@falai/agent 3.4.5 → 4.0.0-alpha.2

package/docs/reference/signals.md

@@ -1,368 +0,0 @@
----
-title: "Signals"
-description: "Typed event detectors that run before and after each LLM turn, with optional structured extraction and rate-limited firing."
-type: reference
-order: 9
----
-
-# Signals
-
-> **Where this is introduced:** [Turn pipeline](../concepts/pipeline.md)
-
-A `Signal` is a typed event detector that runs around an LLM turn. Each signal pairs a *condition* (`when` / `if`) with a *handler* that returns a `SignalDirective`. Signals optionally carry an `extract` schema for structured data extraction beyond boolean detection, run at two pipeline phases (PRE-SIGNAL parallel with routing, POST-SIGNAL after the LLM), are rate-limited (`once` / `always` / `cooldown`), and surface their firings on `AgentResponse.triggeredSignals`.
-
-Signals operate in two modes:
-
-- **Detection mode** — no `extract`. Boolean match/no-match via a batched classifier call. Multiple signals share one LLM call.
-- **Extraction mode** — `extract` set. Structured data extraction merged into the same batched call. The handler receives `ctx.extracted` typed by the `TExtract` generic.
-
-When `agent.signals` is empty or undefined, both phases short-circuit to zero cost — no extra provider calls.
-
-## Signature
-
-```typescript
-interface Signal<TContext = unknown, TData = unknown, TExtract = void> {
-  // Identity
-  id?: string;
-  title?: string;
-  description?: string;
-
-  // Conditions (v2 when/if split)
-  when?: string | string[];
-  if?: SignalPredicate<TContext, TData> | SignalPredicate<TContext, TData>[];
-
-  // Optional structured extraction
-  extract?: SignalSchema<TExtract>;
-
-  // Phase + handler
-  phase: 'pre' | 'post' | 'both';
-  handler: (ctx: SignalContext<TContext, TData, TExtract>) =>
-    | void
-    | SignalDirective<TContext, TData>
-    | Promise<void | SignalDirective<TContext, TData>>;
-
-  // Rate limiting
-  behavior?: 'once' | 'always' | 'cooldown';
-  cooldownMs?: number;
-
-  // Misc
-  enabled?: boolean;
-  priority?: number;
-}
-
-type SignalSchema<_T = unknown> = Record<string, unknown>;
-
-type SignalPredicate<TContext = unknown, TData = unknown> = (
-  ctx: SignalPredicateContext<TContext, TData>,
-) => boolean | Promise<boolean>;
-
-interface SignalPredicateContext<TContext = unknown, TData = unknown> {
-  data: Partial<TData>;
-  context: TContext;
-  session: SessionState<TData>;
-  history: Event[];
-}
-
-interface SignalContext<TContext = unknown, TData = unknown, TExtract = void> {
-  signal: Signal<TContext, TData, TExtract>;
-  phase: 'pre' | 'post';
-  matched: true;
-  reason: string;
-  extracted: TExtract extends void ? undefined : TExtract;
-
-  session: SessionState<TData>;
-  context: TContext;
-  data: Partial<TData>;
-  history: Event[];
-  lastUserMessage?: string;
-  triggeredAt: Date;
-
-  updateContext(updates: Partial<TContext>): Promise<void>;
-  updateData(updates: Partial<TData>): Promise<void>;
-  dispatch(directive: SignalDirective<TContext, TData>): void;
-}
-
-interface SignalDirective<TContext = unknown, TData = unknown>
-  extends Directive<TContext, TData> {
-  stopOtherSignals?: boolean;
-  replyWith?: string | ((ctx: SignalContext<TContext, TData>) => string);
-}
-
-// As reported on the response surface: replyWith already resolved onto reply
-type ResolvedSignalDirective<TContext = unknown, TData = unknown> =
-  Directive<TContext, TData> & { stopOtherSignals?: boolean };
-
-interface SignalFiring<TContext = unknown, TData = unknown> {
-  id: string;
-  phase: 'pre' | 'post';
-  reason?: string;
-  extracted?: unknown;
-  directive?: ResolvedSignalDirective<TContext, TData>;
-  handlerError?: string;
-  extractionError?: string;
-  durationMs?: number;
-}
-
-interface SignalsState {
-  triggers: Record<string, SignalTriggerState>;
-}
-
-interface SignalTriggerState {
-  firstTriggeredAt: Date;
-  lastTriggeredAt: Date;
-  count: number;
-  lastReason?: string;
-  lastPhase?: 'pre' | 'post';
-}
-```
-
-Signals are wired into the agent through two `AgentOptions` fields:
-
-```typescript
-interface AgentOptions<TContext, TData> {
-  // ... other fields ...
-  signals?: Signal<TContext, TData, unknown>[];
-  signalBatchSize?: number; // default 10
-}
-```
-
-## Fields
-
-### `Signal`
-
-| Field | Type | Required | Default | Notes |
-|-------|------|----------|---------|-------|
-| `id` | `string` | no | auto-generated | Stable identifier within a session. Used for `SignalsState.triggers` keying and on `SignalFiring`. Must be unique across the agent's signals. |
-| `title` | `string` | no | — | Display title shown in logs and traces. |
-| `description` | `string` | no | — | Free-text description; rendered into the classifier prompt. |
-| `when` | `string \| string[]` | no | — | AI-evaluated condition(s), using the same `ConditionWhen` include/exclude syntax as flows, steps, branches, and instructions. Non-prefixed entries render under "TRIGGER WHEN" (OR semantics — any match can trigger). Entries prefixed with `!` are exclusion conditions rendered under "DO NOT TRIGGER WHEN" (OR semantics — any match inhibits firing). |
-| `if` | `SignalPredicate \| SignalPredicate[]` | no | — | Code predicate(s). AND semantics. Free to evaluate. Runs before `when`; if any returns `false`, `when` is skipped (no token cost). |
-| `extract` | `SignalSchema<TExtract>` | no | — | When set, the signal operates in extraction mode. JSON Schema object describing the per-signal `extracted` field merged into the classifier response. The `TExtract` generic carries the resulting type onto `ctx.extracted`. |
-| `phase` | `'pre' \| 'post' \| 'both'` | yes | — | When the signal evaluates. `'pre'` runs in parallel with routing. `'post'` runs after the LLM call, before persistence. `'both'` evaluates in both phases. |
-| `handler` | `(ctx) => void \| SignalDirective \| Promise<…>` | yes | — | Invoked when the signal fires. May return a directive, dispatch one via `ctx.dispatch`, or return void for side-effect-only behavior. |
-| `behavior` | `'once' \| 'always' \| 'cooldown'` | no | `'always'` | Rate-limit / dedup mode. `'once'` fires once per session. `'cooldown'` requires `cooldownMs`. |
-| `cooldownMs` | `number` | conditional | — | Required when `behavior === 'cooldown'`. Suppresses re-firing for this duration after the last trigger. Misconfiguration logs a debug warning and falls back to `'always'`. |
-| `enabled` | `boolean` | no | `true` | When `false`, the signal is filtered out at the start of the phase. |
-| `priority` | `number` | no | `0` | Higher priority handlers run first within a phase. Ties break by declaration order in `agent.signals`. |
-
-### `SignalPredicateContext`
-
-Passed to every `if` predicate. Symmetric with `BranchPredicateContext`.
-
-| Field | Type | Notes |
-|-------|------|-------|
-| `data` | `Partial<TData>` | Collected data so far. Fields not yet collected are `undefined`. |
-| `context` | `TContext` | Ambient agent context. |
-| `session` | `SessionState<TData>` | Full session state, including `session.signals` for inspecting prior triggers. |
-| `history` | `Event[]` | Conversation history as native events. |
-
-A predicate that throws is treated as a non-match — other signals continue evaluating.
-
-### `SignalContext`
-
-Passed to handlers when a signal fires. Symmetric with `ToolContext`.
-
-| Field | Type | Notes |
-|-------|------|-------|
-| `signal` | `Signal<TContext, TData, TExtract>` | The signal definition that fired. |
-| `phase` | `'pre' \| 'post'` | Which phase this firing belongs to. Useful when `signal.phase === 'both'`. |
-| `matched` | `true` | Always `true` when the handler runs. Typed for narrowing. |
-| `reason` | `string` | AI rationale when `when` matched, or `'code-only'` / `'unconditional'`. |
-| `extracted` | `TExtract extends void ? undefined : TExtract` | Extracted data when `signal.extract` is set; `undefined` for detection-only signals. |
-| `session` | `SessionState<TData>` | Session state. Use writers below for mutations. |
-| `context` | `TContext` | Ambient agent context. |
-| `data` | `Partial<TData>` | Collected data (partial). |
-| `history` | `Event[]` | Conversation history. |
-| `lastUserMessage` | `string \| undefined` | Convenience accessor for the most recent user message. |
-| `triggeredAt` | `Date` | Wall-clock timestamp when the handler started. |
-| `updateContext` | `(updates) => Promise<void>` | Shallow-merge into `context`. Same contract as `ToolContext.updateContext`. |
-| `updateData` | `(updates) => Promise<void>` | Shallow-merge into `data`. Same contract as `ToolContext.updateData`. |
-| `dispatch` | `(directive) => void` | Emit a `SignalDirective` onto the per-turn bus. May be called multiple times; emissions merge by Algorithm 4 alongside hook and tool directives. |
-
-### `SignalDirective`
-
-Extends [`Directive`](./directive.md). All position fields (`goTo`, `goToStep`, `complete`, `abort`, `reset`), state writes (`dataUpdate`, `contextUpdate`), `reply`, and pre-LLM fields (`appendPrompt`, `injectTools`, `halt`) are inherited unchanged.
-
-| Added field | Type | Notes |
-|-------------|------|-------|
-| `stopOtherSignals` | `boolean` | When `true`, skip remaining signals in the current phase after this handler. Does not affect the other phase. Consumed inside the signal pipeline — does not enter the directive bus. |
-| `replyWith` | `string \| ((ctx) => string)` | Late-binding `reply`. String form is identical to `Directive.reply`. Function form is evaluated at emit time and projected onto `reply` before bus merging. |
-
-**Post-phase drop rules.** When a signal runs in the post-phase, `appendPrompt`, `injectTools`, and `halt` are dropped with a debug warning — they have no meaning after the LLM call has already completed. Position directives in the post-phase set `session.pendingDirective` for the *next* turn (no mid-turn re-entry).
-
-### `ResolvedSignalDirective`
-
-The directive shape as reported on the response surface (`SignalFiring.directive`). The signal processor resolves `replyWith` onto `reply` and strips it before firings reach `AgentResponse` — so a `ResolvedSignalDirective` is a plain [`Directive`](./directive.md) plus `stopOtherSignals`, with the reply text (if any) always on `reply`. Exported from the package barrel.
-
-### `SignalFiring`
-
-One entry per signal that fired this turn. Populated in fire order across both phases on `AgentResponse.triggeredSignals` (and on the final chunk of `AgentResponseStreamChunk`). Mirrors the observability framing of `executedSteps` and `appliedInstructions`.
-
-| Field | Type | Notes |
-|-------|------|-------|
-| `id` | `string` | The signal's identifier. |
-| `phase` | `'pre' \| 'post'` | Phase the signal fired in. |
-| `reason` | `string \| undefined` | AI rationale, `'code-only'`, or `'unconditional'`. |
-| `extracted` | `unknown` | Extracted payload when in extraction mode. |
-| `directive` | `ResolvedSignalDirective \| undefined` | The directive returned (or dispatched) by the handler, with `replyWith` already resolved onto `reply`. |
-| `handlerError` | `string \| undefined` | Error message if the handler threw. The turn continues — handler errors never break a turn. |
-| `extractionError` | `string \| undefined` | Set when an extraction-mode signal matched but the classifier returned no `extracted` payload (model omitted the field or returned `null`). The handler still runs with `extracted: undefined`; this makes the miss observable instead of silent. Independent of `handlerError`. |
-| `durationMs` | `number \| undefined` | Wall-clock duration of the handler invocation. |
-
-### `SignalsState` and `SignalTriggerState`
-
-Persisted on `session.signals`. Adapters preserve this shape bit-identical.
-
-`SignalsState` has a single field, `triggers: Record<string, SignalTriggerState>`, keyed by `signal.id`.
-
-| `SignalTriggerState` field | Type | Notes |
-|----------------------------|------|-------|
-| `firstTriggeredAt` | `Date` | When this signal first completed its handler successfully in the session. Never updated on subsequent successful triggers. |
-| `lastTriggeredAt` | `Date` | When this signal last completed its handler successfully. Drives `cooldown` arithmetic. |
-| `count` | `number` | Total successful handler completions for this signal in this session. Monotonically increasing. |
-| `lastReason` | `string \| undefined` | The `reason` from the most recent successful trigger. |
-| `lastPhase` | `'pre' \| 'post' \| undefined` | The phase of the most recent successful trigger. |
-
-Handler errors are still surfaced on `SignalFiring.handlerError`, but they do not update `SignalsState.triggers`. A `behavior: 'once'` or cooldown signal can therefore retry after an app-side handler failure.
-
-### `signalBatchSize`
-
-Maximum signals per batched classifier call. Default `10`. When more LLM-conditioned signals are eligible after gating, they split into parallel batches of this size — each batch is one provider call, all batches run via `Promise.all`. Setting this to `0` or a non-integer throws `FlowConfigurationError` at agent construction.
-
-## Phase semantics
-
-| Capability | Pre-phase | Post-phase |
-|---|---|---|
-| `halt` | Skips the LLM call | Dropped with debug warning |
-| `reply` / `replyWith` | Replaces the message that would have been generated | Replaces the message that was just generated |
-| `goTo` / `goToStep` | Overrides routing for this turn | Sets `pendingDirective` for next turn |
-| `complete` | This turn closes the flow | Same |
-| `appendPrompt` | Injected into this turn's response prompt | Dropped with debug warning |
-| `injectTools` | Available to this turn's LLM | Dropped with debug warning |
-| `dataUpdate` / `contextUpdate` | Applied before LLM, visible in prompt | Applied, persisted |
-| `extract` results | Available to handler pre-LLM | Available to handler post-LLM |
-| `stopOtherSignals` | Stops further pre-signals | Stops further post-signals |
-
-### Resolution precedence within a turn
-
-Per turn, position decisions are resolved in this fixed order:
-
-1. **`session.pendingDirective`** (set last turn or via `agent.dispatch()`) — consumed first.
-2. **PRE-SIGNAL phase directives** (parallel with routing). Position fields override routing; `halt` discards routing entirely.
-3. **AI routing** — used only if (1) and (2) produced no position field.
-4. **Auto-step chain.**
-5. **Step branches.**
-6. **Linear successor / AI step selection** — final fallback.
-7. **POST-SIGNAL phase directives** — set `pendingDirective` for the *next* turn.
-
-## Examples
-
-### 1. Pre-phase halt with verbatim reply (escalation)
-
-A handoff signal that intercepts the turn before the LLM runs, notifies a side channel, and replies with a fixed message. Cooldown prevents re-notification within the hour.
-
-```typescript
-import type { Signal } from "@falai/agent";
-
-type Ctx = { conversationId: string };
-type Data = { topic?: string };
-
-export const humanHandoff: Signal<Ctx, Data> = {
-  id: "human_handoff",
-  title: "Human handoff requested",
-  description: "Lead asks to talk to a real person.",
-  when: [
-    "the user explicitly asks to talk to a human, agent, or representative",
-    "the user explicitly says they do not want to talk to a bot",
-    "!casual mentions of people (e.g., 'my colleague said')",
-    "!general frustration without an explicit handoff request",
-  ],
-  phase: "pre",
-  behavior: "cooldown",
-  cooldownMs: 60 * 60_000, // 1 hour
-  priority: 100,
-  async handler({ session, lastUserMessage }) {
-    await notifyTeam(session.id, lastUserMessage);
-    return {
-      halt: true,
-      replyWith: "I'm connecting you with someone from our team. They'll reach out shortly.",
-      stopOtherSignals: true,
-    };
-  },
-};
-```
-
-### 2. Post-phase extraction (entity capture)
-
-A signal that runs after every LLM turn, extracts a structured payload, and writes it back into collected data. No `when` — it's gated by `if` on session state, then unconditionally extracts when eligible.
-
-```typescript
-import type { Signal } from "@falai/agent";
-
-type Data = { leadStage?: "cold" | "warm" | "hot" | "closing" };
-
-export const leadStage: Signal<unknown, Data, { stage: Data["leadStage"]; confidence: number }> = {
-  id: "lead_stage",
-  title: "Lead stage classification",
-  if: ({ session }) => (session.history?.length ?? 0) > 5,
-  extract: {
-    type: "object",
-    properties: {
-      stage: { type: "string", enum: ["cold", "warm", "hot", "closing"] },
-      confidence: { type: "number", minimum: 0, maximum: 1 },
-    },
-    required: ["stage", "confidence"],
-  },
-  phase: "post",
-  behavior: "always",
-  async handler({ extracted, updateData }) {
-    if (extracted.confidence >= 0.6) {
-      await updateData({ leadStage: extracted.stage });
-    }
-  },
-};
-```
-
-### 3. Cooldown behavior
-
-A pre-phase signal that nudges the next prompt with a tone hint, suppressed for two minutes after each fire so a frustrated user isn't re-flagged on every turn.
-
-```typescript
-import type { Signal } from "@falai/agent";
-
-export const frustrationDetected: Signal = {
-  id: "frustration_detected",
-  when: "the user is visibly frustrated, impatient, or upset",
-  phase: "pre",
-  behavior: "cooldown",
-  cooldownMs: 2 * 60_000, // 2 minutes
-  async handler({ dispatch }) {
-    dispatch({
-      appendPrompt: ["The user is frustrated. Lead with empathy and acknowledge the issue."],
-    });
-  },
-};
-```
-
-## Errors
-
-Misuse surfaces as typed errors:
-
-- `FlowConfigurationError` — duplicate `id` across `agent.signals`, invalid `extract` schema (not a JSON Schema object), `signalBatchSize` not a positive integer, or a position directive (`goTo` / `goToStep`) that resolves to an unknown flow or step.
-- `DataValidationError` — `dataUpdate` (or `extracted` written via `updateData`) violates the agent schema.
-
-Soft failures handled in-band (no thrown error, turn continues):
-
-- **Handler throws** — recorded as `firings[i].handlerError`; iteration continues with the next signal.
-- **Classifier call fails** — all LLM-conditioned signals in that batch are treated as non-match; code-only and unconditional signals continue normally.
-- **Extraction-mode signal matched but no payload returned** — recorded as `firings[i].extractionError` with a WARN log; the handler still runs with `extracted: undefined`. Usually means the `extract` schema used a keyword the provider ignores (e.g. `nullable: true` instead of `type: ['string', 'null']`), so the model omitted the field.
-- **Post-phase emits pre-LLM-only fields** (`halt` / `appendPrompt` / `injectTools`) — dropped with debug warning.
-- **`behavior: 'cooldown'` with no `cooldownMs`** — debug warning at construction; runtime treats as `'always'`.
-
-## Related
-
-- [Turn pipeline](../concepts/pipeline.md) — where signal phases sit in the turn lifecycle.
-- [Directives](../concepts/directives.md) — the Directive shape and SignalDirective extension.
-- [Directive](./directive.md) — base shape for all position and state fields.
-- [createAgent](./create-agent.md) — `signals` and `signalBatchSize` options.
-- [Errors](./errors.md) — `FlowConfigurationError` format contract.

package/examples/02-data-extraction.ts

@@ -1,90 +0,0 @@
-/**
- * Data extraction — schema, collect, skip, requires, requiredFields; pre-extraction in action.
- * Run with the message "I want X for Y people on Z" and watch all three fields populate from one turn.
- * Read after: docs/start/03-collect-data.md
- */
-
-import { createAgent, GeminiProvider } from "@falai/agent";
-
-// ─── Schema ──────────────────────────────────────────────────────────────────
-
-interface BookingData {
-    roomType: "single" | "double" | "suite";
-    guests: number;
-    checkIn: string;
-}
-
-const schema = {
-    type: "object" as const,
-    properties: {
-        roomType: {
-            type: "string" as const,
-            enum: ["single", "double", "suite"],
-            description: "Type of room requested",
-        },
-        guests: {
-            type: "number" as const,
-            description: "Number of guests",
-        },
-        checkIn: {
-            type: "string" as const,
-            description: "Check-in date in YYYY-MM-DD format",
-        },
-    },
-    required: ["roomType", "guests", "checkIn"],
-};
-
-// ─── Agent ───────────────────────────────────────────────────────────────────
-
-const agent = createAgent<Record<string, never>, BookingData>({
-    name: "Booking Agent",
-    provider: new GeminiProvider({
-        apiKey: process.env.GEMINI_API_KEY!,
-        model: "gemini-3.5-flash-lite",
-    }),
-    schema,
-    flows: [
-        {
-            title: "Book a room",
-            when: "The user wants to book a hotel room",
-            requiredFields: ["roomType", "guests", "checkIn"],
-            steps: [
-                {
-                    id: "ask_room",
-                    prompt: "What type of room would you like? (single, double, or suite)",
-                    collect: ["roomType"],
-                    skip: [({ data }) => data.roomType !== undefined],
-                },
-                {
-                    id: "ask_guests",
-                    prompt: "How many guests will be staying?",
-                    collect: ["guests"],
-                    requires: ["roomType"],
-                    skip: [({ data }) => data.guests !== undefined],
-                },
-                {
-                    id: "ask_checkin",
-                    prompt: "When would you like to check in?",
-                    collect: ["checkIn"],
-                    requires: ["roomType", "guests"],
-                    skip: [({ data }) => data.checkIn !== undefined],
-                },
-            ],
-        },
-    ],
-});
-
-// ─── Run ─────────────────────────────────────────────────────────────────────
-
-async function main() {
-    // Pre-extraction: all three fields are extracted from a single message.
-    const response = await agent.respond({
-        history: [{ role: "user", content: "I want a double for 2 people on 2025-03-15" }],
-    });
-
-    console.log(response.message);
-    console.log("Collected:", response.session?.data);
-    console.log("Complete:", response.isFlowComplete);
-}
-
-main();

package/examples/05-branching.ts

@@ -1,140 +0,0 @@
-/** @intent Demonstrates step.branches — source-local forks with if, when, and Directive targets.
- *  @teaches BranchEntry, BranchMap, if-only fork, when-only fork, combined if+when, mixed targets
- *  @readAfter docs/guides/branching.md */
-import { createAgent, GeminiProvider } from "../src/index.js";
-import type { BranchMap, Directive, FlowOptions, StepOptions } from "../src/index.js";
-
-// ─── Types ───────────────────────────────────────────────────────────────────
-
-interface Context { tier: "free" | "pro" | "enterprise" }
-interface Data { intent: string; query: string; answer: string }
-
-// ─── Provider ────────────────────────────────────────────────────────────────
-
-const provider = new GeminiProvider({
-    apiKey: process.env.GEMINI_API_KEY!,
-    model: "gemini-3.5-flash-lite",
-});
-
-// ─── Steps with branches ─────────────────────────────────────────────────────
-
-// 1. if-only fork — pure code predicate, zero LLM cost
-const triageStep: StepOptions<Context, Data> = {
-    id: "triage",
-    prompt: "What can I help you with today?",
-    collect: ["intent"],
-    branches: [
-        {
-            label: "pro-fast-track",
-            if: ({ context }) => context.tier === "pro" || context.tier === "enterprise",
-            then: "priority_queue",   // resolves as step id in same flow
-        },
-        {
-            label: "free-default",
-            then: "general_help",     // unconditional fallback (last entry)
-        },
-    ] satisfies BranchMap<Context, Data>,
-};
-
-// 2. when-only fork — AI-evaluated condition (costs tokens)
-const classifyStep: StepOptions<Context, Data> = {
-    id: "general_help",
-    prompt: "Let me look into that for you.",
-    collect: ["query"],
-    branches: [
-        {
-            label: "billing",
-            when: "The user is asking about billing, invoices, or payments",
-            then: "billing_flow",     // resolves as flow id (cross-flow jump)
-        },
-        {
-            label: "technical",
-            when: "The user is asking a technical or API question",
-            then: "tech_support",     // resolves as step id
-        },
-        {
-            label: "fallback",
-            then: "tech_support",
-        },
-    ] satisfies BranchMap<Context, Data>,
-};
-
-// 3. Combined if + when — code runs first (free), AI only if code passes
-const priorityStep: StepOptions<Context, Data> = {
-    id: "priority_queue",
-    prompt: "You have priority support. What do you need?",
-    collect: ["query"],
-    branches: [
-        {
-            label: "enterprise-escalation",
-            if: ({ context }) => context.tier === "enterprise",
-            when: "The user is reporting a production outage or critical bug",
-            then: {
-                goTo: "escalation_flow",
-                dataUpdate: { intent: "escalation" },
-                reply: "Routing you to our on-call engineering team now.",
-            } satisfies Directive<Context, Data>,
-        },
-        {
-            label: "standard-priority",
-            then: "tech_support",
-        },
-    ] satisfies BranchMap<Context, Data>,
-};
-
-// 4. Terminal step — no branches, linear end
-const techSupportStep: StepOptions<Context, Data> = {
-    id: "tech_support",
-    prompt: "I'll find the answer to your technical question.",
-    collect: ["answer"],
-};
-
-// ─── Flows ───────────────────────────────────────────────────────────────────
-
-const supportFlow: FlowOptions<Context, Data> = {
-    id: "support_flow",
-    title: "Support",
-    description: "Routes users through support triage",
-    when: "The user needs help or support",
-    steps: [triageStep, classifyStep, priorityStep, techSupportStep],
-    requiredFields: ["answer"],
-};
-
-const billingFlow: FlowOptions<Context, Data> = {
-    id: "billing_flow",
-    title: "Billing",
-    description: "Handles billing inquiries",
-    when: "The user asks about billing",
-    steps: [{ id: "billing_help", prompt: "Let me pull up your billing info." }],
-};
-
-const escalationFlow: FlowOptions<Context, Data> = {
-    id: "escalation_flow",
-    title: "Escalation",
-    description: "Critical issue escalation for enterprise customers",
-    steps: [{ id: "escalate", prompt: "Connecting you to on-call engineering." }],
-};
-
-// ─── Agent ───────────────────────────────────────────────────────────────────
-
-const agent = createAgent<Context, Data>({
-    name: "Support Bot",
-    provider,
-    context: { tier: "pro" },
-    schema: {
-        type: "object",
-        properties: {
-            intent: { type: "string" },
-            query: { type: "string" },
-            answer: { type: "string" },
-        },
-    },
-    flows: [supportFlow, billingFlow, escalationFlow],
-});
-
-// ─── Run ─────────────────────────────────────────────────────────────────────
-
-const response = await agent.respond({
-    history: [{ role: "user", content: "I need help with my API integration" }],
-});
-console.log(response.message);