@1kbirds/chidori 0.1.26 → 3.4.0

package/README.md

@@ -1,10 +1,142 @@
-# Chidori
+# chidori TypeScript SDK
 
-## Generating Python Client
-- maturin develop --features python
+Zero-dependency TypeScript client for a running `chidori serve` instance.
+Uses the global `fetch` (Node 18+, browsers). Mirrors the Python SDK.
 
-## Generating Nodejs Client
-- npm run build-release
+> **This package is not the runtime.** It's an optional HTTP client for driving
+> the Chidori **runtime** — the `chidori` binary — from a TypeScript app. You
+> don't need it to write or run agents (those are plain `.ts` files the runtime
+> executes directly). Install the runtime separately, no Rust toolchain needed:
+> `curl -fsSL https://raw.githubusercontent.com/ThousandBirdsInc/chidori/main/scripts/install.sh | sh`
+> — see the [project README](https://github.com/ThousandBirdsInc/chidori#%EF%B8%8F-quick-start).
 
+## Install
 
-Python patterns were implemented while referring to https://github.com/pydantic/pydantic-core/tree/main/python/pydantic_core
+The package is published to npm as
+[`@1kbirds/chidori`](https://www.npmjs.com/package/@1kbirds/chidori). The
+unscoped `chidori` npm name belongs to an unrelated project, so **always import
+the scoped name** — never `npm install chidori`:
+
+```bash
+npm install @1kbirds/chidori
+```
+
+```ts
+import { AgentClient, Checkpoint } from "@1kbirds/chidori";
+```
+
+To build from source instead:
+
+```bash
+cd sdk/typescript
+npm install
+npm run build
+```
+
+### Authoring agents and tools
+
+Agent and tool files run *inside* the Chidori runtime, not as a normal Node
+program. They get their authoring types — and the `chidori`/`run` globals — from
+the **virtual** module `chidori:agent`:
+
+```ts
+/// <reference types="@1kbirds/chidori/agent-env" />
+import type { Chidori, ToolDefinition } from "chidori:agent";
+```
+
+There is no installable package behind `chidori:agent`; it is a URL-style scheme
+(like `node:fs`) that the runtime strips and injects at execution time, so the
+unrelated `chidori` npm package can never be pulled in by mistake. The
+`/// <reference …>` line (or a `compilerOptions.types: ["@1kbirds/chidori/agent-env"]`
+entry in `tsconfig.json`) gives editors and `tsc` the types while you author;
+the runtime itself needs nothing installed.
+
+## Usage
+
+```ts
+import { AgentClient, Checkpoint, isSignalQueued } from "@1kbirds/chidori";
+
+const client = new AgentClient("http://localhost:8080");
+
+// Run an agent
+const session = await client.run({ document: "Rust is a systems language." });
+console.log(session.output);
+
+// Save and replay a checkpoint — zero LLM calls on replay
+const checkpoint = await session.checkpoint();
+const replayed = await client.replay(checkpoint);
+
+// Durable TypeScript runs may include snapshot metadata in the checkpoint.
+// The manifest is safe to inspect; raw VM snapshot bytes remain server-side.
+if (checkpoint.snapshotManifest) {
+  console.log(checkpoint.snapshotManifest.pending?.kind);
+  console.log(checkpoint.snapshotManifest.abi.engine_fork);
+}
+
+const manifest = await client.getSnapshotManifest(session.id);
+console.log(manifest.policy.typescript_imports);
+
+// Live streaming: host calls, prompt stream deltas, then `done`
+for await (const evt of client.stream({ document: "hi" })) {
+  if (evt.type === "call") console.log(evt.record.function);
+  if (evt.type === "prompt_delta" && evt.prompt_type === "progress") {
+    process.stdout.write(evt.delta);
+  }
+  if (evt.type === "done") console.log(evt.status, evt.output);
+}
+
+// Paused sessions (from input())
+if (session.status === "paused") {
+  const resumed = await client.resume(session.id, "yes");
+  console.log(resumed.output);
+}
+
+// Multiplayer signals (from chidori.signal / pollSignal): deliver
+// { name, payload?, from? } to a run.
+const result = await client.signal(session.id, {
+  name: "review",
+  payload: { decision: "approve", notes: "LGTM" },
+  from: { kind: "human", id: "mara" },
+});
+if (isSignalQueued(result)) {
+  // run wasn't paused-waiting on this name → enqueued in the durable mailbox (202)
+  console.log("queued at delivery_seq", result.delivery_seq);
+} else {
+  // run was paused-waiting on this name → resolved + resumed (200)
+  console.log(result.status, result.output);
+}
+```
+
+See the top-level `sdk/python/chidori` for the Python equivalent.
+
+## Snapshot-aware checkpoints
+
+`Checkpoint` contains the replay call log plus optional `snapshotManifest`
+metadata. The manifest records the runtime ABI, deterministic policy, source
+hashes, pending host operation, and snapshot file name. Clients can use it to
+display durable-resume state or diagnose why resume is blocked without handling
+the raw `runtime.snapshot` VM bytes.
+
+`client.replay(checkpoint)` uses the call log for deterministic replay. Durable
+resume is exposed through `client.resume(sessionId, response)` for paused
+sessions, recovering through persisted host-promise metadata and the replay
+journal. Replay **is** the resume mechanism by design — the QuickJS live-VM
+snapshot path was removed in #39, not merely deferred, so the manifest carries
+journal/scaffold metadata rather than serialized VM bytes.
+
+Use `client.getSnapshotManifest(sessionId)` when a UI needs only snapshot
+metadata. The endpoint never returns the binary VM snapshot.
+
+## Tests
+
+The SDK ships a dependency-free test suite (Node's built-in `node:test`
+runner) that drives `AgentClient` against a stdlib `node:http` mock server,
+covering run/replay/resume/signal, SSE stream parsing, checkpoint
+serialization, and error handling:
+
+```bash
+npm test   # builds, then runs node --test test/*.test.mjs
+```
+
+End-to-end coverage against a real `chidori serve` binary lives in the Python
+SDK integration tests (`sdk/python/tests/`).

package/dist/agent-env.d.ts

@@ -0,0 +1,38 @@
+/**
+ * Ambient types for the virtual `chidori:agent` module.
+ *
+ * Agent and tool files written for the Chidori runtime import their authoring
+ * types — and, in the global style, the `chidori`/`run` values — from
+ * `chidori:agent`:
+ *
+ * ```ts
+ * import type { Chidori, ToolDefinition } from "chidori:agent";
+ * // …or the global style:
+ * import { chidori, run } from "chidori:agent";
+ * ```
+ *
+ * There is no installable package behind that specifier. It is a virtual module
+ * the runtime injects, much like `node:fs` or `bun:test`: the runtime strips the
+ * import and supplies the real values when it executes the file, so nothing is
+ * resolved from npm. Crucially, the unrelated `chidori` package on npm can never
+ * be pulled in by mistake, because `chidori:agent` is not a registry name at all.
+ *
+ * This file exists only so editors and `tsc` can type agent files. Pull it into
+ * a project with a triple-slash reference at the top of an agent or tool file:
+ *
+ * ```ts
+ * /// <reference types="@1kbirds/chidori/agent-env" />
+ * import type { Chidori } from "chidori:agent";
+ * ```
+ *
+ * or once, project-wide, via tsconfig `compilerOptions.types`:
+ * `["@1kbirds/chidori/agent-env"]`.
+ *
+ * NOTE: this must stay a script file (no top-level `import`/`export`). A bare
+ * `declare module "chidori:agent"` is a *global* ambient declaration; adding a
+ * top-level `export` would turn it into a module augmentation and fail to
+ * resolve, since there is no real `chidori:agent` module to augment.
+ */
+declare module "chidori:agent" {
+    export * from "@1kbirds/chidori/agent";
+}

package/dist/agent-env.js

@@ -0,0 +1 @@
+"use strict";

package/dist/agent.d.ts

@@ -0,0 +1,487 @@
+/**
+ * Type declarations for TypeScript agents executed inside the Chidori runtime.
+ *
+ * These are authoring-time types only. The runtime injects the concrete
+ * `chidori` host object when it evaluates an agent or tool module.
+ */
+export type AgentJson = null | boolean | number | string | AgentJson[] | {
+    [key: string]: AgentJson;
+};
+export type JsonObject = {
+    [key: string]: AgentJson;
+};
+export interface JsonSchema {
+    type?: "object" | "array" | "string" | "number" | "integer" | "boolean" | "null";
+    description?: string;
+    properties?: Record<string, JsonSchema>;
+    items?: JsonSchema;
+    required?: string[];
+    enum?: AgentJson[];
+    default?: AgentJson;
+    additionalProperties?: boolean | JsonSchema;
+    [keyword: string]: unknown;
+}
+export interface ToolDefinition {
+    name: string;
+    description: string;
+    parameters: JsonSchema & {
+        type: "object";
+    };
+}
+export type PromptStreamType = "progress" | "draft" | "subagent" | "final" | (string & {});
+/** Provider prompt-cache lifetime for a cached prefix. */
+export type CacheTtl = "5m" | "1h";
+export interface PromptOptions {
+    type?: PromptStreamType;
+    system?: string;
+    model?: string;
+    maxTokens?: number;
+    max_tokens?: number;
+    maxTurns?: number;
+    max_turns?: number;
+    temperature?: number;
+    tools?: string[];
+    format?: "json" | (string & {});
+    stream?: boolean;
+    /**
+     * Prompt-cache posture. Defaults to on (`"5m"`): the runtime marks the
+     * stable request head (system, tools, conversation prefix) so providers
+     * bill repeated prefixes at the cached rate. `false` disables marking for
+     * this call; `"1h"` (or `{ ttl: "1h" }`) requests the extended TTL.
+     * Caching never changes a response — only how it is billed.
+     */
+    cache?: boolean | CacheTtl | {
+        ttl?: CacheTtl;
+    };
+}
+/** Structured response from `Context.respond()` — mirrors the provider turn. */
+export interface LlmResponseJson {
+    content: string;
+    blocks: AgentJson[];
+    tool_calls: {
+        id: string;
+        name: string;
+        input: AgentJson;
+    }[];
+    stop_reason: string;
+    input_tokens: number;
+    output_tokens: number;
+    cache_creation_tokens: number;
+    cache_read_tokens: number;
+}
+/** Options for `Context.compact()` — explicit, opt-in window compaction. */
+export interface CompactOptions {
+    /** How many of the newest conversation turns to keep verbatim (default 2). */
+    keepTurns?: number;
+    /**
+     * Skip compaction (pure no-op, no host call) while `estimateTokens()` is at
+     * or under this budget — lets a loop call `compact()` unconditionally.
+     */
+    budgetTokens?: number;
+    /** Model for the summarization prompt (defaults like `prompt()`). */
+    model?: string;
+    /** System instructions for the summarizer (a faithful-brief default). */
+    instructions?: string;
+    /** `maxTokens` for the summarization prompt. */
+    maxTokens?: number;
+    /** Cache posture for the summarization prompt (see `PromptOptions.cache`). */
+    cache?: boolean | CacheTtl | {
+        ttl?: CacheTtl;
+    };
+    /** TTL of the fresh cache breakpoint placed on the summary (default "5m"). */
+    ttl?: CacheTtl;
+}
+/**
+ * An immutable, content-addressed, turn-structured prompt context.
+ *
+ * Builder methods return a NEW context that structurally shares this one's
+ * segments — `base.user("a")` and `base.user("b")` are independent and share
+ * `base`'s prefix — which keeps cache prefixes stable and makes forks cheap.
+ * Only `prompt()` / `respond()` perform a durable host call.
+ *
+ * ```ts
+ * const base = chidori.context()
+ *   .system("You are a policy analyst.")
+ *   .doc("corpus", corpusText)
+ *   .cacheBreakpoint("1h");
+ * let ctx = base;
+ * for (const q of questions) {
+ *   ctx = ctx.user(q);
+ *   const { text, context } = await ctx.prompt();
+ *   ctx = context; // assistant turn appended, prefix still shared
+ * }
+ * ```
+ */
+export interface Context {
+    system(text: string): Context;
+    /** Expose registered tools (by name, resolved like `prompt({ tools })`). */
+    tools(names: string[]): Context;
+    /** A large stable reference block, labelled for the trace. */
+    doc(label: string, text: string): Context;
+    user(text: string): Context;
+    assistant(text: string): Context;
+    toolResult(id: string, content: string, isError?: boolean): Context;
+    /**
+     * Freeze everything appended so far as a cacheable prefix (one provider
+     * cache breakpoint). Providers cap breakpoints, so marks are coalesced —
+     * latest wins. Most authors never need this: stable heads are auto-marked.
+     */
+    cacheBreakpoint(ttl?: CacheTtl): Context;
+    /** Send this context; returns the text and the context extended with the
+     * assistant turn (including any internal tool-use exchange). */
+    prompt(options?: PromptOptions): Promise<{
+        text: string;
+        context: Context;
+    }>;
+    /** Single structured turn for author-driven tool loops. */
+    respond(options?: PromptOptions): Promise<{
+        response: LlmResponseJson;
+        context: Context;
+    }>;
+    /**
+     * Summarize the older conversation turns into one durable summary segment
+     * (via a recorded `prompt` host call, so it replays deterministically) and
+     * return a new context: stable head + summary + fresh cache breakpoint +
+     * the kept newest turns. Never automatic — compaction changes what the
+     * model sees, so it is always an explicit author decision. Returns this
+     * context unchanged (without a host call) when there is nothing to compact
+     * or the context is within `budgetTokens`.
+     */
+    compact(options?: CompactOptions): Promise<Context>;
+    /** Stable content hash of the request this context would assemble. */
+    digest(options?: PromptOptions): string;
+    /** Rough local token estimate for window budgeting. */
+    estimateTokens(): number;
+}
+/** One recorded exchange in a {@link Conversation} transcript. */
+export interface ConversationTurn {
+    role: "user" | "assistant";
+    text: string;
+}
+/** Options for `chidori.conversation()`. */
+export interface ConversationOptions {
+    /** System prompt — frozen once as the conversation's cacheable prefix. */
+    system?: string;
+    /** Tool names available on every turn (resolved like `prompt({ tools })`). */
+    tools?: string[];
+    /** Default stream label for each turn's prompt (default `"final"`). */
+    type?: PromptStreamType;
+    /** Default model for each turn (a per-turn override still wins). */
+    model?: string;
+    /** Default output token cap for each turn. */
+    maxTokens?: number;
+    /** Default sampling temperature for each turn. */
+    temperature?: number;
+    /** Default cache posture for each turn (see {@link PromptOptions.cache}). */
+    cache?: boolean | CacheTtl | {
+        ttl?: CacheTtl;
+    };
+    /** TTL of the cache breakpoint frozen over the system/tools head. */
+    cacheTtl?: CacheTtl;
+    /**
+     * Opt-in window management: when set, each turn first runs the same budgeted
+     * `Context.compact()` — a pure no-op until the running tail exceeds budget,
+     * then the older turns fold into one recorded summary segment.
+     */
+    compact?: CompactOptions;
+}
+/** Options for `Conversation.loop()` — an interactive `input()`-driven loop. */
+export interface ConversationLoopOptions {
+    /** Prompt shown to the human each turn (or a function of the turn index). */
+    prompt?: string | ((turn: number) => string);
+    /** Extra options forwarded to `chidori.input()` (defaults to `{ type: "message" }`). */
+    inputOptions?: InputOptions;
+    /** Words that end the loop, case-insensitive (default `["exit", "quit"]`). */
+    exit?: string | string[];
+    /** Hard cap on the number of exchanges before returning. */
+    maxTurns?: number;
+    /** Skip blank input lines instead of sending them (default `true`). */
+    skipEmpty?: boolean;
+    /** Per-turn prompt options applied to every `say()` in the loop. */
+    turn?: PromptOptions;
+    /** Called with the assistant reply (and the user message) after each turn. */
+    onReply?: (reply: string, message: string) => void | Promise<void>;
+    /** Return `true` after a turn to end the loop (checked after `onReply`). */
+    until?: (message: string, reply: string) => boolean;
+}
+/**
+ * A small stateful wrapper over {@link Context} for the most common shape — a
+ * multi-turn chat assistant. It owns the running context (system + tools frozen
+ * as a cacheable prefix) and threads each turn through it, so you write
+ * `chat.say(message)` instead of re-plumbing `ctx = (await
+ * ctx.user(message).prompt()).context` by hand. Every turn is still one durable
+ * `prompt`/`respond` host call that replays for free.
+ *
+ * ```ts
+ * const chat = chidori.conversation({ system: "You are concise." });
+ * const a = await chat.say("Hi, who are you?");
+ * const b = await chat.say("What can you help with?");
+ * // or drive it interactively:
+ * const transcript = await chat.loop({ prompt: "you>" });
+ * ```
+ */
+export interface Conversation {
+    /** The underlying immutable context, for dropping to the lower-level API. */
+    readonly context: Context;
+    /** Number of completed exchanges (user+assistant pairs) so far. */
+    readonly length: number;
+    /** The transcript so far as plain `{ role, text }` entries. */
+    history(): ConversationTurn[];
+    /** Send one user message; resolves to the assistant's reply text. */
+    say(message: string, options?: PromptOptions): Promise<string>;
+    /**
+     * Like `say()`, but resolves to the structured response (`tool_calls`,
+     * `blocks`) for author-driven tool loops. Append tool results with
+     * `chat.context.toolResult(...)`, then call `say()` again.
+     */
+    respond(message: string, options?: PromptOptions): Promise<LlmResponseJson>;
+    /**
+     * Drive an interactive loop: read a human message via `chidori.input()`
+     * (terminal stdin under `chidori run`, a paused session resume under `chidori
+     * serve`), reply with `say()`, and repeat until the user types an exit word
+     * or `until` returns true. Resolves to the full transcript.
+     */
+    loop(options?: ConversationLoopOptions): Promise<ConversationTurn[]>;
+}
+export interface InputOptions {
+    type?: string;
+    default?: string;
+    choices?: string[];
+}
+/**
+ * Who delivered a signal. `kind` distinguishes a human participant from a peer
+ * agent; `id` is the participant identity; `runId` is set when an agent sends
+ * (its own run id), so agent-to-agent coordination is attributable in the trace.
+ */
+export interface SignalSender {
+    kind: "human" | "agent";
+    id: string;
+    runId?: string;
+}
+/**
+ * A named message delivered into a run mid-flight (`docs/signals.md` §6.1). The
+ * inverse of `input()`: an outside party (human or agent) pushes
+ * `{ name, payload, from }` at an agent-declared listen point. Every signal is
+ * recorded in the call log, so the multiplayer session replays deterministically.
+ */
+export interface Signal<T = AgentJson> {
+    name: string;
+    payload: T;
+    from: SignalSender;
+}
+export interface SignalOptions {
+    /**
+     * Resolve to a {@link SignalTimeout} sentinel after this many milliseconds
+     * instead of waiting forever. The deadline is enforced by the supervising
+     * server while the run idles; the recorded result (signal or sentinel)
+     * replays deterministically. Discriminate with `"timedOut" in result`.
+     */
+    timeoutMs?: number;
+}
+/**
+ * The sentinel a `timeoutMs` listen point resolves to when the deadline passes
+ * with no matching delivery (`docs/signals.md` §16, pinned:
+ * resolve-to-sentinel rather than reject). `name` is the single awaited name,
+ * or `null` for a multi-name `signalAny`.
+ */
+export interface SignalTimeout {
+    name: string | null;
+    payload: null;
+    from: null;
+    timedOut: true;
+}
+export interface ParallelOptions {
+    concurrency?: number;
+}
+/**
+ * One `chidori.branch` variant (`docs/branching-execution.md` §6.1). A branch
+ * runs its own continuation source module from the parent's anchored state —
+ * not a re-run of the parent agent — so `source` is required.
+ */
+export interface BranchVariant {
+    /** Branch label, shown in outcomes and the trace. Defaults to `branch-<k>`. */
+    label?: string;
+    /** Branch source module path, resolved like `callAgent` paths. */
+    source: string;
+    /** State handed to the branch as its run input. Defaults to `{}`. */
+    input?: AgentJson;
+}
+export type BranchStatus = "completed" | "paused" | "failed";
+/** The result of one branch sub-run, returned for comparison (not merged). */
+export interface BranchOutcome<T extends AgentJson = AgentJson> {
+    label: string;
+    /**
+     * `<parent run id>-op<branch seq>-branch-<k>` — identifies the branch
+     * sub-run, including for out-of-band `chidori branch-resume` /
+     * `branch-rerun` against its persisted store.
+     */
+    branchId: string;
+    status: BranchStatus;
+    /** The branch's output, when `status` is `"completed"`. */
+    output?: T;
+    /** What the branch is waiting on, when `status` is `"paused"`. */
+    pendingPrompt?: string;
+    /** The failure message, when `status` is `"failed"`. */
+    error?: string;
+}
+export interface BranchOptions {
+    /**
+     * Maximum branches running live at once (cost cap). Defaults to 1 —
+     * sequential. Higher values run variants in concurrent waves; outcome
+     * order always follows variant order.
+     */
+    concurrency?: number;
+}
+export interface RetryOptions {
+    attempts?: number;
+    delayMs?: number;
+    backoff?: "fixed" | "exponential";
+}
+export interface TryCallResult<T> {
+    ok: boolean;
+    value?: T;
+    error?: string;
+}
+export interface TemplateOptions {
+    source?: "file" | "inline";
+}
+export type MemoryAction = "get" | "set" | "delete" | "list" | "clear";
+export type WorkspaceFileStatus = "complete" | "writing" | "failed";
+export interface WorkspaceEntry {
+    path: string;
+    status: WorkspaceFileStatus;
+    sha256: string;
+    bytes: number;
+    language?: string | null;
+    attempt?: number | null;
+    updatedAt?: string | null;
+}
+export interface WorkspaceListOptions {
+    completeOnly?: boolean;
+}
+export interface WorkspaceWriteOptions {
+    language?: string;
+}
+export interface WorkspaceHost {
+    list(options?: WorkspaceListOptions): Promise<WorkspaceEntry[]>;
+    read(path: string): Promise<string>;
+    write(path: string, content: string, options?: WorkspaceWriteOptions): Promise<WorkspaceEntry>;
+    delete(path: string, reason?: string): Promise<void>;
+    remove(path: string, reason?: string): Promise<void>;
+    manifest(): Promise<AgentJson>;
+}
+export type TypeScriptImportPolicy = "none" | "relative" | "project";
+export type DatePolicy = "disabled" | "fixed" | "host";
+export type RandomPolicy = "disabled" | "seeded" | "host";
+export type MapSetSnapshotPolicy = "reject" | "serialize";
+export interface RuntimePolicyConfig {
+    typescript?: {
+        imports?: TypeScriptImportPolicy;
+    };
+    runtime?: {
+        date?: DatePolicy;
+        random?: RandomPolicy;
+    };
+    snapshot?: {
+        mapsSets?: MapSetSnapshotPolicy;
+    };
+}
+export interface Chidori {
+    workspace: WorkspaceHost;
+    /** Start an immutable multi-turn prompt context (optionally seeded). */
+    context(seed?: {
+        system?: string;
+        tools?: string[];
+    }): Context;
+    /**
+     * Start a multi-turn chat assistant — a stateful wrapper over `context()`
+     * that owns the running dialogue. Send turns with `chat.say(message)` or drive
+     * an interactive `input()` loop with `chat.loop()`.
+     */
+    conversation(options?: ConversationOptions): Conversation;
+    prompt(text: string, options?: PromptOptions): Promise<string>;
+    input(message: string, options?: InputOptions): Promise<string>;
+    /**
+     * Pause at a named listen point until a matching signal is delivered (or one
+     * is already queued in the durable mailbox), then resolve to
+     * `{ name, payload, from }`. The inverse of `input()`: the run idles cheaply
+     * on disk and an outside party delivers via `POST /sessions/{id}/signal`.
+     */
+    signal<T = AgentJson>(name: string): Promise<Signal<T>>;
+    signal<T = AgentJson>(name: string, options: SignalOptions): Promise<Signal<T> | SignalTimeout>;
+    /**
+     * Non-blocking: consume a queued signal of this name if present, else resolve
+     * to `null`. Records the result (value or null) at this seq so replay is
+     * deterministic.
+     */
+    pollSignal<T = AgentJson>(name: string): Promise<Signal<T> | null>;
+    /**
+     * Fan-in: pause until ANY of the named signals is delivered (or one is
+     * already queued in the durable mailbox). Resolves to the bare consumed
+     * signal — its `name` says which fired. Pre-arrived candidates are consumed
+     * in delivery order (lowest `delivery_seq` across the whole name set).
+     */
+    signalAny<T = AgentJson>(names: string[]): Promise<Signal<T>>;
+    signalAny<T = AgentJson>(names: string[], options: SignalOptions): Promise<Signal<T> | SignalTimeout>;
+    /**
+     * Durable value checkpoint: run `fn` once and journal its JSON-serializable
+     * result; on replay/resume the recorded value (or error) is returned without
+     * re-running `fn`. Wrap expensive deterministic computation in a step so a
+     * resumed run does not re-pay it. The callback must be pure, synchronous
+     * compute — host effects (`chidori.*`), captured randomness, filesystem
+     * writes, timers, and async callbacks are refused inside a step.
+     */
+    step<T extends AgentJson = AgentJson>(name: string, fn: () => T): Promise<T>;
+    callAgent<TInput extends AgentJson = JsonObject, TOutput extends AgentJson = AgentJson>(path: string, input?: TInput): Promise<TOutput>;
+    /**
+     * Fork the run into one sub-run per variant from the current anchored state
+     * (the VFS plus each variant's explicit `input`), run each variant's own
+     * source module, and return every outcome so the agent can compare and pick.
+     * The whole fan-out is one recorded durable call: a replay of this run
+     * returns the outcomes from cache without re-running the branches.
+     */
+    branch<T extends AgentJson = AgentJson>(variants: BranchVariant[], options?: BranchOptions): Promise<BranchOutcome<T>[]>;
+    tool<TArgs extends JsonObject = JsonObject, TResult extends AgentJson = AgentJson>(name: string, args?: TArgs): Promise<TResult>;
+    parallel<TTasks extends readonly (() => Promise<unknown>)[]>(tasks: TTasks, options?: ParallelOptions): Promise<{
+        [Index in keyof TTasks]: Awaited<ReturnType<TTasks[Index]>>;
+    }>;
+    retry<T>(fn: () => Promise<T>, options?: RetryOptions): Promise<T>;
+    tryCall<T>(fn: () => Promise<T>): Promise<TryCallResult<T>>;
+    template(pathOrText: string, vars?: JsonObject, options?: TemplateOptions): Promise<string>;
+    log(message: string, fields?: JsonObject): Promise<void>;
+    memory<T extends AgentJson = AgentJson>(action: MemoryAction, key?: string, value?: T, options?: JsonObject): Promise<T | AgentJson[] | null>;
+    checkpoint(label?: string, data?: AgentJson): Promise<void>;
+}
+export type AgentFunction<TInput extends AgentJson = JsonObject, TOutput extends AgentJson = AgentJson> = (input: TInput) => TOutput | Promise<TOutput>;
+export type ToolFunction<TArgs extends JsonObject = JsonObject, TResult extends AgentJson = AgentJson> = (args: TArgs) => TResult | Promise<TResult>;
+/**
+ * The chidori host object — the durable surface your agents and tools call
+ * (`chidori.log`, `chidori.prompt`, `chidori.tool`, `chidori.input`, …).
+ *
+ * Import it for typed access; the runtime strips this import and supplies the
+ * real object at execution time, so there's no actual module dependency (and no
+ * need for a `(input, chidori)` second parameter):
+ *
+ * ```ts
+ * import { chidori, run } from "chidori:agent";
+ * run(async (input: { topic: string }) => {
+ *   await chidori.log("starting", { topic: input.topic });
+ *   return { ok: true };
+ * });
+ * ```
+ *
+ * Accessing it from a plain import outside the runtime throws.
+ */
+export declare const chidori: Chidori;
+/**
+ * Define the agent entrypoint. Call it once at the top level of an agent module
+ * with your handler; the runtime invokes the handler with the run input and
+ * uses its return value as the output. This replaces the old "export a function
+ * named `agent`" convention.
+ *
+ * ```ts
+ * import { run } from "chidori:agent";
+ * run(async (input) => ({ greeting: `hello ${input.name}` }));
+ * ```
+ */
+export declare function run<TInput extends AgentJson = JsonObject, TOutput extends AgentJson = AgentJson>(handler: AgentFunction<TInput, TOutput>): void;