open-agents-ai 0.186.62 → 0.186.63

package/README.md

@@ -23,6 +23,7 @@ npm i -g open-agents-ai && oa
 
 An autonomous multi-turn tool-calling agent that reads your code, makes changes, runs tests, and fixes failures in an iterative loop until the task is complete. First launch auto-detects your hardware and configures the optimal model with expanded context window automatically.
 
+
 ## Table of Contents
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -67,6 +68,10 @@ An autonomous multi-turn tool-calling agent that reads your code, makes changes,
 - [Research Citations](#research-citations)
 - [License](#license)
 
+
+<details id="the-organism-not-the-cortex">
+<summary><strong>The Organism, Not the Cortex</strong> — Why the LLM is one organ inside a larger organism</summary>
+
 ## The Organism, Not the Cortex
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -89,6 +94,12 @@ An LLM is a high-bandwidth associative generative core — closer to a cortex-li
 
 Don't chase larger models. Build the organism around whatever model you have.
 
+</details>
+
+
+<details id="how-it-works">
+<summary><strong>How It Works</strong> — Multi-turn autonomous tool-calling loop in action</summary>
+
 ## How It Works
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -105,6 +116,12 @@ Agent: [Turn 1] file_read(src/auth.ts)
 
 The agent uses tools autonomously in a loop — reading errors, fixing code, and re-running validation until the task succeeds or the turn limit is reached.
 
+</details>
+
+
+<details id="features">
+<summary><strong>Features</strong> — 61 tools, voice, vision, P2P mesh, self-play, COHERE cognitive stack</summary>
+
 ## Features
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -209,6 +226,12 @@ D8AgCTrxpDKD5meJ2bpAfVwcST3NF3EPuy9xczYycnXn
 0x81Ce81F0B6B5928E15d3a2850F913C88D07051ec
 ```
 
+</details>
+
+
+<details id="enterprise--headless-mode">
+<summary><strong>Enterprise & Headless Mode</strong> — REST API, background jobs, JSON output, auth scopes, tool profiles</summary>
+
 ## Enterprise & Headless Mode
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -720,6 +743,12 @@ Open `http://localhost:11435/` in a browser when `oa serve` is running. Zero ext
 
 Free for non-commercial use under CC-BY-NC-4.0. For enterprise/commercial licensing, contact [zoomerconsulting.com](https://zoomerconsulting.com).
 
+</details>
+
+
+<details id="architecture">
+<summary><strong>Architecture</strong> — AgenticRunner core loop with structured context assembly</summary>
+
 ## Architecture
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -742,6 +771,12 @@ User task → assembleContext(c_instr, c_state, c_know) → LLM → tool_calls
 - **Context-aware** — dynamic compaction, Memex archiving, session persistence, model-tier scaling
 - **Brute-force** — optional auto re-engagement when turn limit is hit (keeps going until task_complete or user abort)
 
+</details>
+
+
+<details id="context-engineering">
+<summary><strong>Context Engineering</strong> — C = A(c_instr, c_know, c_tools, c_mem, c_state, c_query) structured assembly</summary>
+
 ## Context Engineering
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -768,6 +803,12 @@ Key design decisions grounded in research:
 
 Research provenance: grounded in "A Survey of Context Engineering for LLMs" (context assembly equation), "Modular Prompt Optimization" (section-local textual gradients), "Reasoning Up the Instruction Ladder" (priority hierarchy), "GEPA" (reflective prompt evolution), and "Prompt Flow Integrity" (least-privilege context passing).
 
+</details>
+
+
+<details id="model-tier-awareness">
+<summary><strong>Model-Tier Awareness</strong> — Dynamic tool sets, prompts, and limits that scale with model size</summary>
+
 ## Model-Tier Awareness
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -805,6 +846,12 @@ All context-dependent values scale automatically with the actual context window
 | Tool output cap | 2K-8K chars (scales with context) |
 | File read limits | 80-120 line cap for small/medium context windows |
 
+</details>
+
+
+<details id="auto-expanding-context-window">
+<summary><strong>Auto-Expanding Context Window</strong> — RAM/VRAM detection creates optimized model variants automatically</summary>
+
 ## Auto-Expanding Context Window
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -820,6 +867,12 @@ On startup and `/model` switch, Open Agents detects your RAM/VRAM and creates an
 | 8GB+ | 8K tokens |
 | < 8GB | 4K tokens |
 
+</details>
+
+
+<details id="tools-61">
+<summary><strong>Tools (61)</strong> — File I/O, shell, web, vision, memory, agents, COHERE, P2P, x402</summary>
+
 ## Tools (61)
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -928,6 +981,12 @@ The agent has 4 web tools. Pick the right one:
 
 **Structured extraction**: Pass `extract_schema='{"price": "number", "name": "string"}'` to `web_crawl` for best-effort regex-based field extraction from page content.
 
+</details>
+
+
+<details id="ralph-loop--iteration-first-design">
+<summary><strong>Ralph Loop — Iteration-First Design</strong> — Iterative retry loop where errors become learning data</summary>
+
 ## Ralph Loop — Iteration-First Design
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -954,6 +1013,12 @@ The loop tracks iteration history, generates completion reports saved to `.aiwg/
 /ralph-abort      # Cancel running loop
 ```
 
+</details>
+
+
+<details id="task-control">
+<summary><strong>Task Control</strong> — Pause, stop, resume, destroy, and session context persistence</summary>
+
 ## Task Control
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -995,6 +1060,12 @@ When you launch `oa` in a workspace that has saved session context from a previo
 
 Type `y` to restore — the previous session context will be prepended to your next task, giving the agent full continuity. Type `n` (or anything else) to start fresh. The prompt only appears on fresh starts, not on `/update` resumes (which auto-restore context).
 
+</details>
+
+
+<details id="cohere-cognitive-framework">
+<summary><strong>COHERE Cognitive Framework</strong> — 8-layer cognitive stack with distributed inference, identity, and reflection</summary>
+
 ## COHERE Cognitive Framework
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1075,6 +1146,12 @@ The identity kernel maintains a persistent self-model across sessions, the refle
 | L8 | Darwin Gödel Machine: Open-Ended Self-Improvement (2025) | [arxiv:2505.22954](https://arxiv.org/abs/2505.22954) |
 | L8 | i-MENTOR: Intrinsic Motivation Exploration (2025) | [arxiv:2505.17621](https://arxiv.org/abs/2505.17621) |
 
+</details>
+
+
+<details id="agent-immune-system--constraint-enforcement--pressure-resistance">
+<summary><strong>Agent Immune System — Constraint Enforcement & Pressure Resistance</strong> — Behavioral constraints, pressure-aware decision gates, and audit logging</summary>
+
 ## Agent Immune System — Constraint Enforcement & Pressure Resistance
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1136,6 +1213,12 @@ User (frustrated): "fix this broken shit"
   → Model fixes the architecture instead of adding a prompt hack
 ```
 
+</details>
+
+
+<details id="context-compaction--research-backed-memory-management">
+<summary><strong>Context Compaction — Research-Backed Memory Management</strong> — 6 compaction strategies, Memex archive, SNR tracking, deep context mode</summary>
+
 ## Context Compaction — Research-Backed Memory Management
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1264,6 +1347,12 @@ Compaction summaries include:
 
 This ensures the agent can resume coherently after compaction without re-reading files or re-running commands.
 
+</details>
+
+
+<details id="personality-core--sac-framework-style-control">
+<summary><strong>Personality Core — SAC Framework Style Control</strong> — Five-dimension behavioral intensity from silent operator to teacher mode</summary>
+
 ## Personality Core — SAC Framework Style Control
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1314,6 +1403,12 @@ The personality system draws on:
 - **Linear Personality Probing** ([arXiv:2512.17639](https://arxiv.org/abs/2512.17639)) — Prompt-level steering completely dominates activation-level interventions
 - **The Prompt Report** ([arXiv:2406.06608](https://arxiv.org/abs/2406.06608)) — Positive framing outperforms negated instructions for behavioral control
 
+</details>
+
+
+<details id="emotion-engine--affective-state-modulation">
+<summary><strong>Emotion Engine — Affective State Modulation</strong> — Circumplex affect model with valence, arousal, dominance axes</summary>
+
 ## Emotion Engine — Affective State Modulation
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1378,6 +1473,12 @@ The emotion system is informed by peer-reviewed and preprint research:
 
 8. **EmotionBench** — Huang et al. ([arXiv:2308.03656](https://arxiv.org/abs/2308.03656), 2023). LLMs cannot maintain emotional state across turns implicitly — argues for explicit external mood state representation (which this engine implements).
 
+</details>
+
+
+<details id="voice-feedback-tts">
+<summary><strong>Voice Feedback (TTS)</strong> — GLaDOS, Overwatch, Kokoro, LuxTTS voice clone with emotion-driven prosody</summary>
+
 ## Voice Feedback (TTS)
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1571,6 +1672,12 @@ The stochastic narration engine generates spoken descriptions of what the agent
 - **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
 
+</details>
+
+
+<details id="listen-mode--live-bidirectional-audio">
+<summary><strong>Listen Mode — Live Bidirectional Audio</strong> — Real-time Whisper transcription with hands-free auto-submit</summary>
+
 ## Listen Mode — Live Bidirectional Audio
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1609,6 +1716,12 @@ The `transcribe-cli` dependency auto-installs in the background on first use. On
 
 **File transcription**: Drag-and-drop audio/video files (`.mp3`, `.wav`, `.mp4`, `.mkv`, etc.) onto the terminal to transcribe them. Results are saved to `.oa/transcripts/`.
 
+</details>
+
+
+<details id="vision--desktop-automation-moondream">
+<summary><strong>Vision & Desktop Automation (Moondream)</strong> — Local VLM for screenshots, point-and-click, browser automation, OCR</summary>
+
 ## Vision & Desktop Automation (Moondream)
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1797,6 +1910,12 @@ Supports `apt` (Debian/Ubuntu), `dnf` (Fedora), `pacman` (Arch), and `brew` (mac
 - Moondream Station (local) — runs entirely on your machine, no API keys needed
 - Moondream Cloud API — set `MOONDREAM_API_KEY` for cloud inference
 
+</details>
+
+
+<details id="interactive-tui">
+<summary><strong>Interactive TUI</strong> — REPL with slash commands, mid-task steering, animated metrics bar</summary>
+
 ## Interactive TUI
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -1902,6 +2021,12 @@ The steering sub-agent uses the same model and backend as the main agent with `m
 - **LATS** (Zhou et al., 2024) — mid-execution replanning with user-provided value signals improves task completion on complex multi-step problems
 - **AutoGen** (Wu et al., 2023) — human-in-the-loop patterns work best when user messages are expanded into structured instructions, reducing ambiguity for the primary agent
 
+</details>
+
+
+<details id="telegram-bridge--sub-agent-per-chat">
+<summary><strong>Telegram Bridge — Sub-Agent Per Chat</strong> — Per-chat sub-agents with admin passthrough, media handling, and streaming</summary>
+
 ## Telegram Bridge — Sub-Agent Per Chat
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2035,6 +2160,12 @@ The bridge automatically handles Telegram's rate limits (HTTP 429) with exponent
 
 **Combined with blessed mode** — `/full-send-bless` + `/telegram` creates a persistent, always-on agent that processes Telegram messages around the clock while keeping the model warm.
 
+</details>
+
+
+<details id="x402-payment-rails--nexus-p2p">
+<summary><strong>x402 Payment Rails & Nexus P2P</strong> — EVM wallets, EIP-3009 USDC transfers, metered inference, budget policies</summary>
+
 ## x402 Payment Rails & Nexus P2P
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2094,6 +2225,12 @@ nexus(action='budget_set', auto_approve_below='0.01')  # Auto-approve micropayme
 - All outbound messages scanned for key material before sending
 - Keys NEVER appear in tool output, logs, or LLM context
 
+</details>
+
+
+<details id="sponsored-inference--share-your-gpu-with-the-world">
+<summary><strong>Sponsored Inference — Share Your GPU With the World</strong> — 5-step wizard to share models via secure branded relay</summary>
+
 ## Sponsored Inference — Share Your GPU With the World
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2160,6 +2297,12 @@ Consumer OA ──→ Cloudflared Tunnel ──→ Sponsor Proxy ──→ Ollam
 
 The tunnel fix uses debounced restarts with exponential cooldown (10s → 20s → 40s), stopping auto-restart after 3 consecutive failures to prevent Cloudflare rate limiting. Progress indicators emit every 5 seconds during startup, and specific error messages are shown for common failure modes (ENOENT, port conflict, 429, DNS).
 
+</details>
+
+
+<details id="cohere-distributed-mind">
+<summary><strong>COHERE Distributed Mind</strong> — Multi-node mesh with NATS pub/sub, peer review, collective learning</summary>
+
 ## COHERE Distributed Mind
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2226,6 +2369,12 @@ Inbound queries are scanned for prompt injection attempts before processing:
 - Remote constraints from peer nodes (CM-07, published every 5 minutes)
 - Blocked queries increment `queriesErrors` and are silently dropped
 
+</details>
+
+
+<details id="dream-mode--creative-idle-exploration">
+<summary><strong>Dream Mode — Creative Idle Exploration</strong> — NREM/REM sleep cycles with autoresearch swarm on GPU</summary>
+
 ## Dream Mode — Creative Idle Exploration
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2302,6 +2451,12 @@ If the Python scripts are invoked directly (without `uv run`), they self-bootstr
 
 If no GPU is detected, the REM stage falls back to the standard multi-agent creative exploration (Visionary + Pragmatist + Cross-Pollinator + Synthesizer).
 
+</details>
+
+
+<details id="blessed-mode--infinite-warm-loop">
+<summary><strong>Blessed Mode — Infinite Warm Loop</strong> — Keep model warm in VRAM, auto-cycle tasks, Default Mode Network</summary>
+
 ## Blessed Mode — Infinite Warm Loop
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2341,6 +2496,12 @@ Each DMN cycle runs a lightweight LLM agent (15 max turns, temperature 0.4) with
 
 **Research basis**: Reflexion ([arXiv:2303.11366](https://arxiv.org/abs/2303.11366)), Self-Rewarding LMs ([arXiv:2401.10020](https://arxiv.org/abs/2401.10020)), Generative Agents ([arXiv:2304.03442](https://arxiv.org/abs/2304.03442)), STOP ([arXiv:2310.02226](https://arxiv.org/abs/2310.02226)), Voyager ([arXiv:2305.16291](https://arxiv.org/abs/2305.16291))
 
+</details>
+
+
+<details id="docker-sandbox--collective-intelligence">
+<summary><strong>Docker Sandbox & Collective Intelligence</strong> — Container isolation, multi-agent testbed, self-play loop</summary>
+
 ## Docker Sandbox & Collective Intelligence
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2458,6 +2619,12 @@ Nodes share identity kernel updates via `nexus.cohere.kernel.delta` on NATS. Ado
 4. **Tool Use = Quality** — Agents using `web_search` produced current, verifiable data. Non-tool responses were generic.
 5. **Identity Divergence** — Different task exposure → different specializations. Intern gained `web-research` from heavy search; Director gained nothing (still loading).
 
+</details>
+
+
+<details id="code-sandbox">
+<summary><strong>Code Sandbox</strong> — Isolated JS, Python, Bash, TypeScript execution in subprocess or Docker</summary>
+
 ## Code Sandbox
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2476,6 +2643,12 @@ Supports JavaScript, TypeScript, Python, and Bash. Two execution modes:
 - **Subprocess** (default) — runs in a child process with timeout and output limits
 - **Docker** — runs in an isolated container when `docker` is available
 
+</details>
+
+
+<details id="structured-data-tools">
+<summary><strong>Structured Data Tools</strong> — Generate and parse CSV, TSV, JSON, Markdown tables, Excel files</summary>
+
 ## Structured Data Tools
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2504,6 +2677,12 @@ Agent: read_structured_file(path="report.md")
 
 Detects binary formats (XLSX, PDF, DOCX) and suggests conversion tools.
 
+</details>
+
+
+<details id="multi-provider-web-search">
+<summary><strong>Multi-Provider Web Search</strong> — DuckDuckGo, Tavily, and Jina AI with auto-detection</summary>
+
 ## Multi-Provider Web Search
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2521,6 +2700,12 @@ export TAVILY_API_KEY=tvly-...   # Enable Tavily (optional)
 export JINA_API_KEY=jina_...     # Enable Jina AI (optional)
 ```
 
+</details>
+
+
+<details id="task-templates">
+<summary><strong>Task Templates</strong> — Specialized system prompts for code, document, analysis, and plan tasks</summary>
+
 ## Task Templates
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2534,6 +2719,12 @@ Set a task type to get specialized system prompts, recommended tools, and output
 /task-type plan       # Planning — emphasizes steps, dependencies, risks
 ```
 
+</details>
+
+
+<details id="human-expert-speed-ratio">
+<summary><strong>Human Expert Speed Ratio</strong> — Real-time Exp: Nx gauge calibrated across 47 tool baselines</summary>
+
 ## Human Expert Speed Ratio
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2575,6 +2766,12 @@ Color coding: green (2x+ faster), yellow (1-2x, comparable), red (<1x, slower th
 
 All 47 tools have calibrated baselines ranging from 3s (`task_stop`) to 180s (`codebase_map`). Unknown tools default to 20s.
 
+</details>
+
+
+<details id="cost-tracking--session-metrics">
+<summary><strong>Cost Tracking & Session Metrics</strong> — Token cost estimation for 15+ providers with LLM-as-judge evaluation</summary>
+
 ## Cost Tracking & Session Metrics
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2591,6 +2788,12 @@ Cost tracking supports 15+ providers including Groq, Together AI, OpenRouter, Fi
 
 Work evaluation uses five task-type-specific rubrics (code, document, analysis, plan, general) scoring correctness, completeness, efficiency, code quality, and communication on a 1-5 scale.
 
+</details>
+
+
+<details id="configuration">
+<summary><strong>Configuration</strong> — CLI flags, env vars, config files, project context, and .oa/ directory</summary>
+
 ## Configuration
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2623,6 +2826,12 @@ Create `AGENTS.md`, `OA.md`, or `.open-agents.md` in your project root for agent
 └── pending-task.json  # Saved task state for /stop and /update resume
 ```
 
+</details>
+
+
+<details id="model-support">
+<summary><strong>Model Support</strong> — Qwen3.5-122B primary target, any Ollama or OpenAI-compatible model</summary>
+
 ## Model Support
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2637,6 +2846,12 @@ oa --backend vllm --backend-url http://localhost:8000/v1 "add tests"
 oa --backend-url http://10.0.0.5:11434 "refactor auth"
 ```
 
+</details>
+
+
+<details id="supported-inference-providers">
+<summary><strong>Supported Inference Providers</strong> — 14 providers from local Ollama to Groq, Chutes, OpenRouter, and P2P mesh</summary>
+
 ## Supported Inference Providers
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2795,6 +3010,12 @@ When you've used multiple endpoints, the agent automatically builds a failover c
 
 No configuration needed — the cascade is built from your endpoint usage history. Works across local Ollama, cloud providers, and P2P peers.
 
+</details>
+
+
+<details id="evaluation-suite">
+<summary><strong>Evaluation Suite</strong> — 23 web nav + 46 coding + 35 enterprise tasks with pass^k reliability</summary>
+
 ## Evaluation Suite
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -2986,6 +3207,12 @@ The PoT (Program-of-Thought) guidance achieves **100% code generation rate** —
 - **~80 tokens of prompt additions** (PoT math guidance + search-when-uncertain) took the eval from 41.2% to 100% across all tiers — no fine-tuning required.
 - 4B models match 9B/27B on structured domain tasks (healthcare, DevOps, e-commerce) but need search tools for specialized regulatory knowledge.
 
+</details>
+
+
+<details id="aiwg-integration">
+<summary><strong>AIWG Integration</strong> — AI-augmented SDLC with 85+ agents, structured memory, and traceability</summary>
+
 ## AIWG Integration
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -3005,6 +3232,12 @@ oa "analyze this project's SDLC health and set up documentation"
 | **85+ Agents** | Specialized AI personas (Test Engineer, Security Auditor, API Designer) |
 | **Traceability** | @-mention system links requirements to code to tests |
 
+</details>
+
+
+<details id="research-citations">
+<summary><strong>Research Citations</strong> — 32 papers (2023-2026) grounding self-play, memory, identity, and containers</summary>
+
 ## Research Citations
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -3058,6 +3291,12 @@ The COHERE collective intelligence system, self-play idle loop, identity evoluti
 | LatentMAS: Latent-Space Collaboration | [2511.20639](https://arxiv.org/abs/2511.20639) | Nov 2025 | Future: 4x faster, 70-84% token reduction |
 | Agent-Kernel Microkernel Architecture | [2512.01610](https://arxiv.org/abs/2512.01610) | Dec 2025 | Architecture: 10k agent coordination |
 
+</details>
+
+
+<details id="license">
+<summary><strong>License</strong> — CC BY-NC 4.0 with enterprise licensing available</summary>
+
 ## License
 
 <div align="right"><a href="#top">back to top</a></div>
@@ -3065,3 +3304,5 @@ The COHERE collective intelligence system, self-play idle loop, identity evoluti
 [Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)](https://creativecommons.org/licenses/by-nc/4.0/)
 
 Free for non-commercial use. For enterprise/commercial licensing, contact [zoomerconsulting.com](https://zoomerconsulting.com).
+
+</details>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-agents-ai",
-  "version": "0.186.62",
+  "version": "0.186.63",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",