@betterlu/agentlu 1.11.0 → 2.0.0

package/README.md

@@ -10,7 +10,19 @@ const client = new OpenAI({ baseURL: "http://localhost:8912/v1", apiKey });
 const r = await client.responses.create({ model: "fast", input: "hi" });
 ```
 
-This client adds get-or-create registration, typed `runJson` / `runText`, use-case catalogs, the two `/v1` tiers (`llm.*` / `agent.*`), error classification (`err.kind`, `isUnavailable`), background `wait`, bulk uploads, client-side `timeoutMs`, and auto-retry on `409 thread_busy`.
+This client adds get-or-create registration, typed `runJson` / `runText`, use-case catalogs, error classification (`err.kind`, `isUnavailable`), background `wait`, bulk uploads, client-side `timeoutMs`, and auto-retry on `409 thread_busy`.
+
+## SDK 2.0 surface
+
+SDK 2.0 removes the unused direct-call `llm.*` and `typedStream` wrappers, the
+Briefs publishing/reader helpers, and the discovery helpers for agents, models,
+tools, health and the current client. The matching `/v1/briefs`, `/v1/agents`,
+`/v1/models`, `/v1/tools` and `/v1/me` routes are no longer served. Use the
+ordinary `runText` / `runJson` methods with `agent: "direct"` for direct calls.
+
+`/v1/health` remains available and `waitForReady` still polls it; `/v1/threads`
+and the SDK `threads` helpers remain available. The server also retains
+`/v1/chat/completions` for OpenAI-compatible callers.
 
 ## Install
 
@@ -35,10 +47,11 @@ console.log(response.output_text);
 
 ## Two tiers
 
-`/v1` has two documented tiers, and 1.8 exposes both as namespaces. Pick one and
-stay there (see the `agentlu-api` skill for the full contract):
+`/v1` supports direct and agent runs through the same run methods. Use
+`agent: "direct"` for one profile-free call or name an agent for its profile
+and tools (see the `agentlu-api` skill for the full contract):
 
-| | `alu.llm.*` (direct) | `alu.agent.*` (agent) |
+| | Direct | Agent |
 |---|---|---|
 | What runs | one profile-free provider call | profile + tool families + agent loop |
 | Session | none (`store: false`) | a real thread you can continue |
@@ -46,21 +59,16 @@ stay there (see the `agentlu-api` skill for the full contract):
 | Use it for | extraction, classification, short answers | research, phone/browser work, context |
 
 ```ts
-// Direct tier.
-const { text } = await alu.llm.text("Name three colours.");
-const { data } = await alu.llm.json<{ items: string[] }>(
-  { type: "object", properties: { items: { type: "array", items: { type: "string" } } }, required: ["items"] },
-  "two milk, one bread",
-);
-for await (const event of alu.llm.stream("Count to three")) console.log(event.type);
+// Direct tier: explicit opt-in, no tools or session.
+const { text } = await alu.runText({ agent: "direct", input: "Name three colours." });
 
 // Agent tier (default agent `api`; pin fast/thorough/… or "<app>/<name>").
 const run = await alu.agent.run({ agent: "fast", input: "what changed in node 24?" });
 const done = await alu.agent.background({ input: "research this deeply" }, { timeoutMs: 600_000 });
 ```
 
-`alu.run` / `runJson` / `runText` / `stream` remain the agent-tier aliases they
-always were - nothing breaks by upgrading.
+`alu.run` / `runJson` / `runText` / `stream` remain available for Responses
+runs, including calls that explicitly select the direct tier.
 
 ## Errors: `kind` and fail-open
 
@@ -72,7 +80,7 @@ saturation, timeout, rate limit):
 import { isUnavailable } from "@betterlu/agentlu";
 
 try {
-  await alu.llm.text(input);
+  await alu.runText({ agent: "direct", input });
 } catch (err) {
   if (isUnavailable(err)) return fallbackAnswer();   // degrade gracefully
   throw err;
@@ -236,7 +244,7 @@ Plain text (throws unless `status === "completed"` with non-empty `output_text`)
 const { text } = await ai.tag.runText({ input: "summarize this" });
 ```
 
-`timeoutMs` aborts the HTTP attempt and, when `timeout_ms` is omitted, is also sent as the server-side run budget. Constructor `new AgentLu({ timeoutMs: 30_000 })` is the default for calls that omit it.
+`timeoutMs` aborts the HTTP attempt and, when `timeout_ms` is omitted, is also sent as the server-side run budget. Constructor `new AgentLu({ timeoutMs: 30_000 })` is the default for calls that omit it. The server rejects `timeout_ms` below 15 000 ms and `max_output_tokens` below 256 with a 400 naming the minimum; `/v1/chat/completions` applies the same minimum to `max_tokens` / `max_completion_tokens`. Registered defaults below either minimum are clamped up with a once-per-client warning.
 
 Upload-and-run:
 
@@ -295,8 +303,8 @@ if (answers.team.type === "choice") return lookupOrder(answers.team.choice);
 | `score` | `score`, `confidence`, `legend`, `probabilities` per step |
 
 `decisions.ask()` is the same call returning only the answers map.
-`decisions.askOrNull()` is fail-open (`null` on unconfigured/error). Discover the
-model in `GET /v1/models` (`harness: "systemone"`); it is only callable on
+`decisions.askOrNull()` is fail-open (`null` on unconfigured/error). The
+decision model is only callable on
 `POST /v1/decisions`, sync, no streaming. A missing key surfaces as HTTP 503
 `systemone_unconfigured` (not retried by the SDK). Use-case runners expose the
 same `decisions` helpers on their own key.
@@ -371,7 +379,7 @@ feature the SDK lacks: add it here and publish).
 2. Bump the version: `npm version <x.y.z> --no-git-tag-version` (minor for new
    methods/fields, patch for fixes, major for breaking changes).
 3. Add a `CHANGELOG.md` entry for that version.
-4. Verify: `npm run build` here and `npx vitest run tests/sdk` from the repo root.
+4. Verify: `npm run build` here and `npm test -- tests/sdk` from the repo root.
 5. Publish: `npm publish` (runs `prepublishOnly` → build; npm user `betterlu`).
    Check with `npm view @betterlu/agentlu version`.
 6. Commit the bump with the change, then update consumers to the new range

package/dist/generated.d.ts

@@ -8,77 +8,6 @@
  * contract. They describe the bytes on the wire (snake_case); the SDK's own
  * types are the ergonomic view over them.
  */
-export interface WireBriefPublishInput {
-    "source": string;
-    "key": string;
-    "kind": "edition" | "report" | "note" | "story";
-    "audience": "larry" | "misa" | "both";
-    "title": string;
-    "dek"?: string;
-    "body"?: string;
-    "bodyFormat"?: "markdown" | "html";
-    "entries"?: Array<{
-        "title": string;
-        "url"?: string;
-        "image"?: string;
-        "chips"?: string[];
-        "body"?: string;
-    }>;
-    "images"?: string[];
-    "url"?: string;
-    "keep"?: boolean;
-    "ttlDays"?: number;
-    "collection"?: string;
-    "sourceLabel"?: string;
-    "sourceEmoji"?: string;
-    "coverArt"?: boolean;
-    "jobRunId"?: string;
-    "artifactId"?: string;
-    "publishedAt"?: string;
-}
-export interface WireBriefDetail {
-    "id": string;
-    "source": string;
-    "sourceLabel": string;
-    "sourceEmoji": string | null;
-    "kind": "edition" | "report" | "note" | "story";
-    "audience": "larry" | "misa" | "both";
-    "title": string;
-    "dek": string | null;
-    "coverUrl": string | null;
-    "coverArtStatus": "none" | "pending" | "ready" | "failed";
-    "edition": {
-        "hero": string;
-        "highlights": string[];
-    } | null;
-    "keep": boolean;
-    "collection": string | null;
-    "readingMinutes": number | null;
-    "publishedAt": string;
-    "bodyFormat": "markdown" | "html";
-    "bodyHtml": string | null;
-    "frameUrl": string | null;
-    "entries": Array<{
-        "title": string;
-        "url"?: string;
-        "image"?: string;
-        "chips"?: string[];
-        "body"?: string;
-        "bodyHtml"?: string;
-    }> | null;
-    "images": string[];
-    "url": string | null;
-    "jobRunId": string | null;
-    "artifactId": string | null;
-    "expiresAt": string | null;
-}
-export interface WireBriefPublishResult {
-    "id": string;
-    "url": string;
-    "coverUrl": string | null;
-    "coverArtStatus": "none" | "pending" | "ready" | "failed";
-    "updated": boolean;
-}
 export interface WireHouseholdSignalInput {
     "source": string;
     "kind": string;
@@ -97,7 +26,7 @@ export interface WireHouseholdSignalResult {
 export interface WireCreateResponse {
     /** Execution backend. */
     "backend"?: "pi";
-    /** Model id, or an agent name (stock OpenAI clients can only set `model`). Harness-qualified ids run other runtimes: `google/<id>` selects an Antigravity model on the Google agent. See GET /v1/models and GET /v1/models?harness=google. */
+    /** Model id, or an agent name (stock OpenAI clients can only set `model`). Harness-qualified ids run other runtimes: `google/<id>` selects an Antigravity model on the Google agent. */
     "model"?: string;
     "input"?: string | Array<{
         "role": "user" | "assistant" | "system" | "developer";
@@ -130,7 +59,7 @@ export interface WireCreateResponse {
     "conversation"?: string | {
         "id": string;
     };
-    /** Optional and discouraged. Omit it to use the model's own output limit; values here are applied verbatim. If the limit is reached, the HTTP 200 Response has status `incomplete`, reason `max_output_tokens`, and may retain partial output. */
+    /** Optional and discouraged. Minimum 256 tokens; omit it to use the model's own output limit; values here are applied verbatim. A registered client default below the minimum is clamped up. If the limit is reached, the HTTP 200 Response has status `incomplete`, reason `max_output_tokens`, and may retain partial output. */
     "max_output_tokens"?: number;
     "temperature"?: number;
     "text"?: {
@@ -156,13 +85,13 @@ export interface WireCreateResponse {
     "tool_choice"?: "auto" | "none" | "required";
     "metadata"?: Record<string, string>;
     "include"?: Array<"reasoning" | "rounds" | "subagents" | "artifacts">;
-    /** Which agent runs the request. The recommended general machine agent is `api`, which has no persona: `instructions` replace its system prompt. Pin `fast` for the chat-tuned version. `direct` selects the direct tier: the profile-free `api` agent with no tools and no session (see the Tiers section). `household/<slug>` runs a household agent built in the Agents studio (list them at GET /v1/agents). An `<app>/<name>` id runs an agent the caller's app declared at registration. `research` is an alias for `thorough` with Report output (a sourced report artifact). Retired ids (`mealu`, `nestlu`, `kitchen`, `waterlu`, `money`, `swarm`, `storylu`, `google-research`, `google-image`, `android`) return HTTP 400 `unknown_agent` with a pointer to their successor; the Android phone is now the `android` tool family on `assistant`. */
-    "agent"?: "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom" | "direct" | string | "mealu" | "nestlu" | "kitchen" | "waterlu" | "money" | "swarm" | "stage" | "roleplay" | "storylu" | "google-research" | "google-image" | "android" | "research";
+    /** Agent id. Built-in agents, `direct`, the `research` alias, declared app agents (`<app>/<name>`), and household agents (`household/<slug>`) are accepted; unknown ids return HTTP 400 `unknown_agent`. */
+    "agent"?: string;
     /** Thinking level. Explicit `effort: "off"` is honored; other requests are clamped to the selected model's supported capabilities. The SDK does not promise provider benchmark parity. */
     "reasoning"?: {
         "effort"?: "off" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max";
     };
-    /** Wall clock for this run. Default: the use-case tier's (45s instant / 180s standard / 600s deep). */
+    /** Wall clock for this run. Minimum 15000 ms (15 s); a registered client default below it is clamped up. Default: the use-case tier's (45s instant / 180s standard / 600s deep). */
     "timeout_ms"?: number;
     /** Provider attempts for this run (1 = none). Bounds model-chain fallbacks. Default: the use-case tier's (instant 1, standard 2, deep 3) when the client declares a profile; without one the run keeps the pre-tier budget (the direct tier walks its whole chain, the agent tier does not fall back). */
     "attempts"?: number;
@@ -174,7 +103,7 @@ export interface WireCreateResponse {
     "end_user"?: string;
 }
 export interface WireChatCompletionRequest {
-    /** An agentlu model id (GET /v1/models) or `direct`. Omit it to let the model chain pick. An unknown id is HTTP 400 `unknown_model`. */
+    /** An AgentLu model id or `direct`. Omit it to let the model chain pick. An unknown id is HTTP 400 `unknown_model`. */
     "model"?: string;
     /** system/developer messages become the run's instructions; the rest is folded into the prompt with role tags. Tool messages are rejected (unsupported_tool): the direct tier runs no tools. */
     "messages": Array<{
@@ -235,9 +164,9 @@ export interface WireRegisterClientRequest {
     /** Run defaults applied when a request omits the field: request > registered client default > use-case tier > agent/profile default. The built-in `store` default is true. Register one client per use-case (e.g. mealu-enrich) with its system prompt and tools baked in. */
     "defaults"?: {
         "backend"?: "pi";
-        "agent"?: "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom" | "direct" | string | "mealu" | "nestlu" | "kitchen" | "waterlu" | "money" | "swarm" | "stage" | "roleplay" | "storylu" | "google-research" | "google-image" | "android" | "research";
+        "agent"?: "auto" | "fast" | "thorough" | "assistant" | "google" | "api" | "digest" | "herald" | "cradle" | "custom" | "direct" | string | "research";
         "model"?: string;
-        "modelRoute"?: "chat.quick" | "chat.household" | "chat.deep" | "chat.code" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
+        "modelRoute"?: "chat.quick" | "chat.household" | "chat.deep" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
         "instructions"?: string;
         "temperature"?: number;
         "store"?: boolean;
@@ -286,12 +215,11 @@ export interface WireRegisterClientRequest {
 /** Run defaults applied when a request omits the field: request > registered client default > use-case tier > agent/profile default. The built-in `store` default is true. Register one client per use-case (e.g. mealu-enrich) with its system prompt and tools baked in. */
 export interface WireClientDefaults {
     "backend"?: "pi";
-    /** Which agent runs the request. The recommended general machine agent is `api`, which has no persona: `instructions` replace its system prompt. Pin `fast` for the chat-tuned version. `direct` selects the direct tier: the profile-free `api` agent with no tools and no session (see the Tiers section). `household/<slug>` runs a household agent built in the Agents studio (list them at GET /v1/agents). An `<app>/<name>` id runs an agent the caller's app declared at registration. `research` is an alias for `thorough` with Report output (a sourced report artifact). Retired ids (`mealu`, `nestlu`, `kitchen`, `waterlu`, `money`, `swarm`, `storylu`, `google-research`, `google-image`, `android`) return HTTP 400 `unknown_agent` with a pointer to their successor; the Android phone is now the `android` tool family on `assistant`. */
-    "agent"?: "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom" | "direct" | string | "mealu" | "nestlu" | "kitchen" | "waterlu" | "money" | "swarm" | "stage" | "roleplay" | "storylu" | "google-research" | "google-image" | "android" | "research";
-    /** ModelConfig id (GET /v1/models). Stock clients may also use an agent name here; prefer explicit `agent` for machine calls. */
+    "agent"?: "auto" | "fast" | "thorough" | "assistant" | "google" | "api" | "digest" | "herald" | "cradle" | "custom" | "direct" | string | "research";
+    /** Configured model id. Stock clients may also use an agent name here; prefer explicit `agent` for machine calls. */
     "model"?: string;
     /** Task route whose pool order and effort apply to automatic model selection. Mutually exclusive with model. Explicit request models, builtin chat agents, and agent model pins win; app-owned and api agents inherit the client route. */
-    "modelRoute"?: "chat.quick" | "chat.household" | "chat.deep" | "chat.code" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
+    "modelRoute"?: "chat.quick" | "chat.household" | "chat.deep" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
     /** System prompt for every run of this client. */
     "instructions"?: string;
     "temperature"?: number;
@@ -322,8 +250,9 @@ export interface WireClientDefaults {
         "url": string;
         "allowed_tools"?: string[];
     }>;
+    /** Default wall clock when the request omits `timeout_ms`. Values below 15000 ms are clamped up at run time (with a once-per-client warning). */
     "timeoutMs"?: number;
-    /** Default output cap when the request omits `max_output_tokens`. The use-case tier supplies one when this is unset (instant 2048, deep 8192, standard none); requests always win. Honored for defaults written after this release: a value stored earlier stays inert until the app re-declares its defaults (or the console edits them), so a re-enable can never silently cap runs that used to complete. */
+    /** Default output cap when the request omits `max_output_tokens`. Values below 256 tokens are clamped up at run time (with a once-per-client warning). The use-case tier supplies one when this is unset (instant 2048, deep 8192, standard none); requests always win. Honored for defaults written after this release: a value stored earlier stays inert until the app re-declares its defaults (or the console edits them), so a re-enable can never silently cap runs that used to complete. */
     "maxOutputTokens"?: number;
     /** Provider attempts per run (1 = none). Bounds model-chain fallbacks, and is the background queue's retry budget. Defaults to the tier's (instant 1, standard 2, deep 3) when `use_case_profile.priority` is declared; without one the run keeps the pre-tier budget. */
     "attempts"?: number;
@@ -352,7 +281,7 @@ export interface WireAppAgentDeclaration {
 }
 export interface WireCreateThreadRequest {
     "title"?: string;
-    "agent"?: "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom" | "research" | string;
+    "agent"?: "auto" | "fast" | "thorough" | "assistant" | "google" | "api" | "digest" | "herald" | "cradle" | "custom" | "research" | string;
     "model"?: string;
     "person"?: string | null;
     "owner"?: "client" | "person";
@@ -388,7 +317,7 @@ export interface WireResponse {
     "created_at": number;
     "status": "queued" | "in_progress" | "completed" | "failed" | "incomplete" | "cancelled";
     "backend": "pi";
-    "agent": "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom";
+    "agent": "auto" | "fast" | "thorough" | "assistant" | "google" | "api" | "digest" | "herald" | "cradle" | "custom";
     "model": string;
     "conversation"?: {
         "id": string;
@@ -458,7 +387,7 @@ export interface WireResponse {
         "thread_id"?: string | null;
         "retries"?: number;
         "effective_model"?: string | null;
-        "model_route"?: "chat.quick" | "chat.household" | "chat.deep" | "chat.code" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
+        "model_route"?: "chat.quick" | "chat.household" | "chat.deep" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
         "service_tier"?: "standard" | "priority";
         "execution_path"?: string | null;
         "queue_wait_ms"?: number | null;
@@ -578,7 +507,7 @@ export interface WireThread {
     "created_at": number;
     "updated_at": number;
     "title": string;
-    "agent": "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom";
+    "agent": "auto" | "fast" | "thorough" | "assistant" | "google" | "api" | "digest" | "herald" | "cradle" | "custom";
     "model": string | null;
     /** "api" for threads created through /v1; "chat" when a human started it. */
     "origin": string;
@@ -617,9 +546,9 @@ export interface WireClient {
     "allow_person_threads"?: boolean;
     "defaults"?: {
         "backend"?: "pi";
-        "agent"?: "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom" | "direct" | string | "mealu" | "nestlu" | "kitchen" | "waterlu" | "money" | "swarm" | "stage" | "roleplay" | "storylu" | "google-research" | "google-image" | "android" | "research";
+        "agent"?: "auto" | "fast" | "thorough" | "assistant" | "google" | "api" | "digest" | "herald" | "cradle" | "custom" | "direct" | string | "research";
         "model"?: string;
-        "modelRoute"?: "chat.quick" | "chat.household" | "chat.deep" | "chat.code" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
+        "modelRoute"?: "chat.quick" | "chat.household" | "chat.deep" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
         "instructions"?: string;
         "temperature"?: number;
         "store"?: boolean;
@@ -683,51 +612,6 @@ export interface WireError {
         "retryable": boolean;
     };
 }
-export interface WireAgentInfo {
-    /** A builtin agent id, or `household/<slug>` for an agent built in the household studio. */
-    "id": string;
-    "object": "agent";
-    "profile": string;
-    "description": string;
-    "tools": string[];
-    /** Tool families a /v1 caller may request for this agent. */
-    "available_tools": Array<"web_search" | "fetch_content" | "hublu" | "memory" | "image" | "browser" | "android">;
-    "supports_images": boolean;
-    /** Long-running agents are best used with `background: true`. */
-    "long_running": boolean;
-    /** Runtime that serves this agent. */
-    "harness": "pi" | "google";
-    /** Household agents only: the capability ids the agent was built with. */
-    "capabilities"?: string[];
-}
-export interface WireModelInfo {
-    "id": string;
-    "object": "model";
-    "owned_by": string;
-    "label": string;
-    "input": Array<"text" | "image" | "video">;
-    "available": boolean;
-    /** True for the default entry within this model's harness. */
-    "default": boolean;
-    /** Runtime that serves this model. */
-    "harness": "pi" | "google" | "systemone";
-    /** Reasoning levels the model accepts (google models: the values the ACP thought_level option advertises). */
-    "efforts"?: Array<"off" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max">;
-    /** Present on decision models: how to call them. */
-    "decision"?: {
-        "endpoint": "/v1/decisions";
-        "question_types": Array<"noul" | "choice" | "score">;
-    };
-}
-export interface WireToolInfo {
-    "type": "web_search" | "fetch_content" | "hublu" | "memory" | "image" | "browser" | "android";
-    "object": "tool";
-    "description": string;
-    /** Concrete tool names this family expands to. */
-    "expands_to": string[];
-    /** False when the backing service (e.g. hublu) is unreachable right now. */
-    "available": boolean;
-}
 export interface WireStreamEvent {
     "type": string;
     "sequence_number"?: number;
@@ -743,7 +627,7 @@ export interface WireResponseList {
         "created_at": number;
         "status": "queued" | "in_progress" | "completed" | "failed" | "incomplete" | "cancelled";
         "backend": "pi";
-        "agent": "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom";
+        "agent": "auto" | "fast" | "thorough" | "assistant" | "google" | "api" | "digest" | "herald" | "cradle" | "custom";
         "model": string;
         "conversation"?: {
             "id": string;
@@ -808,7 +692,7 @@ export interface WireResponseList {
             "thread_id"?: string | null;
             "retries"?: number;
             "effective_model"?: string | null;
-            "model_route"?: "chat.quick" | "chat.household" | "chat.deep" | "chat.code" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
+            "model_route"?: "chat.quick" | "chat.household" | "chat.deep" | "device" | "background.digest" | "utility" | "memory.extract" | "memory.judge" | "memory.consolidate";
             "service_tier"?: "standard" | "priority";
             "execution_path"?: string | null;
             "queue_wait_ms"?: number | null;
@@ -878,7 +762,7 @@ export interface WireThreadList {
         "created_at": number;
         "updated_at": number;
         "title": string;
-        "agent": "auto" | "fast" | "thorough" | "assistant" | "code" | "google" | "image" | "api" | "digest" | "herald" | "cradle" | "custom";
+        "agent": "auto" | "fast" | "thorough" | "assistant" | "google" | "api" | "digest" | "herald" | "cradle" | "custom";
         "model": string | null;
         "origin": string;
         "end_user"?: string | null;
@@ -929,52 +813,3 @@ export interface WireFileList {
     /** Opaque keyset cursor; null on the last page. */
     "next_cursor": string | null;
 }
-export interface WireAgentList {
-    "object": "list";
-    "data": Array<{
-        "id": string;
-        "object": "agent";
-        "profile": string;
-        "description": string;
-        "tools": string[];
-        "available_tools": Array<"web_search" | "fetch_content" | "hublu" | "memory" | "image" | "browser" | "android">;
-        "supports_images": boolean;
-        "long_running": boolean;
-        "harness": "pi" | "google";
-        "capabilities"?: string[];
-    }>;
-    /** Opaque keyset cursor; null on the last page. */
-    "next_cursor": string | null;
-}
-export interface WireModelList {
-    "object": "list";
-    "data": Array<{
-        "id": string;
-        "object": "model";
-        "owned_by": string;
-        "label": string;
-        "input": Array<"text" | "image" | "video">;
-        "available": boolean;
-        "default": boolean;
-        "harness": "pi" | "google" | "systemone";
-        "efforts"?: Array<"off" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max">;
-        "decision"?: {
-            "endpoint": "/v1/decisions";
-            "question_types": Array<"noul" | "choice" | "score">;
-        };
-    }>;
-    /** Opaque keyset cursor; null on the last page. */
-    "next_cursor": string | null;
-}
-export interface WireToolList {
-    "object": "list";
-    "data": Array<{
-        "type": "web_search" | "fetch_content" | "hublu" | "memory" | "image" | "browser" | "android";
-        "object": "tool";
-        "description": string;
-        "expands_to": string[];
-        "available": boolean;
-    }>;
-    /** Opaque keyset cursor; null on the last page. */
-    "next_cursor": string | null;
-}

package/dist/index.d.ts

@@ -1,6 +1,3 @@
-import type { WireBriefPublishInput, WireBriefPublishResult, WireBriefDetail } from "./generated.js";
-export type { WireBriefPublishInput, WireBriefPublishResult, WireBriefDetail } from "./generated.js";
-export type { WireBriefPublishInput as BriefPublishInput, WireBriefPublishResult as BriefPublishResult, WireBriefDetail as BriefDetail } from "./generated.js";
 /**
  * @betterlu/agentlu - client for the agentlu service API.
  *
@@ -12,13 +9,10 @@ export type { WireBriefPublishInput as BriefPublishInput, WireBriefPublishResult
  * get-or-create registration, typed `runJson` / `runText`, use-case catalogs,
  * background `wait`, and auto-retry on `409 thread_busy`.
  *
- * Two tiers, two namespaces (1.8):
- *
- *   - `alu.llm.text/json/stream(...)` - the direct tier: one profile-free
- *     provider call, no tools, no session.
- *   - `alu.agent.run/stream/background(...)` - the agent tier: a profile,
+ * The agent tier is exposed through `alu.agent.run/stream/background(...)`:
+ * a profile,
  *     tool families, threads and background runs.
- *   - `alu.run/runJson/runText/stream` stay as aliases of the agent tier.
+ * `alu.run/runJson/runText/stream` stay as aliases of the agent tier.
  *
  * Errors carry a coarse `kind` (`unavailable | timeout | rate_limited |
  * invalid_request | schema | incomplete | internal`) and `isUnavailable(err)`
@@ -28,7 +22,7 @@ export type { WireBriefPublishInput as BriefPublishInput, WireBriefPublishResult
  * `Wire*` types (see `generated.ts`, regenerate with `npm run gen:contract`).
  */
 import type { WireHouseholdSignalInput, WireHouseholdSignalResult, WireAppAgentDeclaration, WireClientDefaults, WireUseCaseProfile } from "./generated.js";
-export type { WireHouseholdSignalInput, WireHouseholdSignalResult, WireAppAgentDeclaration, WireAgentInfo, WireChatCompletion, WireChatCompletionRequest, WireClient, WireClientDefaults, WireCreateDecision, WireCreateResponse, WireCreateThreadRequest, WireDecision, WireError, WireFile, WireModelInfo, WireRegisterClientRequest, WireResponse, WireStreamEvent, WireThread, WireThreadMessage, WireToolInfo, WireUseCaseProfile, } from "./generated.js";
+export type { WireHouseholdSignalInput, WireHouseholdSignalResult, WireAppAgentDeclaration, WireChatCompletion, WireChatCompletionRequest, WireClient, WireClientDefaults, WireCreateDecision, WireCreateResponse, WireCreateThreadRequest, WireDecision, WireError, WireFile, WireRegisterClientRequest, WireResponse, WireStreamEvent, WireThread, WireThreadMessage, WireUseCaseProfile, } from "./generated.js";
 export declare const DEFAULT_BASE_URL = "http://localhost:8912/v1";
 /**
  * Accept an origin (`http://host:8912`) or a `/v1` base and normalize to the
@@ -428,40 +422,6 @@ export interface UploadManyOptions extends UploadOptions {
     /** Parallel uploads. Default 3. */
     concurrency?: number;
 }
-/**
- * Direct-tier request (`llm.text` / `llm.json` / `llm.stream`): one
- * profile-free provider call, no tools, no session. `instructions` is the
- * whole system prompt.
- */
-export interface LlmRequest {
-    input: string;
-    instructions?: string;
-    model?: string;
-    temperature?: number;
-    timeoutMs?: number;
-    maxOutputTokens?: number;
-    metadata?: Record<string, string>;
-    signal?: AbortSignal;
-    maxAttempts?: number;
-    idempotencyKey?: string;
-}
-export interface LlmJsonRequest extends LlmRequest {
-    /** Format name sent to the provider (default "Result"). */
-    schemaName?: string;
-}
-export interface LlmApi {
-    text: (request: string | LlmRequest) => Promise<{
-        text: string;
-        response: ResponseObject;
-    }>;
-    json: <T = unknown>(schema: Record<string, unknown>, request: string | LlmJsonRequest) => Promise<{
-        data: T;
-        response: ResponseObject;
-    }>;
-    stream: (request: string | LlmRequest) => AsyncIterable<StreamEvent> & {
-        finalText: () => Promise<string>;
-    };
-}
 /** Agent-tier entry points (`alu.agent.*`). */
 export interface AgentTierApi {
     run: (opts: RunOptions) => Promise<{
@@ -502,30 +462,6 @@ export interface ResponseObject {
     [k: string]: unknown;
 }
 export declare function isTerminalStatus(status: string): boolean;
-/**
- * One SSE event from `/v1/responses` with `stream: true`. The event names and
- * shapes mirror the OpenAI Responses stream; `delta` is the incremental text
- * for `response.output_text.delta`.
- */
-export interface StreamEvent {
-    type: string;
-    sequence_number?: number;
-    /** Incremental text on `response.output_text.delta`. */
-    delta?: string;
-    [k: string]: unknown;
-}
-/**
- * Accumulate a raw stream into typed events and the final text. `events()`
- * yields every event; `finalText()` resolves after the stream ends with the
- * concatenated `response.output_text.delta` payloads.
- *
- *   const stream = alu.typedStream({ input: "hi" });
- *   for await (const ev of stream) { ... }
- *   const text = stream.finalText();
- */
-export declare function typedStream(source: AsyncIterable<Record<string, unknown>>): AsyncIterable<StreamEvent> & {
-    finalText: () => Promise<string>;
-};
 export declare class AgentLu {
     readonly baseUrl: string;
     apiKey: string | undefined;
@@ -571,7 +507,6 @@ export declare class AgentLu {
         kind?: string;
     }>;
     private uploadForApiKey;
-    me(): Promise<unknown>;
     /**
      * One-shot helper. When `schema` is given, returns the validated object as
      * `data` and throws `AgentLuError("schema_validation_failed")` if the server
@@ -597,30 +532,12 @@ export declare class AgentLu {
     /** `runText` that returns `null` instead of throwing on any failure. */
     runTextOrNull(opts: RunOptions): Promise<string | null>;
     stream(opts: RunOptions): AsyncGenerator<Record<string, unknown>>;
-    /**
-     * Typed stream: yields `StreamEvent`s and exposes `finalText()` with the
-     * concatenated output text after the stream ends.
-     */
-    typedStream(opts: RunOptions): AsyncIterable<StreamEvent> & {
-        finalText: () => Promise<string>;
-    };
-    /**
-     * Direct tier (`agent: "direct"`): one profile-free provider call - no
-     * profile, no tools, no session. Use it for extraction, classification and
-     * short answers; use `agent.*` when the run needs tools or a thread.
-     *
-     *   const { text } = await alu.llm.text("Name three colours.");
-     *   const { data } = await alu.llm.json(schema, { input, schemaName: "Items" });
-     */
-    readonly llm: LlmApi;
     /**
      * Agent tier: a profile, tool families, threads and background runs. The
      * default agent is `api` (no persona - `instructions` are the system prompt);
      * pin `fast`/`thorough`/... or an app-owned `<app>/<name>` id when needed.
      */
     readonly agent: AgentTierApi;
-    /** Build a direct-tier run body from a short-form `llm.*` request. */
-    private llmOptions;
     /**
      * Enqueue a background run (`background: true`) and poll it to a terminal
      * status with `responses.wait`. Resolves with the terminal Response row -
@@ -671,10 +588,6 @@ export declare class AgentLu {
         delete: (id: string) => Promise<unknown>;
         messages: (id: string) => Promise<unknown>;
     };
-    readonly briefs: {
-        publish: (input: WireBriefPublishInput, opts?: RequestOptions) => Promise<WireBriefPublishResult>;
-        get: (id: string, opts?: RequestOptions) => Promise<WireBriefDetail>;
-    };
     readonly household: {
         signal: (input: WireHouseholdSignalInput, opts?: RequestOptions) => Promise<WireHouseholdSignalResult>;
     };
@@ -695,10 +608,6 @@ export declare class AgentLu {
      * overrides the shared 5-attempt policy, and `signal` cancels the call.
      */
     readonly decisions: DecisionApi;
-    agents(): Promise<unknown>;
-    models(): Promise<unknown>;
-    tools(): Promise<unknown>;
-    health(): Promise<unknown>;
     /**
      * Wait until the service answers `/health`, retrying while it is unreachable
      * (offline, 502/503/504, timeout) with jittered backoff. Resolves `true` once
@@ -767,10 +676,6 @@ export declare class UseCaseRunner {
     /** `runText` that returns `null` instead of throwing on any failure. */
     runTextOrNull(opts: RunOptions): Promise<string | null>;
     stream(opts: RunOptions): AsyncGenerator<Record<string, unknown>>;
-    /** Typed stream for this use-case: events + `finalText()`. */
-    typedStream(opts: RunOptions): AsyncIterable<StreamEvent> & {
-        finalText: () => Promise<string>;
-    };
     /**
      * Upload with this use-case's key, so the file is owned by the same
      * service client that will reference it via `run({ fileIds })`.

package/dist/index.js

@@ -1,6 +1,4 @@
 export const DEFAULT_BASE_URL = "http://localhost:8912/v1";
-/** Wire agent id for the direct tier (not a chat mode). */
-const DIRECT_AGENT_ID = "direct";
 /** Default agent for the agent tier: the profile-free machine agent. */
 const AGENT_TIER_AGENT = "api";
 /**
@@ -193,44 +191,6 @@ async function uploadManyWith(files, opts, upload) {
 export function isTerminalStatus(status) {
     return status === "completed" || status === "failed" || status === "incomplete" || status === "cancelled";
 }
-/**
- * Accumulate a raw stream into typed events and the final text. `events()`
- * yields every event; `finalText()` resolves after the stream ends with the
- * concatenated `response.output_text.delta` payloads.
- *
- *   const stream = alu.typedStream({ input: "hi" });
- *   for await (const ev of stream) { ... }
- *   const text = stream.finalText();
- */
-export function typedStream(source) {
-    let text = "";
-    let done = false;
-    const it = (async function* () {
-        try {
-            for await (const raw of source) {
-                const ev = raw;
-                if (ev.type === "response.output_text.delta" && typeof ev.delta === "string") {
-                    text += ev.delta;
-                }
-                yield ev;
-            }
-        }
-        finally {
-            done = true;
-        }
-    })();
-    return Object.assign(it, {
-        finalText: async () => {
-            // Drain if the caller never iterated to the end.
-            if (!done) {
-                for await (const _ of it) {
-                    void _;
-                }
-            }
-            return text;
-        },
-    });
-}
 function env(name) {
     return typeof process !== "undefined" ? process.env[name] : undefined;
 }
@@ -361,9 +321,6 @@ export class AgentLu {
         });
         return (await res.json());
     }
-    async me() {
-        return this.request("GET", "/me");
-    }
     /**
      * One-shot helper. When `schema` is given, returns the validated object as
      * `data` and throws `AgentLuError("schema_validation_failed")` if the server
@@ -405,29 +362,6 @@ export class AgentLu {
     async *stream(opts) {
         yield* this.streamRunForApiKey(opts, this.apiKey);
     }
-    /**
-     * Typed stream: yields `StreamEvent`s and exposes `finalText()` with the
-     * concatenated output text after the stream ends.
-     */
-    typedStream(opts) {
-        return typedStream(this.stream(opts));
-    }
-    /**
-     * Direct tier (`agent: "direct"`): one profile-free provider call - no
-     * profile, no tools, no session. Use it for extraction, classification and
-     * short answers; use `agent.*` when the run needs tools or a thread.
-     *
-     *   const { text } = await alu.llm.text("Name three colours.");
-     *   const { data } = await alu.llm.json(schema, { input, schemaName: "Items" });
-     */
-    llm = {
-        text: (request) => this.runText(this.llmOptions(request)),
-        json: (schema, request) => {
-            const schemaName = typeof request === "string" ? undefined : request.schemaName;
-            return this.runJson({ ...this.llmOptions(request), schema, schemaName: schemaName ?? "Result" });
-        },
-        stream: (request) => this.typedStream(this.llmOptions(request)),
-    };
     /**
      * Agent tier: a profile, tool families, threads and background runs. The
      * default agent is `api` (no persona - `instructions` are the system prompt);
@@ -438,11 +372,6 @@ export class AgentLu {
         stream: (opts) => this.stream(withAgentTierDefault(opts)),
         background: (opts, wait) => this.runBackground(withAgentTierDefault(opts), wait),
     };
-    /** Build a direct-tier run body from a short-form `llm.*` request. */
-    llmOptions(request) {
-        const opts = typeof request === "string" ? { input: request } : request;
-        return { ...opts, agent: DIRECT_AGENT_ID, store: false, tools: [] };
-    }
     /**
      * Enqueue a background run (`background: true`) and poll it to a terminal
      * status with `responses.wait`. Resolves with the terminal Response row -
@@ -728,10 +657,6 @@ export class AgentLu {
         delete: (id) => this.request("DELETE", `/threads/${encodeURIComponent(id)}`),
         messages: (id) => this.request("GET", `/threads/${encodeURIComponent(id)}/messages`),
     };
-    briefs = {
-        publish: (input, opts) => this.request("POST", "/briefs", input, opts),
-        get: (id, opts) => this.request("GET", `/briefs/${encodeURIComponent(id)}`, undefined, opts),
-    };
     household = {
         signal: (input, opts) => this.request("POST", "/household/signals", input, opts),
     };
@@ -763,18 +688,6 @@ export class AgentLu {
             }
         },
     };
-    async agents() {
-        return this.request("GET", "/agents", undefined, { auth: false });
-    }
-    async models() {
-        return this.request("GET", "/models", undefined, { auth: false });
-    }
-    async tools() {
-        return this.request("GET", "/tools", undefined, { auth: false });
-    }
-    async health() {
-        return this.request("GET", "/health", undefined, { auth: false });
-    }
     /**
      * Wait until the service answers `/health`, retrying while it is unreachable
      * (offline, 502/503/504, timeout) with jittered backoff. Resolves `true` once
@@ -1017,10 +930,6 @@ export class UseCaseRunner {
             yield* this.operations.stream(this.apiKey, opts);
         }
     }
-    /** Typed stream for this use-case: events + `finalText()`. */
-    typedStream(opts) {
-        return typedStream(this.stream(opts));
-    }
     /**
      * Upload with this use-case's key, so the file is owned by the same
      * service client that will reference it via `run({ fileIds })`.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@betterlu/agentlu",
-  "version": "1.11.0",
+  "version": "2.0.0",
   "description": "Client for the agentlu service API (/v1). Or just use the openai SDK pointed at http://localhost:8912/v1.",
   "type": "module",
   "main": "./dist/index.js",