModels('xai') as readonly Model<Api>[]
+    const catalog = getBuiltinModels('opencode') as readonly Model<Api>[]
     const completions = catalog.find(model => model.api === 'openai-completions')
     const responses = catalog.find(model => model.api === 'openai-responses')
-    if (completions === undefined || responses === undefined) throw new Error('xai no longer ships a mixed catalog')
+    if (completions === undefined || responses === undefined) throw new Error('opencode ships no mixed catalog')
 
     const models = modelsOf({
-      xai: {
+      opencode: {
         compat: { supportsReasoningEffort: false },
         models: [{ id: completions.id }, { id: responses.id }],
       },
-    }, 'xai')
+    }, 'opencode')
 
     expect((models.get(completions.id)?.compat as OpenAICompletionsCompat).supportsReasoningEffort).toBe(false)
     expect(models.get(responses.id)?.compat).toEqual(responses.compat)
@@ -876,18 +874,18 @@ describe('compat switches', () => {
   })
 
   it('lands each route switch only on the models whose protocol declares it', () => {
-    const catalog = getBuiltinModels('xai') as readonly Model<Api>[]
+    const catalog = getBuiltinModels('opencode') as readonly Model<Api>[]
     const completions = catalog.find(model => model.api === 'openai-completions')
     const responses = catalog.find(model => model.api === 'openai-responses')
-    if (completions === undefined || responses === undefined) throw new Error('xai no longer ships a mixed catalog')
+    if (completions === undefined || responses === undefined) throw new Error('opencode ships no mixed catalog')
 
     const models = modelsOf({
-      xai: {
+      opencode: {
         // Both protocols take the first switch; only completions takes the second.
         compat: { supportsDeveloperRole: false, thinkingFormat: 'openai' },
         models: [{ id: completions.id }, { id: responses.id }],
       },
-    }, 'xai')
+    }, 'opencode')
 
     const onCompletions = models.get(completions.id)?.compat as OpenAICompletionsCompat
     expect(onCompletions.supportsDeveloperRole).toBe(false)

packages/llm/llm-pi-ai/tests/compat-upgrade.spec.ts

@@ -0,0 +1,61 @@
+import { describe, expect, it } from 'vitest'
+import { Config, resolveProfiles } from '../src/config.ts'
+import { THINKING_TOKEN_BUDGET_FIELDS } from '../src/catalog.ts'
+
+function configured(compat: Record<string, unknown>, api = 'openai-completions'): Config {
+  return Config({ providers: { gateway: {
+    api, baseURL: 'https://gateway.test/v1', models: [{ id: 'model' }], compat,
+  } } })
+}
+
+function resolved(compat: Record<string, unknown>, api = 'openai-completions') {
+  return resolveProfiles(configured(compat, api).providers).get('gateway')?.piProvider.getModels()[0]?.compat
+}
+
+describe('pi-ai gateway compatibility declarations', () => {
+  it.each(THINKING_TOKEN_BUDGET_FIELDS)('preserves reasoning-budget field %s', (field) => {
+    expect(resolved({ thinkingTokenBudgetField: field })).toMatchObject({ thinkingTokenBudgetField: field })
+  })
+
+  it.each([-2, 0, 2])('preserves scheduler priority %s', (priority) => {
+    expect(resolved({ vllmPriority: priority })).toMatchObject({ vllmPriority: priority })
+  })
+
+  it('preserves a disabled Responses output-cap parameter', () => {
+    expect(resolved({ supportsMaxOutputTokens: false }, 'openai-responses'))
+      .toMatchObject({ supportsMaxOutputTokens: false })
+  })
+
+  it.each(['chatTemplateKwargs', 'chatTemplateArgs'])('accepts thinking.budget in %s', (field) => {
+    const value = { budget: { $var: 'thinking.budget' } }
+    expect(resolved({ [field]: value })).toMatchObject({ [field]: value })
+  })
+
+  it.each([
+    { thinkingTokenBudgetField: 'unknown' },
+    { vllmPriority: 'high' },
+    { vllmPriority: 0.5 },
+    { supportsMaxOutputTokens: 'false' },
+    { chatTemplateKwargs: { budget: { $var: 'thinking.unknown' } } },
+    { chatTemplateArgs: { budget: { $var: 'thinking.unknown' } } },
+  ])('rejects invalid schema values: %j', (compat) => {
+    expect(() => configured(compat)).toThrow()
+  })
+
+  it.each([
+    { thinkingTokenBudgetField: 'thinking_budget' },
+    { vllmPriority: 0 },
+    { supportsMaxOutputTokens: false },
+  ])('rejects fields on a protocol that cannot consume them: %j', (compat) => {
+    expect(() => resolved(compat, 'anthropic-messages')).toThrow(/compat/)
+  })
+
+  it.each(['supportsMidConvoEffort', 'allowedFallbackModels'])('withholds catalog-owned %s', (field) => {
+    expect(() => resolved({ [field]: true }, 'anthropic-messages'))
+      .toThrow(/which is not configurable here/)
+  })
+
+  it('keeps generic additions absent unless configured', () => {
+    expect(resolved({})).toBeUndefined()
+  })
+})

packages/llm/llm-pi-ai/tests/convert.spec.ts

@@ -415,6 +415,45 @@ describe('toPiContext', () => {
     })
   })
 
+  it.each(['high', ''])('preserves provider thinking level %j through durable projection and replay', (providerThinkingLevel) => {
+    const state = toPiReplayState(assistant({
+      api: 'anthropic-messages',
+      provider: 'anthropic',
+      model: 'claude-opus-5',
+      providerThinkingLevel,
+      content: [{ type: 'text', text: 'done' }],
+    }))
+    expect(state).toEqual({
+      response: {
+        kind: 'pi-ai',
+        version: 2,
+        api: 'anthropic-messages',
+        provider: 'anthropic',
+        model: 'claude-opus-5',
+        providerThinkingLevel,
+        stopReason: 'stop',
+      },
+      blocks: [{ type: 'text' }],
+    })
+    const replayState: unknown = JSON.parse(JSON.stringify(state))
+    const onDegrade = vi.fn()
+    const context = toPiContext({
+      provider: 'anthropic',
+      model: 'claude-opus-5',
+      messages: [createMessage({
+        role: 'assistant',
+        content: [{ type: 'text', text: 'done' }],
+        source: { kind: 'model', provider: 'anthropic', model: 'claude-opus-5', replayState },
+      })],
+    }, undefined, onDegrade)
+    expect(context.messages[0]).toMatchObject({
+      api: 'anthropic-messages',
+      providerThinkingLevel,
+      content: [{ type: 'text', text: 'done' }],
+    })
+    expect(onDegrade).not.toHaveBeenCalled()
+  })
+
   it('replays all native block kinds when optional metadata is absent', () => {
     const state = toPiReplayState(assistant({
       content: [
@@ -450,6 +489,8 @@ describe('toPiContext', () => {
     })
     expect(context.messages[0]).not.toHaveProperty('responseModel')
     expect(context.messages[0]).not.toHaveProperty('responseId')
+    expect(state.response).not.toHaveProperty('providerThinkingLevel')
+    expect(context.messages[0]).not.toHaveProperty('providerThinkingLevel')
   })
 
   it('degrades unsupported replay-state versions to provider-neutral history', () => {
@@ -612,6 +653,8 @@ describe('toPiContext', () => {
     ['unknown stop reason', { ...validReplay, response: { ...validResponse, stopReason: 'pause' } }, 'unknown stopReason'],
     ['non-string response model', { ...validReplay, response: { ...validResponse, responseModel: 1 } }, 'responseModel must be a string'],
     ['non-string response id', { ...validReplay, response: { ...validResponse, responseId: 1 } }, 'responseId must be a string'],
+    ['non-string provider thinking level', { ...validReplay, response: { ...validResponse, providerThinkingLevel: 1 } }, 'providerThinkingLevel must be a string'],
+    ['null provider thinking level', { ...validReplay, response: { ...validResponse, providerThinkingLevel: null } }, 'providerThinkingLevel must be a string'],
     ['missing blocks', { response: validResponse }, 'blocks must be an array'],
     ['non-array blocks', { ...validReplay, blocks: 'text' }, 'blocks must be an array'],
     ['number block', { ...validReplay, blocks: [1] }, 'block 0 must be an object'],

pnpm-lock.yaml

@@ -6255,8 +6255,8 @@ importers:
         specifier: link:../../../vendor/schemastery
         version: link:../../../vendor/schemastery
       '@earendil-works/pi-ai':
-        specifier: ^0.84.2
-        version: 0.84.2(@modelcontextprotocol/sdk@1.29.0(zod@4.4.3))(ws@8.21.0)(zod@4.4.3)
+        specifier: ^0.85.1
+        version: 0.85.1(@modelcontextprotocol/sdk@1.29.0(zod@4.4.3))(ws@8.21.0)(zod@4.4.3)
     devDependencies:
       '@deepseek-ai/cordis':
         specifier: workspace:^
@@ -10565,8 +10565,8 @@ packages:
       '@modelcontextprotocol/sdk': ^1.29.0
       zod: ^4.0.0
 
-  '@anthropic-ai/sdk@0.91.1':
-    resolution: {integrity: sha512-LAmu761tSN9r66ixvmciswUj/ZC+1Q4iAfpedTfSVLeswRwnY3n2Nb6Tsk+cLPP28aLOPWeMgIuTuCcMC6W/iw==}
+  '@anthropic-ai/sdk@0.123.0':
+    resolution: {integrity: sha512-Y9oX9mPNGZClHQOFqrWRk43Srcu/UHuPq3rfxxOq7JgW0gi+lJA2MAOK4Ul3k/+AUrwRWFJvd0tK3oC0Pw25dw==}
     hasBin: true
     peerDependencies:
       zod: ^3.25.0 || ^4.0.0
@@ -10897,13 +10897,13 @@ packages:
       search-insights:
         optional: true
 
-  '@earendil-works/pi-ai@0.84.2':
-    resolution: {integrity: sha512-6MzsrYIYNVlE7SfpbL2yYb67Qo58p/7Q+xWG1RZvoX1P80aRCHSod2/13aFpxkow1lPO2LEh3c495J0Gwmyjig==}
+  '@earendil-works/pi-ai@0.85.1':
+    resolution: {integrity: sha512-+VgVIJDkDO2efYJKEEqvPTH4zmnIaXdAppGbO+vKFA9qy5PdhFiAenuFAkU+oiCSfOC4dMHDyrjdQeL4ZoC5CQ==}
     engines: {node: '>=22.19.0'}
     hasBin: true
 
-  '@earendil-works/pi-telemetry@0.84.2':
-    resolution: {integrity: sha512-wg5caea7uIv1BHRBm2Y116RvFG4oSAiP5qk9tA2463PDGIr4K8M1Ceyyg5DOpF/shUUl0gk826yQJAeAcHYB9g==}
+  '@earendil-works/pi-telemetry@0.85.1':
+    resolution: {integrity: sha512-Bg/YN6kA7Swja/NQxka8xFdecb4E/auIEGF2G5A25EaQXhRnPj300/7/KpgsDDMYUzHTDAv4RyUxaQPJKW81Rw==}
     engines: {node: '>=22.19.0'}
 
   '@emnapi/core@1.10.0':
@@ -11930,10 +11930,6 @@ packages:
     resolution: {integrity: sha512-CmVa4ImJ+ynfrPMNaAXHET6Bhb44SwzmfyVJFq9ni2jgXJR/l7C6gfVFddNmHP+ZOkP9cf4f9DBe68qVLTHc9w==}
     engines: {node: '>=8.0.0'}
 
-  '@opentelemetry/api@1.9.0':
-    resolution: {integrity: sha512-3giAOQvZiH5F9bMlMiv8+GSPMeqg0dbaeo58/0SlA9sxSqZhnUtxzX9/2FzyhS9sWQf5S0GJE0AKBrFqjpeYcg==}
-    engines: {node: '>=8.0.0'}
-
   '@opentelemetry/api@1.9.1':
     resolution: {integrity: sha512-gLyJlPHPZYdAk1JENA9LeHejZe1Ti77/pTeFm/nMXmQH/HFZlcS/O2XJB+L8fkbrNSqhdtlvjBVjxwUYanNH5Q==}
     engines: {node: '>=8.0.0'}
@@ -12741,6 +12737,9 @@ packages:
     resolution: {integrity: sha512-R8Rdn8Hy72KKcebgLiv8jQcQkXoLMOGGv5uI1/k0l+snqkOzQ1R0ChUBCxWMlBsFMekWjq0wRudIweFs7sKT5A==}
     engines: {node: '>=14.0.0'}
 
+  '@stablelib/base64@1.0.1':
+    resolution: {integrity: sha512-1bnPQqSxSuc3Ii6MhBysoWCg58j97aUjuCSZrGSmDxNqtytIi0k8utUenAwTZN4V5mXXYGsVUI9zeBqy+jBOSQ==}
+
   '@standard-schema/spec@1.1.0':
     resolution: {integrity: sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w==}
 
@@ -14124,6 +14123,9 @@ packages:
   fast-levenshtein@2.0.6:
     resolution: {integrity: sha512-DCXu6Ifhqcks7TZKY3Hxp3y6qphY5SJZmrWMDrKcERSOXWQdMhU9Ig/PYrzyw/ul9jOIyh0N4M0tbC5hodg8dw==}
 
+  fast-sha256@1.3.0:
+    resolution: {integrity: sha512-n11RGP/lrWEFI/bWdygLxhI+pVeo1ZYIVwvvPkW7azl/rOy+F3HYRZ2K5zeE9mmkhQppyv9sQFx0JM9UabnpPQ==}
+
   fast-uri@3.1.3:
     resolution: {integrity: sha512-i70LwGWUduXqzicKXWshooq+sWL1K3WUU5rKZNG/0i3a1OSoX3HqhH5WbWwTmqWfor4urUakGPiRQcleRZTwOg==}
 
@@ -15717,6 +15719,9 @@ packages:
   stackback@0.0.2:
     resolution: {integrity: sha512-1XMJE5fQo1jGH6Y/7ebnwPOBEkIEnT4QF32d5R1+VXdXveM0IBMJt8zfaxX1P3QhVwrYe+576+jkANtSS2mBbw==}
 
+  standardwebhooks@1.1.1:
+    resolution: {integrity: sha512-bCbX9ZEyFkWPsRz7Bl3NuQUJohmwGSev/yhr7vhaGPlc4AfIrspIRa6cPTBuI1ItmrTDJ4d/S2hCsfe4+vQGnQ==}
+
   statuses@2.0.2:
     resolution: {integrity: sha512-DvEy55V3DB7uknRo+4iOGT5fP1slR8wQohVdknigZPMpMstaKJQWhwiYBACJE3Ul2pTnATihhBYnRhZQHGBiRw==}
     engines: {node: '>= 0.8'}
@@ -16549,9 +16554,10 @@ snapshots:
       '@anthropic-ai/claude-agent-sdk-win32-arm64': 0.3.241
       '@anthropic-ai/claude-agent-sdk-win32-x64': 0.3.241
 
-  '@anthropic-ai/sdk@0.91.1(zod@4.4.3)':
+  '@anthropic-ai/sdk@0.123.0(zod@4.4.3)':
     dependencies:
       json-schema-to-ts: 3.1.1
+      standardwebhooks: 1.1.1
     optionalDependencies:
       zod: 4.4.3
 
@@ -17016,13 +17022,12 @@ snapshots:
     transitivePeerDependencies:
       - '@algolia/client-search'
 
-  '@earendil-works/pi-ai@0.84.2(@modelcontextprotocol/sdk@1.29.0(zod@4.4.3))(ws@8.21.0)(zod@4.4.3)':
+  '@earendil-works/pi-ai@0.85.1(@modelcontextprotocol/sdk@1.29.0(zod@4.4.3))(ws@8.21.0)(zod@4.4.3)':
     dependencies:
-      '@anthropic-ai/sdk': 0.91.1(zod@4.4.3)
+      '@anthropic-ai/sdk': 0.123.0(zod@4.4.3)
       '@aws-sdk/client-bedrock-runtime': 3.1048.0
-      '@earendil-works/pi-telemetry': 0.84.2
+      '@earendil-works/pi-telemetry': 0.85.1
       '@google/genai': 1.52.0(@modelcontextprotocol/sdk@1.29.0(zod@4.4.3))
-      '@opentelemetry/api': 1.9.0
       '@smithy/node-http-handler': 4.7.3
       http-proxy-agent: 7.0.2
       https-proxy-agent: 7.0.6
@@ -17037,7 +17042,7 @@ snapshots:
       - ws
       - zod
 
-  '@earendil-works/pi-telemetry@0.84.2': {}
+  '@earendil-works/pi-telemetry@0.85.1': {}
 
   '@emnapi/core@1.10.0':
     dependencies:
@@ -17802,8 +17807,6 @@ snapshots:
     dependencies:
       '@opentelemetry/api': 1.9.1
 
-  '@opentelemetry/api@1.9.0': {}
-
   '@opentelemetry/api@1.9.1': {}
 
   '@opentelemetry/core@2.10.0(@opentelemetry/api@1.9.1)':
@@ -18362,6 +18365,8 @@ snapshots:
       '@smithy/util-buffer-from': 2.2.0
       tslib: 2.8.1
 
+  '@stablelib/base64@1.0.1': {}
+
   '@standard-schema/spec@1.1.0': {}
 
   '@stylistic/eslint-plugin@5.10.0(eslint@10.5.0(jiti@2.7.0))':
@@ -19942,6 +19947,8 @@ snapshots:
 
   fast-levenshtein@2.0.6: {}
 
+  fast-sha256@1.3.0: {}
+
   fast-uri@3.1.3: {}
 
   fast-xml-builder@1.2.0:
@@ -21830,6 +21837,11 @@ snapshots:
 
   stackback@0.0.2: {}
 
+  standardwebhooks@1.1.1:
+    dependencies:
+      '@stablelib/base64': 1.0.1
+      fast-sha256: 1.3.0
+
   statuses@2.0.2: {}
 
   std-env@4.1.0: {}

pnpm-workspace.yaml

@@ -52,7 +52,8 @@ allowBuilds:
 minimumReleaseAgeExclude:
   # Fresh pi-ai releases carry the model catalog updates that are the whole
   # point of bumping it; waiting out the release age would defeat that.
-  - '@earendil-works/pi-ai@0.84.2'
+  - '@earendil-works/pi-ai@0.85.1'
+  - '@earendil-works/pi-telemetry@0.85.1'
   - node-addon-native-custom-loader@0.1.4
   - node-addon-require-builtin-darwin-arm64@0.1.4
   - node-addon-require-builtin-darwin-x64@0.1.4

snapshots/session/pi-ai-replay-metadata/session.v2.jsonl

@@ -0,0 +1,16 @@
+{"type":"session","version":2,"id":"{{session:1}}","createdAt":1783600629539,"cwd":"{{cwd}}","isSeeded":false,"delegationDepth":0}
+{"type":"permission/preset","data":{"preset":"danger-full-access"}}
+{"type":"sandbox/mode","data":{"mode":"danger-full-access"}}
+{"type":"approval/policy","data":{"policy":"never"}}
+{"type":"agent/inbox/spliced","data":{"target":"next-turn","start":0,"inserted":[{"content":[{"type":"text","text":"Reply with exactly the word: PONG. Do not use any tools."}],"source":{"kind":"user"},"role":"user","id":"{{message:1}}"}]}}
+{"type":"turn/start","data":{"turn":1}}
+{"type":"agent/inbox/spliced","data":{"target":"next-turn","start":0,"removedCount":1,"inserted":[]}}
+{"type":"step/start","data":{"turn":1,"step":1}}
+{"type":"user/message","data":{"content":[{"type":"text","text":"Reply with exactly the word: PONG. Do not use any tools."}],"source":{"kind":"user"},"role":"user","id":"{{message:1}}"},"surfaceOp":"append"}
+{"type":"user/message","data":{"content":[{"type":"text","text":"Current runtime context. This snapshot supersedes earlier runtime-context snapshots.\n\nCurrent DSH file policy: danger-full-access. The DSH file sandbox does not restrict file modifications by available operations.\n\nApproval prompts are disabled in this session: actions that require approval are rejected automatically — do not request sandbox escalation (do not set `sandbox_permissions`)."}],"source":{"kind":"plugin","plugin":"@deepseek-ai/dsh-system-prompt","form":"snapshot","sections":[{"name":"sandbox:policy","text":"Current DSH file policy: danger-full-access. The DSH file sandbox does not restrict file modifications by available operations."},{"name":"approval:policy","text":"Approval prompts are disabled in this session: actions that require approval are rejected automatically — do not request sandbox escalation (do not set `sandbox_permissions`)."}]},"role":"user","id":"{{message:2}}"},"surfaceOp":"append"}
+{"type":"session/title","data":{"title":"Reply with exactly the word:","messageSeqs":[7],"source":{"kind":"fallback"}}}
+{"type":"request/header","data":{"header":{"config":{"provider":"deepseek-official","model":"deepseek-v4-flash"},"system":"{{system}}","tools":"{{tools}}"},"reason":"initial"}}
+{"type":"request/context","data":{"provider":"deepseek-official","model":"deepseek-v4-flash"}}
+{"type":"assistant/message","data":{"turn":1,"step":1,"message":{"role":"assistant","content":[{"type":"text","text":"PONG"}],"source":{"kind":"model","provider":"deepseek-official","model":"deepseek-v4-flash","replayState":{"response":{"kind":"pi-ai","version":2,"api":"openai-completions","provider":"deepseek-official","model":"deepseek-v4-flash","providerThinkingLevel":"high","stopReason":"stop"},"blocks":[{"type":"text"}]}},"id":"{{message:3}}"},"stream":[{"type":"chunk","time":0,"chunk":{"type":"block-start","index":0,"blockType":"text"}},{"type":"text-chunks","time0":1788618852857,"index":0,"dt":[],"texts":["PONG"]},{"type":"chunk","time":0,"chunk":{"type":"block-end","index":0,"block":{"type":"text","text":"PONG"}}},{"type":"chunk","time":0,"chunk":{"type":"finish","reason":{"kind":"stop"},"replayState":{"response":{"kind":"pi-ai","version":2,"api":"openai-completions","provider":"deepseek-official","model":"deepseek-v4-flash","providerThinkingLevel":"high","stopReason":"stop"},"blocks":[{"type":"text"}]}}}]},"surfaceOp":"append"}
+{"type":"step/end","data":{"turn":1,"step":1}}
+{"type":"turn/end","data":{"turn":1,"reason":{"kind":"completed"}}}

snapshots/session/pi-ai-replay-metadata/snapshot.yml

@@ -0,0 +1,7 @@
+version: 1
+scenario: pi-ai-replay-metadata
+profile: headless
+composition: default
+recording: authored
+header:
+  class: default