open-agents-ai 0.61.0 → 0.62.1

package/README.md

@@ -40,6 +40,7 @@ An autonomous multi-turn tool-calling agent that reads your code, makes changes,
 - **Autoresearch Swarm** — 5-agent GPU experiment loop during REM sleep: Researcher, Monitor, Evaluator, Critic, Flow Maintainer autonomously run ML training experiments, keep improvements, discard regressions
 - **Live Listen** — bidirectional voice communication with real-time Whisper transcription
 - **Live Voice Session** — `/listen` with `/voice` enabled spawns a cloudflared tunnel with a real-time WebSocket audio endpoint. A floating presence UI shows live transcription, connected users, and audio visualization. Echo cancellation prevents TTS feedback loops
+- **Call Sub-Agent** — each WebSocket caller gets a dedicated AgenticRunner for low-latency voice-to-voice loops, with admin/public access tiers and bidirectional activity sharing with the main agent
 - **Telegram Voice** — `/voice` enabled via Telegram forwards TTS audio as voice messages alongside text responses. Incoming voice messages are auto-transcribed and handled as text
 - **Neural TTS** — hear what the agent is doing via GLaDOS or Overwatch ONNX voices, with personality-driven expressiveness
 - **Personality Core** — SAC framework-based style control (concise/balanced/verbose/pedagogical) that shapes agent response depth, voice expressiveness, and system prompt behavior
@@ -1051,6 +1052,34 @@ When `/voice` is enabled and the Telegram bridge is active:
 
 Cloudflared is automatically installed at startup alongside other dependencies (moondream, tesseract, transcribe-cli). The install is non-blocking and runs in the background.
 
+### Call Sub-Agent Architecture
+
+Each WebSocket caller in a live voice session gets a **dedicated AgenticRunner** — a fully independent agent instance that handles the voice-to-text-to-LLM-to-TTS-to-reply pipeline with minimal latency.
+
+**Access tiers** — callers connect at one of two privilege levels:
+
+| Tier | URL | Tool Access | Max Turns |
+|------|-----|-------------|-----------|
+| **Admin** | `wss://…?key=<session-key>` | Full tool set (12 tools: file read/write/edit, shell, grep, glob, list directory, web search/fetch, memory read/write/search) | 15 |
+| **Public** | `wss://…` (no key) | Read-only tools (6 tools: file read, grep, glob, list directory, memory read/search) | 5 |
+
+The **session key** is a `crypto.randomBytes(16)` hex string generated per TUI session and displayed in the terminal when the voice session starts. Passing it as the `?key=` URL parameter on the WebSocket connection upgrades the caller to admin access.
+
+**ActivityFeed** — the main TUI agent and all call sub-agents share a bidirectional ring buffer (max 100 entries). Tool calls and results from call sub-agents surface in the main terminal waterfall, and the main agent's activity is visible to connected callers. Each entry carries timestamp, source (main/call), sourceId, tool name, success status, and a summary. Admin callers see verbose timestamped activity; public callers see surface-level summaries.
+
+**Per-client lifecycle** — on WebSocket connect, a `CallSubAgent` is instantiated with its own `AgenticRunner`, `OllamaAgenticBackend`, and conversation history. Transcripts are queued FIFO if the agent is mid-response, ensuring nothing is dropped. On disconnect, the sub-agent is disposed and removed from the active client map.
+
+### Content-Aware Voice Narration
+
+The stochastic narration engine generates spoken descriptions of what the agent is doing for TTS output. Instead of preset phrases, it uses:
+
+- **Variant pools** — 6-10 phrasings per tool per personality tier (terse/conversational/chatty), selected randomly with no back-to-back repeats
+- **Context modifiers** — tracks session state (consecutive errors, file revisits, progress beats) to add natural transitions like "Third time's the charm" or "Coming back to"
+- **Content digests** — extracts key details from actual tool result content (ETH balances, test results, error messages, wallet addresses, status tags, version numbers) and weaves them into the spoken narration. Instead of "Got it", the agent says "Got it — 2.5 ETH, address 0x9fe7F838..." or "That worked, 42 tests passed"
+- **Cross-tool context** — the digest from a tool result optionally carries forward into the next tool call description, so the agent can say "Checking that file, following up on 2.5 ETH" instead of repeating a generic opener
+- **Personality scaling** — terse mode (level 1-2) uses short functional descriptions; conversational (3) adds natural phrasing; chatty (4-5) adds theatrical commentary and content references
+- **Natural silence** — on bland successes without notable content, ~40% of the time the narration is skipped entirely for a more natural rhythm
+
 ## Personality Core — SAC Framework Style Control
 
 The personality system controls how the agent communicates — from silent operator to teacher mode. It's based on the **SAC framework** (arXiv:2506.20993) which models personality along five behavioral intensity dimensions rather than binary trait toggles.

package/dist/index.js

@@ -26999,6 +26999,7 @@ function resetNarrationContext() {
   narration.lastFile = "";
   narration.filesSeen.clear();
   narration.lastVariantIdx = {};
+  narration.lastResultDigest = "";
 }
 function pick(key, variants) {
   if (variants.length === 1)
@@ -27234,31 +27235,102 @@ function describeToolCall(toolName, args, personality = 2) {
       break;
     }
   }
-  if (!prefix)
-    return base;
-  if (prefix.endsWith(". ") || prefix.endsWith("! "))
-    return prefix + base;
-  if (prefix && base.length > 0) {
-    return prefix + base.charAt(0).toLowerCase() + base.slice(1);
+  let result;
+  if (!prefix) {
+    result = base;
+  } else if (prefix.endsWith(". ") || prefix.endsWith("! ")) {
+    result = prefix + base;
+  } else if (base.length > 0) {
+    result = prefix + base.charAt(0).toLowerCase() + base.slice(1);
+  } else {
+    result = base;
   }
-  return base;
+  if (narration.lastResultDigest && personality >= 3 && Math.random() < 0.4) {
+    const ctx = narration.lastResultDigest;
+    const bridges = tier === "chatty" ? [` \u2014 earlier I saw ${ctx}`, `, building on what we found: ${ctx}`, `. Given that we got ${ctx}`] : [` \u2014 saw ${ctx} earlier`, `, following up on ${ctx}`];
+    result += pick("ctx_lastresult", bridges);
+    narration.lastResultDigest = "";
+  }
+  return result;
 }
-function describeToolResult(toolName, success, personality = 2) {
+function extractResultDigest(toolName, content) {
+  if (!content || content.length < 5)
+    return "";
+  const text = content.slice(0, 2e3);
+  const nuggets = [];
+  const ethMatch = text.match(/([\d.]+)\s*ETH/i);
+  if (ethMatch)
+    nuggets.push(`${ethMatch[1]} ETH`);
+  const tokenMatch = text.match(/([\d,.]+)\s*(USDC|USDT|DAI|BTC|SOL|MATIC|tokens?)\b/i);
+  if (tokenMatch && !nuggets.length)
+    nuggets.push(`${tokenMatch[1]} ${tokenMatch[2]}`);
+  const addrMatch = text.match(/(0x[0-9a-fA-F]{8})[0-9a-fA-F]+/);
+  if (addrMatch)
+    nuggets.push(`address ${addrMatch[1]}...`);
+  const httpMatch = text.match(/(?:status|code)[:\s]*(\d{3})\b/i);
+  if (httpMatch)
+    nuggets.push(`status ${httpMatch[1]}`);
+  const testPass = text.match(/(\d+)\s*(?:tests?\s*)?pass(?:ed|ing)?/i);
+  const testFail = text.match(/(\d+)\s*(?:tests?\s*)?fail(?:ed|ing|ures?)?/i);
+  if (testPass || testFail) {
+    const parts = [];
+    if (testPass)
+      parts.push(`${testPass[1]} passed`);
+    if (testFail)
+      parts.push(`${testFail[1]} failed`);
+    nuggets.push(parts.join(", "));
+  }
+  const errMatch = text.match(/(?:error|Error|ERROR)[:\s]+(.{5,50}?)(?:\n|$)/);
+  if (errMatch)
+    nuggets.push(`error: ${errMatch[1].trim()}`);
+  const statusMatch = text.match(/\[(running|complete|completed|failed|pending|success|stopped|error)\]/i);
+  if (statusMatch)
+    nuggets.push(statusMatch[1].toLowerCase());
+  const versionMatch = text.match(/\bv?(\d+\.\d+\.\d+)\b/);
+  if (versionMatch && !nuggets.some((n) => n.includes(versionMatch[1]))) {
+    nuggets.push(`version ${versionMatch[1]}`);
+  }
+  const matchCount = text.match(/(\d+)\s*(?:matches|results|occurrences)/i);
+  if (matchCount)
+    nuggets.push(`${matchCount[1]} matches`);
+  const sizeMatch = text.match(/([\d.]+)\s*(GB|MB|KB|bytes)\b/i);
+  if (sizeMatch)
+    nuggets.push(`${sizeMatch[1]} ${sizeMatch[2]}`);
+  const gasMatch = text.match(/gas[:\s]*([\d,]+)/i);
+  if (gasMatch)
+    nuggets.push(`gas ${gasMatch[1]}`);
+  const exitMatch = text.match(/exit\s*(?:code)?[:\s]*(\d+)/i);
+  if (exitMatch && exitMatch[1] !== "0")
+    nuggets.push(`exit code ${exitMatch[1]}`);
+  const digest = nuggets.slice(0, 3).join(", ");
+  return digest.length > 100 ? digest.slice(0, 97) + "..." : digest;
+}
+function describeToolResult(toolName, success, personality = 2, resultContent) {
   if (toolName === "task_complete")
     return "";
   const tier = getTier(personality);
+  const digest = resultContent ? extractResultDigest(toolName, resultContent) : "";
+  if (digest) {
+    narration.lastResultDigest = digest;
+  }
   if (success) {
     narration.consecutiveErrors = 0;
-    if (personality >= 3 && Math.random() < 0.4)
+    if (!digest && personality >= 3 && Math.random() < 0.4)
       return "";
-    return pick(`result_ok_${tier}`, RESULT_SUCCESS_VARIANTS[tier] ?? RESULT_SUCCESS_VARIANTS.terse);
+    const base = pick(`result_ok_${tier}`, RESULT_SUCCESS_VARIANTS[tier] ?? RESULT_SUCCESS_VARIANTS.terse);
+    if (digest && personality >= 2) {
+      const connectors = tier === "chatty" ? [` \u2014 ${digest}`, `, I see ${digest}`, `. Looks like ${digest}`, `, showing ${digest}`] : tier === "conv" ? [` \u2014 ${digest}`, `, ${digest}`, `. Shows ${digest}`] : [`: ${digest}`];
+      return base + pick("digest_conn_ok", connectors);
+    }
+    return base;
   }
   narration.consecutiveErrors++;
   narration.totalErrors++;
-  if (narration.consecutiveErrors >= 3) {
-    return pick(`result_multifail_${tier}`, RESULT_MULTI_FAIL_VARIANTS[tier] ?? RESULT_MULTI_FAIL_VARIANTS.terse);
+  const failBase = narration.consecutiveErrors >= 3 ? pick(`result_multifail_${tier}`, RESULT_MULTI_FAIL_VARIANTS[tier] ?? RESULT_MULTI_FAIL_VARIANTS.terse) : pick(`result_fail_${tier}`, RESULT_FAIL_VARIANTS[tier] ?? RESULT_FAIL_VARIANTS.terse);
+  if (digest && personality >= 2) {
+    return `${failBase} \u2014 ${digest}`;
   }
-  return pick(`result_fail_${tier}`, RESULT_FAIL_VARIANTS[tier] ?? RESULT_FAIL_VARIANTS.terse);
+  return failBase;
 }
 function describeTaskComplete(summary, completed, personality = 2) {
   const truncated = summary.length > 300 ? summary.slice(0, 300) + "..." : summary;
@@ -27848,7 +27920,8 @@ Error: ${err instanceof Error ? err.message : String(err)}`);
       lastTool: "",
       lastFile: "",
       filesSeen: /* @__PURE__ */ new Set(),
-      lastVariantIdx: {}
+      lastVariantIdx: {},
+      lastResultDigest: ""
     };
     FILE_READ_VARIANTS = {
       terse: (f) => [`Reading ${f}`, `Opening ${f}`, `Loading ${f}`],
@@ -34542,8 +34615,8 @@ ${entry.fullContent}`
             const sizeStr = resultLen > 0 ? ` | ${resultLen.toLocaleString()} chars (~${Math.ceil(resultLen / 4).toLocaleString()} tokens)` : "";
             renderVerbose(`${event.toolName ?? "unknown"}: ${durStr}${sizeStr}`);
           }
-          if (voice?.enabled && !(event.success ?? true)) {
-            const desc = describeToolResult(event.toolName ?? "unknown", false, vLevel);
+          if (voice?.enabled) {
+            const desc = describeToolResult(event.toolName ?? "unknown", event.success ?? false, vLevel, event.content ?? void 0);
             if (desc) {
               renderVoiceText(desc);
               voice.speak(desc);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-agents-ai",
-  "version": "0.61.0",
+  "version": "0.62.1",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",