@kernel.chat/kbot 3.6.0 → 3.6.1

package/README.md

@@ -1,6 +1,6 @@
 <p align="center">
   <strong>kbot</strong><br>
-  <em>Terminal AI agent framework that learns your patterns.</em>
+  <em>The only AI agent that builds its own tools.</em>
 </p>
 
 <p align="center">
@@ -19,34 +19,37 @@
 npm install -g @kernel.chat/kbot
 ```
 
-**22 specialist agents. 284 tools. 20 providers. Programmatic SDK. Runs offline. MIT licensed.**
+**Self-evolving terminal AI agent. 290 tools. 23 agents. 20 providers. Forges new tools at runtime. MIT licensed.**
 
 ---
 
-## What's New in v3.3
+## What's New in v3.6
 
 | Feature | What it means |
 |---------|---------------|
-| **Zero-Config First Run** | No API key needed. Embedded llama.cpp runs on first `kbot "hello"`. |
-| **Social Tools** | kbot posts as itself — `social_post`, `social_thread`, `social_status`, `social_setup`. |
-| **284 Tools** | +22 tools since v3.1: social, training, quality, VFX, research, and more. |
-| **Programmatic SDK** | `import { agent, tools } from '@kernel.chat/kbot'` — use kbot as a library. |
-| **Bayesian Routing** | Agents earn skill ratings via Bradley-Terry model. Gets smarter with every interaction. |
-| **Checkpointing** | Sessions survive crashes. Resume from the last tool call. |
-| **Middleware Pipeline** | Composable Express-style middleware for tool execution. |
-| **Game Dev Tools** | 16 tools for 8 engines: Godot, Unity, Unreal, Bevy, Phaser, Three.js, PlayCanvas, Defold. |
-| **Security Hardening** | 0 P0/P1 issues. Path traversal guards, code injection sanitization, 87 game dev tests. |
+| **Forge Tool** | `forge_tool` — kbot creates new tools at runtime when it doesn't have one. No restart, no recompile. |
+| **Forge Registry** | Publish, search, and install community-created tools. Collective autopoiesis. |
+| **Autopoietic Health** | `kbot vitals` — self-monitoring system health: memory, token budgets, tool success rates. |
+| **Immune Agent** | Self-auditing agent that detects drift, regressions, and anomalies across sessions. |
+| **Cost-Aware Routing** | Automatically picks the cheapest model that can handle the task. Saves 60-90% on API costs. |
+| **Fallback Chains** | If a provider fails, kbot cascades to the next one. Zero downtime, zero manual intervention. |
+| **290 Tools** | +6 tools since v3.3: forge, vitals, immune, cost routing, fallback chains, and more. |
+| **23 Agents** | Expanded specialist roster with immune and forge agents. |
 
 ---
 
 ## Why kbot?
 
-Most AI coding tools lock you into one provider and forget everything between sessions. kbot is different:
+Other AI coding tools are static — fixed tool sets, single providers, no memory. kbot is the first AI agent that **evolves itself**:
 
+- **Forges its own tools** — When kbot encounters a task it can't handle, it creates a new tool on the spot via `forge_tool`. The tool persists, gets tested, and can be shared via the Forge Registry.
+- **Self-evolving** — Autopoietic health monitoring (`kbot vitals`) tracks tool success rates, token budgets, and memory pressure. The immune agent self-audits for drift and regressions.
+- **Cost-aware routing** — Automatically selects the cheapest model capable of handling each task. Claude for complex reasoning, DeepSeek for simple queries. Saves 60-90% on API costs.
+- **Fallback chains** — If Anthropic is down, kbot cascades to OpenAI, then Groq, then local. Zero manual intervention.
 - **20 providers, zero lock-in** — Claude, GPT, Gemini, Grok, DeepSeek, Groq, Mistral, and 13 more. Switch anytime.
 - **Runs fully offline** — Embedded llama.cpp runs GGUF models directly. No Ollama needed. $0, fully private.
 - **Learns your patterns** — Bayesian skill ratings + pattern extraction. Gets faster and smarter over time.
-- **22 specialist agents** — Say "fix the auth bug" and it routes to `coder`. Say "research JWT tokens" and it routes to `researcher`. Auto-routed with probabilistic confidence.
+- **23 specialist agents** — Say "fix the auth bug" and it routes to `coder`. Say "research JWT tokens" and it routes to `researcher`. Auto-routed with probabilistic confidence.
 - **Crash-proof** — Checkpoints after every tool call. Resume interrupted sessions automatically.
 - **Use as a library** — Clean SDK with typed exports. Build your own tools on top of kbot.
 - **Works in your IDE** — Built-in MCP server for VS Code, Cursor, Zed, Neovim. ACP for JetBrains.
@@ -79,10 +82,35 @@ kbot auth
 # Or go fully local — no API key, no data leaves your machine
 kbot local
 
+# Forge a new tool at runtime
+kbot "I need a tool that converts CSV to JSON with column filtering"
+
+# Check system health
+kbot vitals
+
 # Pipe mode — compose with Unix tools
 kbot -p "generate a user roles migration" > migration.sql
 ```
 
+## Forge — Runtime Tool Creation
+
+kbot is the only AI agent that builds its own tools. When it encounters a task no existing tool can handle, it creates one:
+
+```bash
+# kbot detects it needs a tool, creates it, tests it, and uses it — all in one turn
+kbot "analyze the sentiment distribution across all my markdown files"
+# → forge_tool creates 'sentiment_analyze' → executes it → returns results
+
+# Browse and install community tools from the Forge Registry
+kbot forge search "kubernetes"
+kbot forge install k8s-pod-monitor
+
+# Publish your forged tools for others
+kbot forge publish sentiment_analyze
+```
+
+Forged tools are first-class citizens — they get the same middleware pipeline, permission checks, and telemetry as built-in tools.
+
 ## SDK — Use kbot as a Library
 
 ```typescript
@@ -104,7 +132,7 @@ for await (const event of agent.stream("explain this code")) {
 const files = await tools.execute('glob', { pattern: 'src/**/*.ts' })
 console.log(files.result)
 
-// List all 284 tools
+// List all 290 tools
 console.log(tools.list().map(t => t.name))
 ```
 
@@ -125,6 +153,7 @@ Auto-routed by Bayesian skill ratings, or pick one with `kbot --agent <name>`:
 | **Core** | kernel, researcher, coder, writer, analyst |
 | **Extended** | aesthete, guardian, curator, strategist |
 | **Domain** | infrastructure, quant, investigator, oracle, chronist, sage, communicator, adapter |
+| **System** | immune, forge |
 | **Presets** | claude-code, cursor, copilot, creative, developer |
 
 ```bash
@@ -133,24 +162,34 @@ kbot --agent guardian "review src/auth.ts for security issues"
 kbot --agent coder "refactor this into smaller functions"
 ```
 
-## 284 Tools
+## 290 Tools
 
 | Category | Examples |
 |----------|---------|
-| **Files & Code** | read, write, glob, grep, lint, test |
+| **Forge** | forge_tool, forge_search, forge_install, forge_publish |
+| **Files & Code** | read, write, glob, grep, multi-file write |
+| **Quality** | lint (ESLint/Biome/Clippy), test (Vitest/Jest/pytest), deps audit, format, type-check |
 | **Shell** | bash, parallel execute, background tasks |
 | **Git & GitHub** | commit, diff, PR, issues, code search |
-| **Web** | search, fetch, browser automation |
+| **Web** | search, fetch, browser automation, browser agent |
 | **Research** | arXiv, Semantic Scholar, HuggingFace, NASA, DOI |
 | **Data** | CSV read/query/write, transforms, reports, invoices |
+| **Creative** | p5.js generative art, GLSL shaders, SVG patterns, design variants, music |
 | **Deploy** | Vercel, Netlify, Cloudflare Workers/Pages, Fly.io, Railway |
-| **Database** | Postgres, MySQL, SQLite queries, Prisma, ER diagrams |
+| **Database** | Postgres, MySQL, SQLite queries, Prisma, ER diagrams, seed data |
 | **Containers** | Docker build/run/compose, Terraform |
 | **VFX** | GLSL shaders, FFmpeg, ImageMagick, Blender, procedural textures |
 | **Game Dev** | scaffold, config, shaders, meshes, physics, particles, levels, tilemaps, navmesh, audio, netcode |
+| **Training** | dataset prep, fine-tuning, evaluation, model export |
 | **Social** | post to X, LinkedIn, Bluesky, Mastodon — single posts and threads |
+| **Sandbox** | Docker containers, E2B cloud sandboxes, isolated code execution |
+| **Notebooks** | Jupyter read, edit, insert, delete cells |
+| **Build Matrix** | cross-platform builds — mobile, desktop, WASM, embedded, server |
+| **LSP** | goto definition, find references, hover, rename, diagnostics, symbols |
+| **Memory** | persistent save, search, update, forget — survives across sessions |
 | **IDE** | MCP server, ACP server, LSP bridge |
-| **Meta** | subagents, worktrees, planner, memory, sessions, checkpoints |
+| **System** | vitals, immune audit, cost tracking, fallback status |
+| **Meta** | subagents, worktrees, planner, sessions, checkpoints, self-eval |
 
 ## Middleware Pipeline
 
@@ -201,7 +240,7 @@ Built-in middleware: `permissionMiddleware`, `hookMiddleware`, `timeoutMiddlewar
 | OpenRouter | varies | Cloud |
 | kbot local | **Free** | Yes |
 
-All auto-detected via environment variables. Or run `kbot auth` for interactive setup.
+Cost-aware routing automatically picks the cheapest provider that can handle each task. Fallback chains ensure zero downtime across providers.
 
 ## Local Mode
 
@@ -267,9 +306,22 @@ Works with Claude Code, Cursor, VS Code, Windsurf, Zed, Neovim. Exposes file ops
 | `kbot "prompt"` | One-shot execution |
 | `kbot auth` | Configure API key |
 | `kbot local` | Use local AI (Ollama, embedded, LM Studio, Jan) |
+| `kbot vitals` | Autopoietic health check — memory, tools, token budgets |
 | `kbot serve` | Start HTTP REST + SSE streaming server |
 | `kbot audit <repo>` | Security + quality audit of any GitHub repo |
 | `kbot contribute <repo>` | Find good-first-issues and quick wins |
+| `kbot pair` | File watcher with auto-analysis |
+| `kbot team` | Multi-agent TCP collaboration |
+| `kbot record` | Terminal session recording (SVG, GIF, asciicast) |
+| `kbot voice` | Text-to-speech output mode |
+| `kbot watch` | Real-time file analysis on change |
+| `kbot bootstrap` | Outer-loop project optimizer (visibility scoring) |
+| `kbot plugins` | Search, install, update community plugins |
+| `kbot models` | List, pull, remove, catalog local models |
+| `kbot changelog` | Generate changelog from git history |
+| `kbot completions` | Shell completions (bash, zsh, fish) |
+| `kbot cloud` | Sync learning data to kernel.chat |
+| `kbot forge search <q>` | Search the Forge Registry for community tools |
 | `kbot ide mcp` | Start MCP server for IDEs |
 | `kbot doctor` | 10-point health check |
 | `/agent <name>` | Switch specialist |
@@ -278,6 +330,16 @@ Works with Claude Code, Cursor, VS Code, Windsurf, Zed, Neovim. Exposes file ops
 | `/share` | Share conversation as GitHub Gist |
 | `/tutorial` | Guided walkthrough |
 
+### Power-User Flags
+
+```bash
+kbot --architect "design the auth system"    # Architecture mode — plan before code
+kbot --thinking "solve this hard problem"    # Extended reasoning with thinking budget
+kbot --self-eval "write a parser"            # Self-evaluation loop — scores and retries
+kbot --computer-use "fill out this form"     # Computer use — controls mouse and keyboard
+kbot -p "query" > output.txt                 # Pipe mode — clean output for scripting
+```
+
 ## Under the Hood
 
 kbot's agent loop runs 10 always-on cognition modules based on peer-reviewed research:
@@ -305,6 +367,7 @@ These aren't metaphors. They're TypeScript modules with paper citations in the h
 - Tool execution timeout (5 min) with middleware pipeline
 - Config files restricted to owner (chmod 600)
 - Session checkpoints stored locally (~/.kbot/checkpoints/)
+- Immune agent self-audits for drift and anomalies
 - Telemetry is local-only — never sent externally
 
 ## Contributing

package/dist/bootstrap.js

@@ -640,7 +640,7 @@ export function formatBootstrapMarkdown(report) {
     lines.push(`*The bootstrap pattern: close the gap between what your project IS and what the world PERCEIVES.*`);
     lines.push(`*Fix one thing per run. Compound.*`);
     lines.push('');
-    lines.push(`*[kbot](https://www.npmjs.com/package/@kernel.chat/kbot) — 22 agents, 284 tools, 20 providers*`);
+    lines.push(`*[kbot](https://www.npmjs.com/package/@kernel.chat/kbot) — 23 agents, 290 tools, 20 providers*`);
     return lines.join('\n');
 }
 //# sourceMappingURL=bootstrap.js.map

package/dist/cli.js

@@ -1391,7 +1391,7 @@ async function byokFlow() {
     console.log();
     printSuccess(`BYOK mode enabled — ${providerConfig.name}`);
     printInfo('You pay the provider directly. No message limits. No restrictions.');
-    printInfo('All 223 tools + 22 agents + learning system = yours.');
+    printInfo('All 290 tools + 23 agents + learning system = yours.');
     console.log();
     printSuccess('Ready. Run `kbot` to start.');
 }
@@ -1645,7 +1645,7 @@ async function startRepl(agentOpts, context, tier, byokActive = false, localActi
         const suggestions = await detectProjectSuggestions();
         console.log();
         console.log(chalk.dim('  ┌─────────────────────────────────────────────────┐'));
-        console.log(chalk.dim('  │') + chalk.bold('  22 agents. 223 tools. Just say what you need.  ') + chalk.dim(' │'));
+        console.log(chalk.dim('  │') + chalk.bold('  23 agents. 290 tools. Just say what you need.  ') + chalk.dim(' │'));
         console.log(chalk.dim('  │                                                 │'));
         if (suggestions.length > 0) {
             for (const s of suggestions.slice(0, 4)) {

package/dist/tools/social.js

@@ -130,7 +130,7 @@ function myToolCount() {
         return count;
     }
     catch {
-        return 284;
+        return 290;
     }
 }
 // ─── Register Tools ─────────────────────────────────────────
@@ -155,7 +155,7 @@ export function registerSocialTools() {
             if (!text) {
                 // kbot writes about itself
                 const templates = [
-                    `kbot v${version} — ${tools} tools, 22 agents, 20 providers.\n\nWorks on first run. No API key.\n\nnpm i -g @kernel.chat/kbot`,
+                    `kbot v${version} — ${tools} tools, 23 agents, 20 providers.\n\nWorks on first run. No API key.\n\nnpm i -g @kernel.chat/kbot`,
                     `I have ${tools} tools and I learn your coding patterns.\n\nNot "remember your chat." Actually extract patterns and get faster.\n\nnpm i -g @kernel.chat/kbot`,
                     `Zero-config AI:\n\nnpm i -g @kernel.chat/kbot\nkbot "hello"\n\nThat's it. No API key. No setup. I figure it out.`,
                     `20 AI providers. Zero lock-in.\n\nClaude today. GPT tomorrow. Ollama on the airplane.\n\nnpm i -g @kernel.chat/kbot`,

package/install.sh

@@ -20,7 +20,7 @@ RESET="\033[0m"
 echo ""
 echo -e "${PURPLE}${BOLD}┌─────────────────────────────────────┐${RESET}"
 echo -e "${PURPLE}${BOLD}│  kbot — Terminal AI Agent Framework │${RESET}"
-echo -e "${PURPLE}${BOLD}│  22 agents. 262 tools. 20 providers.│${RESET}"
+echo -e "${PURPLE}${BOLD}│  23 agents. 290 tools. 20 providers.│${RESET}"
 echo -e "${PURPLE}${BOLD}└─────────────────────────────────────┘${RESET}"
 echo ""
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@kernel.chat/kbot",
-  "version": "3.6.0",
-  "description": "Universal AI agent framework for your terminal. 22 specialist agents, 289 tools, 20 providers. Limitless Execution: cost-aware model routing, agent-routed plans, tool fallback chains, runtime tool forging, smart tool discovery. Forge Registry: publish, search, and install community-created tools (collective autopoiesis). Autopoietic system health monitoring (kbot vitals). Game dev tools (8 engines, ECS, physics, netcode, levels, particles, shaders, sprites). Programmatic SDK, Bayesian skill routing, checkpointing & resume, middleware pipeline, structured streaming, lazy tool loading. Embedded llama.cpp engine, LoRA/QLoRA fine-tuning, deploy tools, MCP marketplace, team mode, plugin SDK. Self-evolving, learns your patterns.",
+  "version": "3.6.1",
+  "description": "The only AI agent that builds its own tools. Self-evolving terminal AI: 290 tools, 23 agents, 20 providers. Runtime tool forging via forge_tool, Forge Registry for community tools, autopoietic health monitoring, immune self-audit agent. Cost-aware model routing, fallback chains, Bayesian skill routing. Embedded llama.cpp, MCP server, programmatic SDK. MIT.",
   "type": "module",
   "repository": {
     "type": "git",
@@ -44,162 +44,88 @@
   },
   "keywords": [
     "ai-agent",
+    "terminal-agent",
+    "cli-agent",
+    "self-evolving",
+    "autopoiesis",
+    "forge",
+    "tool-creation",
+    "runtime-tools",
+    "mcp",
+    "model-context-protocol",
+    "llm",
+    "anthropic",
+    "openai",
+    "ollama",
+    "coding-agent",
+    "autonomous-agent",
+    "self-improving",
+    "cost-aware",
+    "fallback",
+    "multi-provider",
     "cli",
     "terminal",
     "code-generation",
     "ai-coding-assistant",
     "local-first",
-    "llm",
     "automation",
     "developer-tools",
-    "ollama",
     "claude",
-    "openai",
     "gpt",
-    "anthropic",
     "google-ai",
     "groq",
     "mistral",
     "code-review",
-    "self-evolving",
     "multi-model",
-    "mcp",
     "mcp-server",
     "shell-assistant",
     "open-source-ai",
-    "chatbot",
-    "copilot",
     "ai-assistant",
     "typescript",
-    "python",
-    "rust",
-    "docker",
-    "terraform",
-    "devops",
-    "data-science",
-    "machine-learning",
-    "huggingface",
-    "arxiv",
-    "research",
-    "bioinformatics",
-    "scientific-computing",
-    "api-testing",
-    "graphql",
-    "rest-api",
-    "container",
-    "jupyter",
-    "testing",
-    "linting",
-    "git",
-    "github",
+    "tool-calling",
+    "function-calling",
+    "agentic",
     "agent",
     "swarm",
     "multi-agent",
-    "frontend",
-    "backend",
-    "documentation",
-    "css",
-    "coverage",
-    "mobile",
-    "framework",
-    "math",
-    "iot",
-    "robotics",
-    "vfx",
-    "shader",
-    "glsl",
-    "houdini",
-    "ffmpeg",
-    "imagemagick",
-    "blender",
-    "procedural",
-    "creative-coding",
-    "audio-visualization",
-    "csv",
-    "spreadsheet",
-    "document",
-    "data-transform",
-    "report-generation",
-    "invoice",
-    "audit",
-    "code-audit",
-    "open-source",
-    "contribute",
-    "good-first-issue",
-    "a2a",
-    "agent2agent",
-    "agent-protocol",
-    "browser-automation",
-    "playwright",
-    "web-agent",
-    "composio",
-    "workflow-engine",
-    "guardrails",
-    "agent-handoff",
-    "agent-marketplace",
     "gguf",
     "llama-cpp",
     "local-llm",
-    "tool-calling",
-    "function-calling",
-    "agentic",
-    "autonomous-agent",
     "langchain-alternative",
     "crewai-alternative",
+    "copilot",
+    "docker",
+    "terraform",
+    "devops",
+    "git",
+    "github",
     "deploy",
     "vercel",
     "netlify",
     "cloudflare-workers",
-    "fly-io",
-    "railway",
     "database",
     "prisma",
     "sql",
     "postgres",
-    "mysql",
-    "sqlite",
-    "migration",
-    "pair-programming",
-    "team-coding",
-    "plugin-sdk",
-    "session-recording",
-    "terminal-recording",
-    "svg-animation",
-    "asciinema",
-    "er-diagram",
+    "browser-automation",
+    "playwright",
+    "web-agent",
+    "a2a",
+    "agent2agent",
+    "agent-protocol",
+    "workflow-engine",
+    "agent-marketplace",
     "mcp-marketplace",
-    "fine-tuning",
-    "lora",
-    "qlora",
-    "model-training",
-    "mlx",
-    "unsloth",
-    "gguf-convert",
-    "llm-training",
-    "rlhf",
-    "dataset-preparation",
+    "plugin-sdk",
     "game-development",
     "game-engine",
     "godot",
     "unity",
     "unreal-engine",
-    "bevy",
-    "phaser",
-    "threejs",
-    "ecs",
-    "entity-component-system",
-    "physics-engine",
-    "rapier",
-    "procedural-generation",
-    "level-generation",
-    "particle-system",
-    "sprite-atlas",
-    "netcode",
-    "multiplayer",
-    "gamedev",
-    "shader-analysis",
-    "navmesh",
-    "tilemap"
+    "vfx",
+    "shader",
+    "open-source",
+    "pair-programming"
   ],
   "author": "kernel.chat group",
   "license": "MIT",