niahere 0.4.3 → 0.4.5

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "niahere",
-  "version": "0.4.3",
+  "version": "0.4.5",
   "description": "A personal AI assistant daemon — chat, scheduled jobs, persona system, extensible via skills.",
   "type": "module",
   "scripts": {

package/src/channels/index.ts

@@ -1,5 +1,5 @@
 import type { Channel } from "../types";
-import { registerChannel, getFactories, trackStarted, clearStarted } from "./registry";
+import { registerChannel, getFactories, trackStarted, clearStarted, getStarted } from "./registry";
 import { log } from "../utils/log";
 import { getConfig } from "../utils/config";
 import { createTelegramChannel } from "./telegram";
@@ -25,10 +25,12 @@ export interface StartResult {
   failed: string[];
 }
 
-export async function startChannels(): Promise<StartResult> {
+export async function startChannels(only?: readonly string[]): Promise<StartResult> {
+  const onlySet = only ? new Set(only) : null;
   const pending = getFactories()
     .map((factory) => factory())
-    .filter((ch): ch is Channel => ch !== null);
+    .filter((ch): ch is Channel => ch !== null)
+    .filter((ch) => onlySet === null || onlySet.has(ch.name));
 
   if (pending.length === 0) return { started: [], failed: [] };
 
@@ -60,6 +62,42 @@ export async function startChannels(): Promise<StartResult> {
   return { started, failed };
 }
 
+/**
+ * Bring the running channel set back in line with configuration. Starts any
+ * configured-but-not-running channels and stops any running-but-unconfigured
+ * ones. A no-op when already in sync, so it is safe to call on a timer.
+ *
+ * This is the recovery path for the boot-time race where channels fail to
+ * start because Postgres isn't ready yet (channel `.start()` reads the DB):
+ * `startChannels` abandons the failed channels with no retry, so without
+ * reconciliation Nia stays alive but deaf on every channel until a manual
+ * restart. The alive monitor calls this every healthy heartbeat.
+ */
+export async function reconcileChannels(): Promise<StartResult> {
+  const wanted = getConfiguredChannelNames();
+  const running = getStarted();
+  const runningNames = new Set(running.map((ch) => ch.name));
+  const wantedSet = new Set(wanted);
+
+  const missing = wanted.filter((name) => !runningNames.has(name));
+  const extra = running.filter((ch) => !wantedSet.has(ch.name));
+  if (missing.length === 0 && extra.length === 0) return { started: [], failed: [] };
+
+  log.warn({ missing, extra: extra.map((ch) => ch.name) }, "channels out of sync, reconciling");
+
+  // A channel was removed from config. stopChannels() tears down the whole
+  // registry (it stops the shared Twilio server and clears all tracking), so a
+  // partial stop isn't possible — rebuild the full configured set instead.
+  if (extra.length > 0) {
+    await stopChannels(running);
+    return wanted.length > 0 ? startChannels() : { started: [], failed: [] };
+  }
+
+  // Only additions: start just the missing channels so healthy ones stay
+  // connected and one persistently-failing channel can't thrash the rest.
+  return startChannels(missing);
+}
+
 export function getConfiguredChannelNames(): string[] {
   const { channels } = getConfig();
   if (!channels.enabled) return [];

package/src/channels/sms.ts

@@ -131,8 +131,7 @@ class SmsChannel implements Channel {
 
   private async sendTo(remoteE164: string, body: string): Promise<void> {
     if (!this.twilio.sid || !this.twilio.secret) {
-      log.warn("sms: twilio sid/secret missing, cannot send");
-      return;
+      throw new Error("sms: twilio sid/secret missing, cannot send");
     }
     try {
       const res = await twilioSendMessage({
@@ -147,6 +146,10 @@ class SmsChannel implements Channel {
       log.info({ to: remoteE164, sid: res.messageSid, status: res.status }, "sms: sent");
     } catch (err) {
       log.error({ err, to: remoteE164 }, "sms: send failed");
+      // Rethrow so deliver() propagates the failure: the inbound handler falls
+      // back to an error reply, and the send_message tool reports "Failed to
+      // send" instead of falsely claiming delivery.
+      throw err;
     }
   }
 

package/src/channels/whatsapp.ts

@@ -121,7 +121,10 @@ class WhatsAppChannel implements Channel {
       const state = await this.getState(from);
       chainLock(state, async () => {
         const newState = await this.restartChat(from);
-        await this.sendTextTo(from, `New conversation started (room ${this.roomPrefix(from)}-${newState.roomIndex}).`);
+        await this.sendTextTo(
+          from,
+          `New conversation started (room ${this.roomPrefix(from)}-${newState.roomIndex}).`,
+        ).catch((err) => log.warn({ err, from }, "whatsapp: failed to send reset confirmation"));
       });
       return new Response(EMPTY_TWIML, { status: 200, headers: { "Content-Type": "text/xml" } });
     }
@@ -231,7 +234,7 @@ class WhatsAppChannel implements Channel {
   // --- Outbound ---
 
   private async sendTextTo(remoteE164: string, body: string): Promise<void> {
-    if (!this.canSend(remoteE164)) return;
+    if (!this.canSend(remoteE164)) throw new Error("whatsapp: cannot send (missing creds or outside 24h window)");
     const converted = toWhatsAppMarkdown(body);
     const chunks = chunkText(converted, CHUNK_LIMIT);
     for (const chunk of chunks) {
@@ -240,14 +243,14 @@ class WhatsAppChannel implements Channel {
   }
 
   private async sendMediaTo(remoteE164: string, data: Buffer, mimeType: string, filename?: string): Promise<void> {
-    if (!this.canSend(remoteE164)) return;
+    if (!this.canSend(remoteE164)) throw new Error("whatsapp: cannot send (missing creds or outside 24h window)");
     const ext = filename ? extOf(filename) : undefined;
     let mediaUrl: string;
     try {
       mediaUrl = await getTwilioServer().serveMedia(new Uint8Array(data), mimeType, ext);
     } catch (err) {
       log.error({ err }, "whatsapp: serveMedia failed");
-      return;
+      throw err;
     }
     await this.postMessage(remoteE164, "", [mediaUrl]);
   }

package/src/chat/engine.ts

@@ -37,11 +37,10 @@ export async function createChatEngine(opts: EngineOptions): Promise<ChatEngine>
   const { room, channel, resume, mcpServers } = opts;
   let systemPrompt = buildSystemPrompt("chat", channel);
 
-  // Inject recent session summaries for continuity
+  // Recent session summaries for continuity. Appended AFTER persona selection
+  // below — the employee/agent branches replace systemPrompt wholesale, so
+  // injecting here would be discarded for those contexts.
   const sessionContext = await getSessionContext(room);
-  if (sessionContext) {
-    systemPrompt += "\n\n" + sessionContext;
-  }
 
   // Context overrides: employee > agent > job > default
   let cwd = homedir();
@@ -86,6 +85,11 @@ export async function createChatEngine(opts: EngineOptions): Promise<ChatEngine>
     }
   }
 
+  // Continuity summaries — appended after persona so employee/agent contexts keep them.
+  if (sessionContext) {
+    systemPrompt += "\n\n" + sessionContext;
+  }
+
   // Watch mode: inject behavior into system prompt
   if (opts.watchBehavior) {
     const { channel: watchChannel, behavior } = opts.watchBehavior;
@@ -312,6 +316,7 @@ export async function createChatEngine(opts: EngineOptions): Promise<ChatEngine>
           log.warn({ room, to: backends[backendIndex]!.name }, "chat provider down, failing over to next backend");
           await teardown(); // close the dead session so ensureSession opens the next backend
           sessionId = null; // a cross-backend session id is meaningless; start fresh
+          userSaved = false; // re-save the user turn under the new backend's session
           continue;
         }
         break;

package/src/core/alive.ts

@@ -1,6 +1,7 @@
 import { log } from "../utils/log";
 import { getConfig } from "../utils/config";
 import { getSql, closeDb } from "../db/connection";
+import { reconcileChannels } from "../channels";
 import { getFailures, type Check } from "./health";
 
 const HEARTBEAT_INTERVAL = 60_000; // 60s
@@ -164,6 +165,23 @@ async function runRecoveryAgent(failures: Check[]): Promise<{ recovered: boolean
   }
 }
 
+/**
+ * Restart any configured channel that isn't currently running. Only acts when
+ * the running set drifts from config (the boot-time postgres race is the common
+ * cause), and notifies the user once when it recovers channels.
+ */
+async function reconcileDeadChannels(): Promise<void> {
+  try {
+    const { started } = await reconcileChannels();
+    if (started.length > 0) {
+      log.warn({ started }, "alive: restarted dead channels");
+      await notifyUser(`Channels were down and have been restarted: ${started.join(", ")}.`);
+    }
+  } catch (err) {
+    log.warn({ err }, "alive: channel reconcile failed");
+  }
+}
+
 async function heartbeat(): Promise<void> {
   const failures = await getFailures();
   const failureNames = failures.map((f) => f.name);
@@ -176,6 +194,10 @@ async function heartbeat(): Promise<void> {
     }
     lastFailures = [];
     recoveryAttempted = false;
+    // The DB is healthy, so any channel that failed to start (e.g. it lost the
+    // boot-time race against Postgres) can be brought back now. No-op when the
+    // running channels already match config.
+    await reconcileDeadChannels();
     return;
   }
 

package/src/core/daemon.ts

@@ -8,8 +8,7 @@ import { isRunning, readPid, removePid, writePid } from "../utils/pid";
 import { ActiveEngine, Job } from "../db/models";
 import { runMigrations } from "../db/migrate";
 import { closeDb, getSql } from "../db/connection";
-import { registerAllChannels, startChannels, stopChannels, getStarted, getConfiguredChannelNames } from "../channels";
-import type { Channel } from "../types";
+import { registerAllChannels, startChannels, stopChannels, getStarted, reconcileChannels } from "../channels";
 import { startScheduler, stopScheduler, recomputeAllNextRuns } from "./scheduler";
 import { startAlive, stopAlive } from "./alive";
 import { createNiaMcpServer } from "../mcp/server";
@@ -282,13 +281,13 @@ export async function runDaemon(): Promise<void> {
   // composition root) so the endpoint module stays free of the handler chain.
   await startMcpEndpoint(NIA_TOOLS);
 
-  // Register and start channels
+  // Register and start channels. The running set is tracked in the channel
+  // registry (getStarted()); the alive monitor reconciles it if any channel
+  // fails to start (e.g. it lost the boot-time race against Postgres).
   registerAllChannels();
-  let channels: Channel[] = [];
   const config = getConfig();
   if (config.channels.enabled) {
-    const result = await startChannels();
-    channels = result.started;
+    await startChannels();
   } else {
     log.info("channels disabled (channels_enabled: false)");
   }
@@ -353,32 +352,7 @@ export async function runDaemon(): Promise<void> {
   process.on("SIGHUP", async () => {
     log.info("received SIGHUP, reloading config");
     resetConfig();
-
-    const running = getStarted();
-    const wantedNames = getConfiguredChannelNames();
-    const runningNames = running.map((channel) => channel.name).sort();
-    const wantChannels = wantedNames.length > 0;
-    const haveChannels = running.length > 0;
-    const needsReconcile =
-      wantChannels &&
-      haveChannels &&
-      (wantedNames.length !== runningNames.length || wantedNames.sort().some((name, i) => name !== runningNames[i]));
-
-    if (wantChannels && !haveChannels) {
-      log.info("SIGHUP: starting channels");
-      const result = await startChannels();
-      channels = result.started;
-    } else if (!wantChannels && haveChannels) {
-      log.info("SIGHUP: stopping channels");
-      await stopChannels(running);
-      channels = [];
-    } else if (needsReconcile) {
-      log.info({ wantedNames, runningNames }, "SIGHUP: reconciling channels");
-      await stopChannels(running);
-      const result = await startChannels();
-      channels = result.started;
-    }
-
+    await reconcileChannels();
     await recomputeAllNextRuns().catch(() => {});
   });
 
@@ -394,7 +368,7 @@ export async function runDaemon(): Promise<void> {
     stopAlive();
     stopScheduler();
     stopMcpEndpoint();
-    await stopChannels(channels);
+    await stopChannels(getStarted());
 
     try {
       if (force) {

package/src/core/scheduler.ts

@@ -40,10 +40,10 @@ async function tick(): Promise<void> {
 
   for (const job of dueJobs) {
     if (!job.always && !isWithinActiveHours()) {
-      try {
-        const nextRun = computeNextRun(job.scheduleType, job.schedule, config.timezone, new Date());
-        if (nextRun) await Job.markRun(job.name, nextRun).catch(() => {});
-      } catch {}
+      // Leave next_run_at untouched so the job stays due and fires as soon as
+      // active hours resume. Rescheduling here would advance a cron job whose
+      // only fire time sits outside the window to the next (also-outside)
+      // occurrence — starving it forever.
       log.info({ job: job.name }, "scheduler: skipping — outside active hours");
       continue;
     }

package/src/db/models/message.ts

@@ -26,31 +26,6 @@ export async function updateDeliveryStatus(id: number, status: DeliveryStatus):
   await sql`UPDATE messages SET delivery_status = ${status} WHERE id = ${id}`;
 }
 
-export async function getUndelivered(
-  room?: string,
-): Promise<Array<{ id: number; room: string; content: string; createdAt: string }>> {
-  const sql = getSql();
-  const rows = room
-    ? await sql`
-        SELECT id, room, content, created_at
-        FROM messages
-        WHERE delivery_status = 'failed' AND is_from_agent = true AND room = ${room}
-        ORDER BY created_at ASC
-      `
-    : await sql`
-        SELECT id, room, content, created_at
-        FROM messages
-        WHERE delivery_status = 'failed' AND is_from_agent = true
-        ORDER BY created_at ASC
-      `;
-  return rows.map((r) => ({
-    id: r.id,
-    room: r.room,
-    content: r.content,
-    createdAt: String(r.created_at),
-  }));
-}
-
 export async function getRecent(limit = 20, room?: string): Promise<RecentMessage[]> {
   const sql = getSql();
   const rows = room

package/src/db/models/session.ts

@@ -171,7 +171,8 @@ export async function accumulateMetadata(id: string, resultMeta: Record<string,
       'total_cache_read_tokens',      COALESCE((metadata->>'total_cache_read_tokens')::int, 0)       + (${delta}->>'total_cache_read_tokens')::int,
       'total_cache_creation_tokens',  COALESCE((metadata->>'total_cache_creation_tokens')::int, 0)   + (${delta}->>'total_cache_creation_tokens')::int,
       'message_count',                COALESCE((metadata->>'message_count')::int, 0)                 + 1,
-      'models_used',                  COALESCE(metadata->'models_used', '[]'::jsonb) || ${modelsDelta},
+      'models_used',                  (SELECT COALESCE(jsonb_agg(DISTINCT e), '[]'::jsonb)
+                                         FROM jsonb_array_elements(COALESCE(metadata->'models_used', '[]'::jsonb) || ${modelsDelta}) AS e),
       'channel',                      COALESCE(metadata->>'channel', ${(resultMeta.channel as string) || null})
     )
     WHERE id = ${id}