@chroxify/eunoe 1.2.1 → 1.3.0

package/src/context/recall/rerank.ts

@@ -1,5 +1,3 @@
-import { existsSync, readFileSync } from "node:fs"
-import { paths } from "../../config/paths"
 import { rerankQuestion } from "../prompts"
 import type { RecallChunk, Reranker } from "."
 
@@ -9,12 +7,6 @@ interface RerankAnswer {
 
 const TIMEOUT_MS = 4_000
 
-export function rerankKey(): string | null {
-  const env = process.env.TYPESAFE_API_KEY?.trim()
-  if (env) return env
-  return existsSync(paths.typesafeKey) ? readFileSync(paths.typesafeKey, "utf8").trim() : null
-}
-
 function rerankState(chunks: RecallChunk[], candidates: number[]): string {
   return candidates
     .map((index, n) => {

package/src/{proxy → core}/compaction.ts

@@ -1,14 +1,15 @@
-import type { Config } from "../config"
-import type { Body, Message } from "../context/content"
-import { type CompactOptions, capTurns, clearToolResults, compactMessages, countToolResults } from "../context/fold/strategies"
-import { splitTurns, turnFingerprint } from "../context/fold/turns"
+import { type Body, type Message, quickHash } from "../context/content"
+import { type CompactOptions, capTurns, clearToolResults, compactMessages, countPinned, countToolResults, foldsOpening } from "../context/fold/strategies"
+import { type ReduceOptions, reduceTurn, splitTurns, turnFingerprint } from "../context/fold/turns"
 import { clearedOutput } from "../context/prompts"
+import type { CoreConfig } from "./config"
 import { DEFAULT_RATIO, type Estimator, estimator, targetSize } from "./estimate"
+import type { RewriteNote } from "./rewrite"
 import { resetCut, type ThreadState } from "./state"
 
 export interface CompactInput {
   body: Body
-  config: Config
+  config: CoreConfig
   state: ThreadState
   window: number | null
   guide: string | null
@@ -19,10 +20,28 @@ export interface CompactInput {
   reduce?: CompactOptions["reduce"]
 }
 
+const FOLDED_TURNS = 4096
+
+const folded = new Map<string, Message[]>()
+
+function reduceRemembered(turn: Message[], options: ReduceOptions): Message[] {
+  const key = `${options.evidence ? 1 : 0}:${quickHash(JSON.stringify(turn))}`
+  const known = folded.get(key)
+  if (known) {
+    folded.delete(key)
+    folded.set(key, known)
+    return known
+  }
+  const reduced = reduceTurn(turn, options)
+  folded.set(key, reduced)
+  if (folded.size > FOLDED_TURNS) folded.delete(folded.keys().next().value as string)
+  return reduced
+}
+
 export interface CompactResult {
   messages: Message[]
   stateChanged: boolean
-  note: Record<string, unknown>
+  note: RewriteNote
 }
 
 interface Rolling {
@@ -32,13 +51,15 @@ interface Rolling {
 
 const PROTECTED_OUTPUTS = 5
 
+const STUCK_STEP = 0.1
+
 function settledCut(turns: Message[][], index: number) {
   let cut = index
   while (cut > 1 && JSON.stringify(turns[cut][0].content).includes("[Request interrupted")) cut -= 1
   return cut
 }
 
-function resolveCut(turns: Message[][], state: ThreadState, config: Config, rollingKeep: number | undefined) {
+function resolveCut(turns: Message[][], state: ThreadState, config: CoreConfig, rollingKeep: number | undefined) {
   let cutIndex = state.cut ? turns.findIndex((turn) => turnFingerprint(turn) === state.cut) : -1
   let stateChanged = false
   if (state.cut && cutIndex === -1) {
@@ -71,17 +92,47 @@ function dueRoll(turns: Message[][], cutIndex: number, rolling: Rolling, size: E
   return next >= 1 && next > cutIndex && backlog > rolling.budget ? next : null
 }
 
-function shrinkToTarget(over: () => boolean, step: () => string, actions: string[], rebuild: () => void) {
+function shrinkToTarget(over: () => boolean, step: () => string, actions: string[], rebuild: () => void): boolean {
   while (over()) {
     const action = step()
     actions.push(action)
-    if (action === "stuck") return
+    if (action === "stuck") return true
     rebuild()
   }
+  return false
+}
+
+const holdsStuck = (state: ThreadState, estimate: number, window: number | null) => window !== null && state.stuck !== undefined && estimate <= state.stuck + STUCK_STEP * window
+
+export type StuckSizes = ReturnType<typeof stuckSizes>
+
+const headLength = (preamble: Message[], first: Message[], state: ThreadState) => preamble.length + 1 + (countPinned([first], new Set(state.pins)) > 0 ? first.length : 1)
+
+function stuckSizes(messages: Message[], headEnd: number, liveFrom: number, size: Estimator) {
+  const of = (list: Message[]) => size.of(list) - size.fixed
+  return { fixed: size.fixed, head: of(messages.slice(0, headEnd)), folded: of(messages.slice(headEnd, liveFrom)), live: of(messages.slice(liveFrom)) }
+}
+
+function cutNote(state: ThreadState, turns: Message[][], cutIndex: number, configuredKeep: number, actions: string[]) {
+  const pins = new Set(state.pins)
+  const keptFrom = cutIndex > 0 ? Math.max(1, cutIndex - (state.keep ?? configuredKeep)) : 0
+  const omitted = cutIndex > 0 ? keptFrom - 1 - countPinned(turns.slice(1, keptFrom), pins) : 0
+  return {
+    action: cutIndex > 0 ? ("compact" as const) : ("full" as const),
+    compacted: actions.includes("cut") || actions.includes("roll"),
+    actions: actions.length > 0 ? actions : undefined,
+    cutTurn: cutIndex > 0 ? cutIndex : null,
+    omitted,
+    folded: cutIndex > 0 ? cutIndex - keptFrom - countPinned(turns.slice(keptFrom, cutIndex), pins) + (foldsOpening(turns[0], keptFrom) && countPinned([turns[0]], pins) === 0 ? 1 : 0) : 0,
+    pinned: cutIndex > 0 ? countPinned(turns.slice(0, cutIndex), pins) : 0,
+    host: state.host,
+    keep: state.keep,
+    cleared: state.cleared,
+  }
 }
 
 export function compact({ body, config, state, window, guide, transcriptDir, mayCompact, omittedBefore = 0, decorate = (m) => m, reduce }: CompactInput): CompactResult {
-  const { turns } = splitTurns(body.messages)
+  const { preamble, turns } = splitTurns(body.messages)
   const rolling = config.eval?.rolling ?? (config.mode === "default" ? { keep: config.keep, budget: config.budget } : undefined)
   const resolved = resolveCut(turns, state, config, rolling?.keep)
   let cutIndex = resolved.cutIndex
@@ -89,7 +140,7 @@ export function compact({ body, config, state, window, guide, transcriptDir, may
 
   const configuredKeep = config.keepTurns === "all" ? Number.MAX_SAFE_INTEGER : config.keepTurns
   const ratio = state.ratio ?? DEFAULT_RATIO
-  const options = { guide, omittedBefore, evidence: config.eval?.evidence ?? true, reduce, capChars: config.turnCap === "off" ? undefined : config.turnCap / ratio }
+  const options = { guide, omittedBefore, evidence: config.eval?.evidence ?? true, reduce: reduce ?? reduceRemembered, capChars: config.turnCap === "off" ? undefined : config.turnCap / ratio, pinned: new Set(state.pins) }
   const liveStart = (list: Message[]) =>
     list.length -
     splitTurns(list)
@@ -133,12 +184,16 @@ export function compact({ body, config, state, window, guide, transcriptDir, may
   if (mayCompact && estimate > limit) {
     const current = turns.length - 1
     if (current > 0 && current > cutIndex) moveCut(current, "cut")
-    shrinkToTarget(
-      () => estimate > target,
-      () => shrinkOnce(state, cutIndex, configuredKeep, liveResults),
-      actions,
-      rebuild,
-    )
+    if (!holdsStuck(state, estimate, window)) {
+      const stuck = shrinkToTarget(
+        () => estimate > target,
+        () => shrinkOnce(state, cutIndex, configuredKeep, liveResults),
+        actions,
+        rebuild,
+      )
+      if (stuck) state.stuck = estimate
+      else delete state.stuck
+    }
     stateChanged = true
   }
 
@@ -146,13 +201,8 @@ export function compact({ body, config, state, window, guide, transcriptDir, may
     messages,
     stateChanged,
     note: {
-      action: cutIndex > 0 ? "compact" : "full",
-      compacted: actions.includes("cut") || actions.includes("roll"),
-      actions: actions.length > 0 ? actions : undefined,
-      cutTurn: cutIndex > 0 ? cutIndex : null,
-      omitted: cutIndex > 0 ? Math.max(1, cutIndex - (state.keep ?? configuredKeep)) - 1 : 0,
-      keep: state.keep,
-      cleared: state.cleared,
+      ...cutNote(state, turns, cutIndex, configuredKeep, actions),
+      stuck: state.stuck === undefined ? undefined : stuckSizes(messages, headLength(preamble, turns[0], state), liveStart(messages), size),
       turns: turns.length,
       messagesIn: body.messages.length,
       messagesOut: messages.length,

package/src/core/config.ts

@@ -0,0 +1,31 @@
+export type Mode = "off" | "compact" | "default"
+
+interface EvalConfig {
+  recallOptions?: { maxChars?: number; maxCharsMulti?: number; perQuery?: number }
+  strategy?: "tail" | "guide" | "trim"
+  rolling?: { keep: number; budget: number }
+  forceCut?: boolean
+  trustHeaders?: boolean
+  evidence?: boolean
+  recall?: boolean
+  rerank?: { model: string; candidates?: number; endpoint?: string }
+}
+
+export interface CoreConfig {
+  mode: Mode
+  compactAt: number
+  keepTurns: number | "all"
+  keep: number
+  budget: number
+  turnCap: number | "off"
+  eval?: EvalConfig
+}
+
+export const CORE_DEFAULTS: CoreConfig = {
+  mode: "default",
+  compactAt: 0.9,
+  keepTurns: "all",
+  keep: 3,
+  budget: 100_000,
+  turnCap: 100_000,
+}

package/src/core/fold.ts

@@ -0,0 +1,69 @@
+import type { Message } from "../context/content"
+import { splitTurns, turnFingerprint } from "../context/fold/turns"
+import { threadKeyOf } from "./rewrite"
+import { resetCut, type ThreadState, type ThreadStore } from "./state"
+
+export interface FoldTarget {
+  store: ThreadStore
+  threadId: string
+  messages: Message[]
+}
+
+export interface FoldOptions {
+  at?: number | "latest"
+  pin?: number[]
+}
+
+export interface Folded {
+  threadKey: string
+  cutTurn: number | null
+  pinned: number[]
+}
+
+function located({ store, threadId, messages }: FoldTarget) {
+  const threadKey = threadKeyOf(threadId, messages)
+  if (!threadKey) throw new RangeError("eunoe can only fold a thread that has at least one user turn")
+  return { threadKey, turns: splitTurns(messages).turns, state: store.read(threadKey) }
+}
+
+function checkedTurn(turn: number, turns: Message[][], from: number): number {
+  if (turns.length <= from) throw new RangeError(`this thread has ${turns.length} turn${turns.length === 1 ? "" : "s"}, so there is no turn ${turn} to fold at`)
+  if (!Number.isInteger(turn) || turn < from || turn >= turns.length) throw new RangeError(`turn ${turn} is outside turns ${from}..${turns.length - 1}`)
+  return turn
+}
+
+function folded(threadKey: string, turns: Message[][], state: ThreadState): Folded {
+  const cutTurn = state.cut ? turns.findIndex((turn) => turnFingerprint(turn) === state.cut) : -1
+  const pins = new Set(state.pins)
+  return { threadKey, cutTurn: cutTurn > 0 ? cutTurn : null, pinned: turns.flatMap((turn, i) => (pins.has(turnFingerprint(turn)) ? [i] : [])) }
+}
+
+function withPins(state: ThreadState, turns: Message[][], indices: number[], pinned: boolean) {
+  const pins = new Set(state.pins)
+  for (const index of indices) {
+    const fingerprint = turnFingerprint(turns[checkedTurn(index, turns, 0)])
+    if (pinned) pins.add(fingerprint)
+    else pins.delete(fingerprint)
+  }
+  if (pins.size > 0) state.pins = [...pins].sort()
+  else delete state.pins
+}
+
+export function foldThread(target: FoldTarget, { at = "latest", pin = [] }: FoldOptions = {}): Folded {
+  const { threadKey, turns, state } = located(target)
+  const cut = checkedTurn(at === "latest" ? turns.length - 1 : at, turns, 1)
+  withPins(state, turns, pin, true)
+  resetCut(state)
+  state.cut = turnFingerprint(turns[cut])
+  state.keep = cut - 1
+  state.host = true
+  target.store.save(threadKey, state)
+  return folded(threadKey, turns, state)
+}
+
+export function pinTurns(target: FoldTarget, indices: number[], pinned = true): Folded {
+  const { threadKey, turns, state } = located(target)
+  withPins(state, turns, indices, pinned)
+  target.store.save(threadKey, state)
+  return folded(threadKey, turns, state)
+}

package/src/core/forward.ts

@@ -0,0 +1,192 @@
+import { type Body, wellFormed } from "../context/content"
+import type { Reranker } from "../context/recall"
+import { CORE_DEFAULTS, type CoreConfig } from "./config"
+import { sizeOf } from "./estimate"
+import { type Rewrite, type RewriteNote, rewriteAsync, type ThreadSources } from "./rewrite"
+import type { ThreadStore } from "./state"
+import { inputTokens, recordOverflow, recordUsage, watchUsage } from "./usage"
+import { createWindowLookup, type WindowLookup } from "./window"
+
+export type ForwardNote = RewriteNote | Record<string, unknown>
+
+export type Prepared = Omit<Rewrite, "note"> & { note: ForwardNote }
+
+export interface ForwarderOptions {
+  store: ThreadStore
+  upstream?: string
+  config?: CoreConfig
+  window?: number | null | WindowLookup
+  threadIdOf?: (request: Request, body: Body) => string | null
+  sources?: (threadId: string, request: Request, body: Body) => ThreadSources | undefined
+  reranker?: Reranker
+  prepare?: (request: Request, body: Body, mayCompact: boolean) => Promise<Prepared>
+  record?: (entry: Record<string, unknown>) => void
+  countTokensTimeoutMs?: number
+}
+
+interface Outgoing {
+  note: ForwardNote
+  threadKey: string | null
+  bodyText: string | undefined
+  sentSize: number
+  givenSize: number
+  fixedSize: number
+}
+
+const COUNT_TOKENS_ATTEMPTS = 3
+
+const DROPPED_REQUEST_HEADERS = ["host", "content-length", "accept-encoding"]
+
+const clipped = (error: string) => error.slice(0, 500)
+
+export function claudeCodeSessionId(headers: Headers, body: Body): string | null {
+  const header = headers.get("x-claude-code-session-id")
+  if (header) return header
+  try {
+    return JSON.parse(body.metadata?.user_id ?? "{}").session_id ?? null
+  } catch {
+    return null
+  }
+}
+
+const defaultThreadId = (request: Request, body: Body) => request.headers.get("x-thread-id") ?? claudeCodeSessionId(request.headers, body)
+
+function forwardableHeaders(headers: Headers) {
+  const out = new Headers(headers)
+  out.delete("content-encoding")
+  out.delete("content-length")
+  return out
+}
+
+function defaultPrepare(options: ForwarderOptions, upstream: string): NonNullable<ForwarderOptions["prepare"]> {
+  const threadIdOf = options.threadIdOf ?? defaultThreadId
+  const given = options.window
+  const lookup = typeof given === "function" ? given : createWindowLookup()
+  const windowFor: (model: unknown, headers: Headers) => Promise<number | null> = typeof given === "number" || given === null ? () => Promise.resolve(given) : (model, headers) => lookup(model, headers, upstream)
+  return async (request, body, mayCompact) => {
+    const threadId = threadIdOf(request, body)
+    const window = await windowFor(body.model, request.headers)
+    const sources = threadId ? options.sources?.(threadId, request, body) : undefined
+    return rewriteAsync({ body, threadId, config: options.config ?? CORE_DEFAULTS, store: options.store, sources, window, mayCompact }, options.reranker)
+  }
+}
+
+const passthrough = (bodyText: string | undefined): Outgoing => ({ note: { action: "pass" }, threadKey: null, bodyText, sentSize: 0, givenSize: 0, fixedSize: 0 })
+
+async function outgoing(prepare: NonNullable<ForwarderOptions["prepare"]>, request: Request, givenText: string, mayCompact: boolean): Promise<Outgoing> {
+  try {
+    const given = JSON.parse(givenText) as Body
+    const result = await prepare(request, given, mayCompact)
+    const sentSize = sizeOf(result.body)
+    return {
+      note: result.note,
+      threadKey: result.threadKey,
+      bodyText: result.body !== given ? JSON.stringify(result.body, wellFormed) : givenText,
+      sentSize,
+      givenSize: result.body === given ? sentSize : sizeOf(given),
+      fixedSize: sizeOf({ ...result.body, messages: [] }),
+    }
+  } catch (error) {
+    return { ...passthrough(givenText), note: { action: "error-passthrough", error: String(error) } }
+  }
+}
+
+function createSender(request: Request, target: string, headers: Headers, timeout: number | null, stall: { count: number }) {
+  return async (body: string | undefined): Promise<Response> => {
+    if (timeout === null) return fetch(target, { method: request.method, headers, body })
+    for (;;) {
+      try {
+        return await fetch(target, { method: request.method, headers, body, signal: AbortSignal.timeout(timeout) })
+      } catch (error) {
+        const timedOut = error instanceof Error && error.name === "TimeoutError"
+        if (!timedOut || stall.count + 1 >= COUNT_TOKENS_ATTEMPTS) throw error
+        stall.count += 1
+      }
+    }
+  }
+}
+
+function delivered(answer: Response, out: Outgoing, metered: boolean, path: string, log: (entry: Record<string, unknown>) => void, store: ThreadStore): Response {
+  const { note, threadKey, sentSize, givenSize, fixedSize } = out
+  const response = new Response(answer.body, { status: answer.status, statusText: answer.statusText, headers: forwardableHeaders(answer.headers) })
+  if (note.action === "pass" || !metered) {
+    if (note.action !== "pass" || Object.keys(note).length > 1) log({ path, status: answer.status, ...note })
+    return response
+  }
+  return watchUsage(response, (usage) => {
+    log({
+      path,
+      status: answer.status,
+      thread: threadKey,
+      ...note,
+      input: inputTokens(usage),
+      cacheRead: usage.cache_read_input_tokens ?? 0,
+      cacheWrite: usage.cache_creation_input_tokens ?? 0,
+      uncached: usage.input_tokens ?? 0,
+      charsIn: givenSize,
+      charsOut: sentSize,
+      charsFixed: fixedSize,
+    })
+    if (threadKey) recordUsage(store, threadKey, sentSize, usage)
+  })
+}
+
+export function createForwarder(options: ForwarderOptions): (request: Request) => Promise<Response> {
+  const upstream = options.upstream ?? "https://api.anthropic.com"
+  const prepare = options.prepare ?? defaultPrepare(options, upstream)
+  const timeout = options.countTokensTimeoutMs ?? 10_000
+
+  async function route(request: Request, began: number): Promise<Response> {
+    const stall = { count: 0 }
+    let prep = 0
+    let upload = 0
+    const log = (entry: Record<string, unknown>) => options.record?.({ ...entry, ms: Date.now() - began, prep, upload, ...(stall.count > 0 && { stalled: stall.count }) })
+    const url = new URL(request.url)
+    const headers = new Headers(request.headers)
+    for (const name of DROPPED_REQUEST_HEADERS) headers.delete(name)
+
+    let givenText: string | undefined
+    if (request.method !== "GET" && request.method !== "HEAD") givenText = await request.text()
+    upload = Date.now() - began
+    const isMessages = request.method === "POST" && url.pathname.startsWith("/v1/messages")
+    const isCountTokens = url.pathname.includes("count_tokens")
+
+    const build = () => (givenText && isMessages ? outgoing(prepare, request, givenText, !isCountTokens) : Promise.resolve(passthrough(givenText)))
+    const send = createSender(request, upstream + url.pathname + url.search, headers, isCountTokens ? timeout : null, stall)
+    let out = await build()
+    prep = Date.now() - began - upload
+    let answer = await send(out.bodyText)
+    const errorResponse = (text: string) => new Response(text, { status: answer.status, statusText: answer.statusText, headers: forwardableHeaders(answer.headers) })
+
+    if (answer.status >= 400 && out.note.action !== "pass") {
+      const errorText = await answer.text()
+      const logged = { path: url.pathname, status: answer.status, thread: out.threadKey, ...out.note }
+      const overflow = out.threadKey && !isCountTokens ? recordOverflow(options.store, out.threadKey, out.sentSize, errorText) : null
+      if (!overflow) {
+        log({ ...logged, error: clipped(errorText) })
+        return errorResponse(errorText)
+      }
+      log({ ...logged, overflow, error: clipped(errorText) })
+      out = await build()
+      out = { ...out, note: { ...out.note, healed: overflow } }
+      answer = await send(out.bodyText)
+      if (answer.status >= 400) {
+        const retryError = await answer.text()
+        log({ path: url.pathname, status: answer.status, thread: out.threadKey, ...out.note, error: clipped(retryError) })
+        return errorResponse(retryError)
+      }
+    }
+
+    return delivered(answer, out, isMessages && !isCountTokens, url.pathname, log, options.store)
+  }
+
+  return async (request) => {
+    const began = Date.now()
+    try {
+      return await route(request, began)
+    } catch (error) {
+      options.record?.({ path: new URL(request.url).pathname, status: 502, action: "upstream-unreachable", error: clipped(String(error)), ms: Date.now() - began })
+      return Response.json({ type: "error", error: { type: "api_error", message: `eunoe could not reach the API: ${String(error)}` } }, { status: 502 })
+    }
+  }
+}

package/src/{proxy → core}/handoff.ts

@@ -1,17 +1,13 @@
-import { statSync } from "node:fs"
-import { type Config, claudeConfigDirs } from "../config"
 import { blocks, type Message } from "../context/content"
 import { FOLD_CAP_CHARS, foldedView, sourceChunks } from "../context/fold"
 import { isPrompt } from "../context/fold/turns"
 import { handoffBlock } from "../context/prompts"
 import type { RecallChunk } from "../context/recall"
-import { findSessionByPrefix } from "../transcript/locate"
-import { sessionMessages } from "../transcript/messages"
 import { isInjected, messageFingerprint } from "./fingerprint"
-import { titleOf } from "./sessions"
+import type { ThreadSources } from "./rewrite"
 import type { Handoff, ThreadState } from "./state"
 
-const sourceCache = new Map<string, { mtime: number; chunks: RecallChunk[] }>()
+const chunksOf = new WeakMap<Message[], { label: string; chunks: RecallChunk[] }>()
 
 export const continueToken = (id: string) => `{ee:continue:${id}}`
 
@@ -28,17 +24,23 @@ export function activeHandoffs(handoffs: Record<string, Handoff | null> = {}): R
   return Object.fromEntries(Object.entries(handoffs).filter((entry): entry is [string, Handoff] => Boolean(entry[1])))
 }
 
-export function attachHandoffs(messages: Message[], config: Config, sessionId: string | null, state: ThreadState): boolean {
+function resolveHandoff(id: string, sources: ThreadSources, state: ThreadState): Handoff | null {
+  const found = sources.findHandoff?.(id) ?? null
+  if (!found || !sources.readHandoff) return null
+  const already = Object.values(state.handoffs ?? {}).some((h) => h?.source === found.source)
+  return already ? null : { label: found.label, source: found.source, text: foldedView(sources.readHandoff(found.source), FOLD_CAP_CHARS) }
+}
+
+export function attachHandoffs(messages: Message[], sources: ThreadSources, state: ThreadState): boolean {
   let changed = false
   for (const message of messages) {
     const id = continueIdOf(message)
     if (!id) continue
     const key = messageFingerprint(message)
     if (state.handoffs && key in state.handoffs) continue
+    const handoff = resolveHandoff(id, sources, state)
     state.handoffs ??= {}
-    const file = findSessionByPrefix(id, claudeConfigDirs(config), sessionId ?? undefined)
-    const already = Object.values(state.handoffs).some((h) => h?.source === file)
-    state.handoffs[key] = file && !already ? { label: titleOf(file) ?? id, source: file, text: foldedView(sessionMessages(file), FOLD_CAP_CHARS) } : null
+    state.handoffs[key] = handoff
     changed = true
   }
   return changed
@@ -53,19 +55,17 @@ export function withHandoffs(messages: Message[], handoffs: Record<string, Hando
   })
 }
 
-export function handoffChunks(handoffs: Record<string, Handoff>): RecallChunk[] {
-  const sources = new Map<string, string>()
-  for (const handoff of Object.values(handoffs)) if (!sources.has(handoff.source)) sources.set(handoff.source, handoff.label)
-  return [...sources].flatMap(([source, label]) => {
-    let mtime: number
-    try {
-      mtime = statSync(source).mtimeMs
-    } catch {
-      return []
-    }
-    const cached = sourceCache.get(source)
-    const chunks = cached?.mtime === mtime ? cached.chunks : sourceChunks(sessionMessages(source), label, true)
-    sourceCache.set(source, { mtime, chunks })
+export function handoffChunks(handoffs: Record<string, Handoff>, sources: ThreadSources): RecallChunk[] {
+  const read = sources.readHandoff
+  if (!read) return []
+  const labels = new Map<string, string>()
+  for (const handoff of Object.values(handoffs)) if (!labels.has(handoff.source)) labels.set(handoff.source, handoff.label)
+  return [...labels].flatMap(([source, label]) => {
+    const messages = read(source)
+    const known = chunksOf.get(messages)
+    if (known?.label === label) return known.chunks
+    const chunks = sourceChunks(messages, label, true)
+    chunksOf.set(messages, { label, chunks })
     return chunks
   })
 }

package/src/core/index.ts

@@ -0,0 +1,14 @@
+export type { Block, Body, Message } from "../context/content"
+export type { RecallChunk, Reranker } from "../context/recall"
+export { jevReranker } from "../context/recall/rerank"
+export type { StuckSizes } from "./compaction"
+export { CORE_DEFAULTS, type CoreConfig, type Mode } from "./config"
+export { type Folded, type FoldOptions, type FoldTarget, foldThread, pinTurns } from "./fold"
+export { claudeCodeSessionId, createForwarder, type ForwarderOptions, type ForwardNote, type Prepared } from "./forward"
+export { continueToken } from "./handoff"
+export { type Rewrite, type RewriteInput, type RewriteNote, rewrite, rewriteAsync, type ThreadSources, threadKeyOf } from "./rewrite"
+export { memoryStore, type ThreadState, type ThreadStore } from "./state"
+export { listSubagentRecall, type Subagent } from "./subagents"
+export { renderTranscriptFiles, type TranscriptOptions, transcriptGuide } from "./transcript"
+export { inputTokens, recordOverflow, recordUsage, type Usage, watchUsage } from "./usage"
+export { createWindowLookup, type WindowLookup } from "./window"

package/src/{proxy → core}/recall.ts

@@ -1,11 +1,10 @@
-import { type Body, blocks, resultText } from "../context/content"
+import { type Body, blocks, type Message, resultText } from "../context/content"
 import { livePrompt, type RecallChunk, type Reranker, recallChunks, recallFor, recallForWith, withRecall } from "../context/recall"
-import { sessionMessages } from "../transcript/messages"
-import { subagentChunks } from "../transcript/subagents"
 import { messageFingerprint } from "./fingerprint"
 import { activeHandoffs, handoffChunks } from "./handoff"
 import type { Rewrite, RewriteInput } from "./rewrite"
-import { readState, saveState } from "./state"
+
+const closingTurn: Message = { role: "user", content: "" }
 
 export const wantsRecall = (input: RewriteInput) => (input.config.eval?.recall ?? true) && input.config.mode !== "off"
 
@@ -13,31 +12,30 @@ export const activeRecalls = (recalls: Record<string, string> = {}) => Object.fr
 
 const touched = (input: RewriteInput, result: Rewrite) => result.threadKey !== null && !(result.body === input.body && result.note.action === "pass")
 
-function droppedChunks(input: RewriteInput, result: Rewrite): RecallChunk[] {
-  const transcript = input.transcript?.jsonl ?? null
+function droppedChunks(input: RewriteInput, result: Rewrite, threadKey: string): RecallChunk[] {
+  const sources = input.sources ?? {}
   const kept = new Map<string, string>()
   for (const message of result.body.messages) for (const b of blocks(message)) if (b.type === "tool_result") kept.set(b.tool_use_id, resultText(b))
   const own = recallChunks(input.body.messages, kept)
-  const session = own.length > 0 || !transcript ? own : recallChunks([...sessionMessages(transcript), { role: "user", content: "" }], kept)
-  const withSubagents = transcript ? [...session, ...subagentChunks(transcript)] : session
-  const handoffs = result.threadKey ? activeHandoffs(readState(result.threadKey).handoffs) : {}
-  return [...handoffChunks(handoffs), ...withSubagents]
+  const session = own.length > 0 || !sources.readHistory ? own : recallChunks([...sources.readHistory(), closingTurn], kept)
+  const withSubagents = sources.listSubagentChunks ? [...session, ...sources.listSubagentChunks()] : session
+  return [...handoffChunks(activeHandoffs(input.store.read(threadKey).handoffs), sources), ...withSubagents]
 }
 
-function recallTarget(original: Body, result: Rewrite): { key: string; text: string } | null {
+function recallTarget(original: Body): { key: string; text: string } | null {
   const live = livePrompt(original.messages)
-  return live && result.threadKey ? { key: messageFingerprint(original.messages[live.index]), text: live.text } : null
+  return live ? { key: messageFingerprint(original.messages[live.index]), text: live.text } : null
 }
 
 export function applyRecall(input: RewriteInput, result: Rewrite): Rewrite {
   if (!touched(input, result) || !result.threadKey) return result
-  const state = readState(result.threadKey)
+  const state = input.store.read(result.threadKey)
   state.recalls ??= {}
-  const target = recallTarget(input.body, result)
+  const target = recallTarget(input.body)
   let note = result.note
   if (target && !(target.key in state.recalls)) {
-    state.recalls[target.key] = recallFor(target.text, droppedChunks(input, result), input.config.eval?.recallOptions) ?? ""
-    saveState(result.threadKey, state)
+    state.recalls[target.key] = recallFor(target.text, droppedChunks(input, result, result.threadKey), input.config.eval?.recallOptions) ?? ""
+    input.store.save(result.threadKey, state)
     note = { ...note, recalled: state.recalls[target.key].length }
   }
   const messages = withRecall(result.body.messages, messageFingerprint, activeRecalls(state.recalls))
@@ -46,13 +44,13 @@ export function applyRecall(input: RewriteInput, result: Rewrite): Rewrite {
 
 export async function prefetchFor(input: RewriteInput, result: Rewrite, rerank: Reranker): Promise<number | null> {
   if (!touched(input, result) || !result.threadKey) return null
-  const target = recallTarget(input.body, result)
+  const target = recallTarget(input.body)
   if (!target) return null
-  const state = readState(result.threadKey)
+  const state = input.store.read(result.threadKey)
   state.recalls ??= {}
   if (target.key in state.recalls) return null
   const options = { candidates: input.config.eval?.rerank?.candidates, ...input.config.eval?.recallOptions }
-  state.recalls[target.key] = (await recallForWith(target.text, droppedChunks(input, result), rerank, options)) ?? ""
-  saveState(result.threadKey, state)
+  state.recalls[target.key] = (await recallForWith(target.text, droppedChunks(input, result, result.threadKey), rerank, options)) ?? ""
+  input.store.save(result.threadKey, state)
   return state.recalls[target.key].length
 }