v0.6.0 - Closed-Loop Engine, Self-Correction & Whatsapp

Jared: The AI COO — v0.6.0 Release

This release introduces a major evolution of the agent engine, establishing self-correcting closed loops, intelligent memory compaction, and auto-improvement capabilities. We've also refactored the WhatsApp channel with interactive safety prompts, added advanced developer skills (TDD and chub API docs integration), and significantly increased test coverage to ensure overall stability.

🔄 Closed-Loop Engine & Self-Correction

• 🩺 Critic Self-Review Cycle: Implemented an automated Quality Assurance Critic loop. Proposed answers are evaluated against a criteria scoring metric (1-10); answers scoring under 6/10 are rejected, forcing the agent to ingest criticism and self-correct on subsequent attempts.
• 📊 Execution Tracing & Diagnostics: All tool executions (arguments, results, success scores, and errors) are now saved directly in the SQLite database (memory.db), enabling complete auditability.
• 🛡️ Interactive Error Recovery: When tools fail, the engine injects diagnostic details and warnings into the agent context, prompting the model to analyze failures in a block, adjust parameters, or pivot to alternative paths.

🧠 Memory Compaction & Context Management

• 📉 Automatic Context Compaction: Prevents context window overflows and reduces token consumption by summarizing session history when a token threshold is exceeded, moving key facts to core memory and truncating history.
• 🧹 Manual Compaction Command: Added the /compact command to trigger history compaction and clean up conversation state on demand.
• 🗺️ Bypass Flag (--nocontext): Added a --nocontext option to prompt execution to completely ignore previous conversation history for isolated, clean-slate tasks.

🧬 Auto-Improvement System

• 📈 Self-Correcting Skills: Added a new auto_improve tool and a jared auto-improve CLI action. Jared analyzes recent execution failures from the trace database and automatically rewrites the corresponding SKILL.md instruction manual to prevent future errors.
• ⚙️ On-Demand Auto-Improvement: Added the /auto-improve slash command to let users manually kick off the self-improvement loop for a specific skill.

🔌 Fully-Featured WhatsApp Channel

• 📱 Robust WhatsApp Client: Enabled full integration via whatsapp-web.js with structured state management, auto-auth caching, and clean client teardown handling on shutdown (SIGINT/SIGTERM/system:shutdown).
• 🔐 Interactive ExecGuard on Chat: Command approval prompts from the execution guard are sent directly to the WhatsApp chat, allowing you to approve or deny executions on the fly.
• 🏎️ Dynamic Typing & Progress Indicators: Emits real-time typing indicators, tool execution stages, subagent status, and throttled token streaming to provide a responsive, live chat experience.

📚 Extended Developer Skills & Tests

• 🧪 TDD (Test-Driven Development) Skill: Added a core skill to guide Jared through Red-Green-Refactor development, vertical tracer bullets, and interface-driven mocking.
• 🌐 chub API Docs Integration: Added the get-api-docs skill using the chub CLI tool (@aisuite/chub), allowing the agent to dynamically search, pull, and annotate third-party library reference docs.
• 🧪 Extensive Test Coverage: Built 7 new test suites covering WhatsApp, the console lexer, closed-loop engine mechanics, execution guard, spawning subagents, and skills loading.

✨ Upgrading

git pull origin main
jared start <project-name>

v0.5.2 - Security & Performance Hardening

Jared: The AI COO — v0.5.2 Release

This release focuses on hardening the core engine, significantly boosting performance through concurrency, and streamlining the codebase. We’ve addressed critical security vulnerabilities and optimized the agent loop to make Jared faster and more secure.

🛡️ Security & Stability

• 🔒 Command Injection Fix: Patched a critical vulnerability in the exec tool to prevent command injection, ensuring shell executions are strictly bound to intended parameters.
• 🎲 Secure Session IDs: Upgraded the session ID generation to use cryptographically secure randomness, preventing session predictability.
• 🩺 Monitoring Skill Revisions: Reverted regressions in the monitoring skill and cleaned up temporary agent artifacts to ensure stable environment health checks.

⚡ Performance Overhaul

• 🏎️ Concurrent Tool Execution: The agent loop now executes multiple tool calls in parallel whenever possible, drastically reducing the latency of complex multi-step tasks.
• 🔌 Parallel MCP Initialization: Optimized the startup sequence by parallelizing the initialization of MCP servers, resulting in much faster boot times for environments with multiple external integrations.
• 📊 Optimized Console Rendering: Improved the performance of formatTable and console log output, ensuring the terminal remains responsive even during high-verbosity operations.

📐 Refactoring & Lean Core

• ⛳ Codebase "Golfing": Performed a massive refactoring of the core engine, bringing the total Lines of Code (LOC) well below 2,500. This ensures Jared remains ultra-lightweight and maintainable without sacrificing a single feature.
• 🧹 File I/O Optimization: Optimized Cron and memory file operations to use asynchronous I/O more efficiently, reducing blocking time during background syncs.

✨ Upgrading

git pull origin main
jared start

v0.5.0 - The Multi Project Structure, The Backlog, Model routing & DevOps Update

Jared: The AI COO — v0.5.0 Release

The v0.5 release marks a major architectural shift. Now you can have different projects inside Jared. All organized by folders. We’ve also unified Jared’s scheduling and task management into a single "Source of Truth," introduced on-the-fly model routing, and expanded Jared’s reach into professional DevOps workflows with new skills and subagent roles.

🚀 Key Highlights

• 📚 Jared is now multi project: You can organize your different agents by projects. Check the readme to understand how to onboard a new project and how to launch Jared for the specific project.
  Hint: jared start becomes: jared start project-name
• 📅 Unified Scheduling & Backlog: Farewell HEARTBEAT.md, hello BACKLOG.md! Jared now uses a single file to manage both your Product Backlog and your Scheduled Tasks (Cron). Jared automatically syncs with this file every minute, allowing you to manage his proactive schedule just by editing a Markdown file.
• 🔌 Explicit Model Routing: You can now override Jared's default model on the fly! Append a provider flag (e.g., --openai, --ollama, --mistral) to any message to route that specific request to a different provider without changing your global configuration.
• 🛠️ DevOps "Hiring" Expansion: We've introduced a new DevOps Engineer identity and two powerful new skills:
  Monitoring: Automatically check environment health via curl with proactive LLM-powered failure analysis.
  Alerting: Perform instant log health checks on local or remote files to identify errors or anomalies.
• 🖥️ CLI Polish & Built-ins:
  jared reset-memory: Now a native built-in command (replacing the old bash script).
  jared stats: Added a new command to view agent performance and usage statistics.
• 🧹 Refined Architecture: The codebase has been leaned out by merging the heartbeat logic directly into a more robust Cron engine, improving reliability for long-running background tasks.
• ✨ Upgrading
  Pull the latest main branch. Note that your HEARTBEAT.md tasks should be migrated to the new BACKLOG.md format (see README for the updated template):

git pull origin main
jared onboard <project-name>
jared start <project-name>

v0.4.0 - Advanced MCP & Telemetry Update

Jared: The AI COO — v0.4.0 Release

The v0.4 release brings much-requested flexibility to how Jared handles external tools via the Model Context Protocol (MCP), introduces token telemetry for better cost tracking, and gives you finer control over Jared's reasoning engine. Oh, and Jared finally has a screenshot of how it looks like !

🚀 Key Highlights

• 🔌 Enhanced MCP Integrations: Major upgrades to the MCP engine! We now support Streamable HTTP transport for modern remote MCP endpoints.

• 🎛️ Toggleable MCP Servers: You can now temporarily disable specific MCP servers by simply adding "enabled": false to your config.json. This prevents their tools from injecting into Jared's context without forcing you to delete your configuration entirely.

• 📊 Token Usage Stats: The console channel just got smarter. We've enhanced the terminal UI formatting and added real-time token usage stats, so you can easily monitor your prompt and completion payload footprint during complex task loops. Use "jared stats" to check the amount of tokens used with every model used

• 🧠 "Thinking" Flag: Added a new "thinking" boolean to config.json. You can now explicitly toggle reasoning blocks on or off depending on the LLM model you are using, giving you tighter control over response latency and formatting.

• ✨ Upgrading
  Pull the latest main branch, review your config.json if you want to try the new thinking toggle, and restart the agent:

git pull origin main
jared start

v0.3.1-alpha - Heartbeat Parsing & Workspace Sandboxing

Jared: The AI COO — v0.3.1 Release

A quick minor release bringing essential QoL improvements to how Jared handles time-based tasks, alongside stricter prompt boundaries for command execution.

🚀 Key Highlights

• ⏱️ Relative Time Parsing in Heartbeat: You can now ask Jared to schedule tasks functionally using relative terminology. If you tell him "remind me in 15 minutes" or "setup the repo in 2 hours", the heartbeat tool will automatically calculate and convert that into absolute clock time (e.g. 11:45 AM) behind the scenes before updating HEARTBEAT.md.

• 🔒 Workspace Sandboxing (Exec Guard): We added an extra layer of explicit prompting to context.js and the exec tool. Jared is now explicitly instructed that, when sandboxed to a workspace, he operates strictly from the workspace root and is prevented from traversing up the directory tree or using absolute external paths.

• 📝 Documentation Polish: Improved formatting in README.md for tool descriptions (documenting the new read_skill_manual and heartbeat tools) and better markdown spacing for HEARTBEAT.md examples.

• ✨ Upgrading

git pull origin main
jared start

v0.3.0-alpha - The Terminal UX & Context Engine Update

Jared: The AI COO — v0.3.0 Release

With the v0.3.0 release, interacting with Jared feels more fluid, responsive, and robust. We have completely overhauled the console experience, significantly upgraded Jared's internal context engine, and rolled out a brand new, highly consistent Heartbeat system to keep Jared's background operations ticking flawlessly.

🚀 Key Highlights

• ✨ Thinking Mode & Terminal Polish: The console channel has received a major glow-up. It now features significantly improved terminal readability and a new "Thinking Mode" with an animated spinner, giving you real-time visual feedback while Jared processes complex workflows.

• 🩺 Remodeled Heartbeat System: The background operations architecture has been rewritten from the ground up (heartbeat.js). The new heartbeat system is far more consistent and reliable, ensuring Jared proactively monitors tasks without skipping a beat.

• 👥 Expanding the AI Team: We're padding out the src/team/ roster. You can now spawn out-of-the-box Product Owner (product_owner.md) and Scrum Master (scrum_master.md) subagents alongside your web developers to manage your agile workflows.

• 🧠 Major Context Engineering: Under-the-hood logic in context.js and SOUL.md has been tightened. Out of the box, Jared's persona is sharper, and the agent requires less hand-holding.

• 📖 Read Skill Manual Tool: Introduced an internal read_skill_manual tool, allowing Jared to dynamically digest and understand new skills dropped into the directory, making extensibility virtually frictionless.

• 📏 Bumping the Cap: To handle this structural growth, our ultra-lightweight guarantee has been expanded—we now allow up to an efficient 3,000 Lines of Code (LOC) across the core engine.

• ✨ Upgrading
  Pull the latest main branch and restart the agent to experience the new terminal UI:

git pull origin main
jared start

v0.2.0-alpha - Ultra-Lightweight AI COO

Jared is expanding beyond a solitary AI COO into a fully-fledged organizational system. The v0.2.0 release introduces role-based AI teaming, a unified Identity architecture, and critical security/CLI stability fixes to keep your digital ops running smoother than ever.

🚀 Key Highlights

• 👥 AI "Hiring" & Team Folder: You can now build out a full roster of specialized AI employees. Drop role definitions (e.g., web_developer.md) into the new src/team/ directory, and Jared can instantly spawn subagents utilizing these distinct roles.
• 🎭 Unified Identity Architecture: We've refactored the old templates system into a clean, centralized identity module. No more bloated templates—just clear, declarative agent identities that the subagent spawn tool can inherit directly.
• 🧠 Enhanced System Soul: Significant improvements have been made to SOUL.md, giving Jared a more refined, contextually aware persona out of the box with improved directive handling.
• 🛡️ Exec Guard Bug Fixes: Fixed a frustrating visual bug in exec-guard.js. User confirmation prompts (allowlist/blocklist) will no longer echo your input twice, resolving conflicting readline interfaces for a much cleaner console experience.
• 🧹 Codebase Optimization: Dropped unused dependencies from package.json and significantly optimized the bun.lock, leaving the project leaner and faster.
• ✨ Upgrading

Pull the latest main branch and you're good to go !

git pull origin main
jared start

v0.1.0-alpha - Ultra-Lightweight AI COO

Jared: The AI COO — v0.1.0-alpha Initial Release

Jared is an opinionated AGAAS (Agentic as a Service) framework designed to be your persistent, always-on AI Chief Organizational Officer. Unlike simple chatbots, Jared is built to handle the "ops cost" of your digital life across multiple channels with a focus on simplicity, security, and extreme efficiency.

🚀 Key Highlights

• 📏 Ultra-Lightweight Core: Delivers a full-featured agent loop, universal provider routing, and persistent memory in under 2,000 lines of code.
• 📱 Universal Multi-Channel Routing: Connect simultaneously via Console, Telegram, Discord, Slack, WhatsApp
• 🧠 Local-First Memory: Native SQLite-backed memory for short-term context and long-term facts/summaries. No cloud vector DB required.
• 🔌 MCP Native: First-class support for the Model Context Protocol to extract tools from any MCP server.
• 🔒 Built-in Security (Exec Guard): A three-layer defense system (allowlist, blocklist, and confirmation mode) for secure shell command execution.
• 🎯 Markdown Skills: Extend Jared's capabilities simply by dropping a SKILL.md file into the src/skills/ directory.
• 💓 Heartbeat & Cron: Proactive task monitoring via HEARTBEAT.md and a robust built-in cron scheduler.
• 🛠 Built-in Tools: Includes native support for exec, search_memory, cron, web_search, web_fetch, and spawn (subagents).

✨ Getting Started

1. Onboard: jared onboard
2. Configure: .jared/config.json
3. Start: jared start