v0.4.3 🐝

Release Date: February 10, 2026
Tag: v0.4.3

🎯 Smarter Agent Building & Better Observability

This release focuses on improving the agent building experience with honest use-case qualification, adding sample agent templates, model selection in quickstart, and introducing structured logging for production observability.

✨ What's New

🧭 Use Case Qualification (Goal Negotiation)

Build the right agent for the right problem. The /hive-create skill now includes a mandatory qualification phase (STEP 2) that helps you understand if Hive is the right fit for your use case before you start building.

• Deep Discovery — Structured questions covering core purpose, interaction model, and external dependencies
• Capability Assessment — "The Good, The Bad, The Ugly" framework fit assessment using actual discovered tools
• Gap Analysis — Requirement-by-requirement table showing supported / partial / not supported status
• Clear Recommendations — One of three verdicts: PROCEED, PROCEED WITH SCOPE ADJUSTMENT, or RECONSIDER
• Explicit Acknowledgment — Users must actively choose to proceed before building begins

🏗️ Build from Sample Agent Templates

Start faster with pre-built agent templates. When creating a new agent, you can now choose from sample templates that demonstrate common patterns and best practices.

• Updated Deep Research Agent as reference implementation
• Template-based agent scaffolding for faster development

🎛️ Model Selection in Quickstart

Choose your model during setup. The quickstart now allows you to select the LLM model you want to use, along with appropriate max_tokens configuration for your chosen model.

📊 Structured Logging for Observability

Production-ready logging. Introducing OTel-compliant structured logging that makes transactions traceable across your agent execution.

• Human-readable format — Clean, focused logs for developer debugging
• Machine-parseable JSON format — Detailed structured logs for observability tools (Datadog, Splunk, etc.)
• Trace context propagation for end-to-end request tracing

🐛 Bug Fixes

• OAuth Token Refresh — Fixed OAuth tokens not refreshing after expiration (#3877)
• Credential Resolution — Fixed integration_id and metadata fields being ignored in AdenCredentialResponse.from_dict (#3965)

🙏 Contributors

Special thanks to our community contributors:

• @anjaliy11 (Anjali Yadav) — Credential integration_id resolution fix (#3965)
• @fermano (Fernando Mano) — Structured logging for observability (#1834)

---

What's Next?

We're working on:

• Automatic testing workflow for agent validation
• Interactive debugging UI for checkpoint inspection
• Hot reloading support for code changes
• Graph version compatibility checking

Stay tuned for v0.4.4! 🚀

v0.4.2 🐝

Release Date: February 8, 2026
Tag: v0.4.2

🎉 Resumable Sessions - Never Lose Progress Again

This release introduces comprehensive resumable session functionality, allowing agents to pause and resume execution without losing progress. Whether interrupted by errors, rate limits, or simply needing to step away, your work is now automatically preserved.

✨ What's New

🎯 Core Feature: Resumable Sessions

Never lose progress again. Agents now automatically save state and can resume execution after any interruption - credential errors, rate limits, or simply closing the terminal.

🚀 New Commands

TUI Commands:

• /resume - Resume the most recent paused/failed session automatically
• /resume <session_id> - Resume a specific session by ID
• /sessions - List all sessions for the current agent with status
• /pause - Immediately pause execution with full state capture (alternative to Ctrl+Z)

CLI Flags:

• --resume-session - Auto-resume latest session when starting agent
• --checkpoint <checkpoint_id> - Resume from specific checkpoint (advanced)

⌨️ Keyboard Shortcuts

• Ctrl+Z - Instantly pause execution (now actually works!)
  • Previously: Just showed a notification without pausing
  • Now: Immediately cancels task and saves full state

💾 Automatic State Saving

State is automatically saved when:

• You press Ctrl+Z or use /pause
• You quit the TUI (close terminal or Ctrl+C)
• A node fails with an error
• Agent hits rate limits or missing credentials
• Execution completes (for iterative workflows)

What gets saved:

• Complete memory snapshot with all collected data
• Full conversation history for interactive nodes
• Execution path (which nodes were executed)
• Current node position (exact point where paused)
• Node visit counts (for feedback loop handling)

📂 Unified Session Storage

• All sessions stored at: ~/.hive/agents/{agent_name}/sessions/
• Each session has its own directory with state.json
• Session status tracked: active, paused, completed, failed, cancelled
• Automatic selection of most recent resumable session

🔧 Changes & Improvements

Behavior Changes

• Pause is now immediate - Uses task cancellation instead of waiting for node boundaries
• TUI quit behavior - Now pauses and saves state (previously cancelled without saving)
• Timeout increased - Session cleanup timeout raised from 0.5s to 5s for reliable state saving

Architecture Updates

• Deprecated pause nodes in favor of client-facing EventLoopNodes
• Replaced request_pause() with immediate task cancellation pattern
• Enhanced ExecutionResult with session_state["memory"] at all exit points

🐛 Bug Fixes

• Memory persistence - ExecutionResult.session_state["memory"] now properly populated at all exit points (Ctrl+Z, failure, exception)
• Resume node selection - Fixed resume starting at intake node instead of paused_at node
• Visit count handling - Fixed double-counting on resume (paused node count now properly adjusted)
• Session selection - Fixed selection of oldest instead of latest session
• Quit state save - Fixed failures due to insufficient timeout
• Ctrl+Z implementation - Implemented actual pause logic (was only showing notification)
• Empty memory on resume - Fixed memory restoration by ensuring session_state["memory"] is populated

⚠️ Known Limitations

• Interactive debugging - Checkpoint recovery (/recover) exists but lacks user-friendly inspection UI
• Hot reloading - Code changes don't reflect in resumed sessions (must complete old execution first)
• Graph structure changes - Resuming after heavy modifications (renamed/deleted nodes, changed edges) has undefined behavior
• Version compatibility - No mechanism to detect mismatches between session schema and agent structure

📚 Documentation

See RESUMABLE_SESSIONS_PR.md for detailed operational guide.

🙏 Contributors

This release was made possible by community contributions from:

Contributions (v0.4.0 → v0.4.2)

• @Acid-OP (Gaurav Kapur) - Fixed node outputs not being written to memory before edge evaluation (#3599, #3694)
• @hamzanajam7 - Documentation improvements (Contributing link, project structure fixes)
• @kramitnke5 (Amit Kumar) - Apollo.io contact and company data enrichment integration (#3167)
• @subhayanmukherjee78 (Subhayan Mukherjee) - Clarified test generation responsibility in hive skill documentation
• @RichardTang-Aden (Richard Tang) - Documentation improvements (README updates, video addition, quickstart.sh messages, uv run migration)

Special thanks to all community members who provided feedback, testing, and bug reports during development.

🔗 Links

• Full Changelog: v0.4.0...v0.4.2
• Pull Request: #4146
• Feature Branch: #4141

---

What's Next?

We're working on:

• Interactive debugging UI for checkpoint inspection
• Hot reloading support for code changes
• Graph version compatibility checking
• MCP tools for session management

Stay tuned for v0.5.0! 🚀

v0.4.1 🐝

🐝 This release brings significant improvements to the developer experience with a new Terminal UI, enhanced debugging capabilities, and comprehensive logging infrastructure.

✨ New Features

Hive TUI (Terminal User Interface)

• Interactive terminal UI with menu support, syntax highlighting, and copy functionality
• Consolidated TUI runner log paths for better organization
• Updated quickstart to include hive tui

Hive Debugger Skill

• New interactive debugging companion for Hive agents
• Identifies runtime issues and proposes solutions
• Instructions updated for hive tui integration

Multi-level Logging

• Comprehensive runtime logging infrastructure
• Runtime log tools for viewing and managing agent execution logs
• Unified storage layer for agent runtime
• Improved execution log details in error traces
• Runtime logging structure documentation

Integration & Tools

• Gmail send tool for sending emails via Gmail API
• Slack credential spec and contribution documentation
• Integration credentials implementation check
• Updated hive-credentials workflow

Developer Experience

• Improved quickstart.sh messages
• Key concepts documentation section
• Video demos added to README

🐛 Bug Fixes

• Fixed quickstart secret key setup
• Fixed MCP server configuration with uv
• Hardened Hive CLI setup process
• Removed deprecated --mock mode from codebase and documentation
• Fixed cursor skill symlinks to claude skill
• Fixed skill ordering and naming issues
• Fixed executor configuration and data tool usage
• Fixed execution log details in error trace
• Fixed gap between lint and format
• Fixed deprecated and outdated tests
• Fixed aden hive url
• Bug fix for crashing agent

📚 Documentation

• Added video demos and GIFs to README
• Improved README with better structure and documentation
• Added key concepts section
• Runtime logging structure documentation
• Updated documentation structure and skill references
• Hive commands rebrand
• Added upstream sync steps to CONTRIBUTING guide
• Fixed typo in aden-credential-sync docs
• Updated links in README
• Removed unused docs

🔧 Refactoring

• Consolidated TUI runner log path
• Unified storage layer for agent runtime
• Moved templates to new location
• Changed gmail to google naming convention

🙏 Contributors

A huge thank you to all the amazing contributors who made this release possible!

Community Contributors

@Ansh0928 (Anshumaan Saraf)

• Added upstream sync steps to CONTRIBUTING guide - helping new contributors get started faster

Patrick

• Fixed typo in aden-credential-sync documentation - attention to detail matters!

---

Thank you to everyone who contributed code, reported issues, and provided feedback. Your contributions make Hive better for everyone! 🐝

🚀 Release v0.4.0

79 commits since v0.3.2 | Target: main @ 80a41b4

---

✨ Highlights

This is a major release introducing the Event Loop Node architecture, an interactive TUI dashboard, ClientIO gateway for client-facing agents, a GitHub tool, Slack tool integration (45+ tools), and a full migration from pip to uv for package management.

---

🆕 Features

🔄 Event Loop Node Architecture

• Implement event loop node framework (WP1-4, WP8, WP9, WP10, WP12) — a new node type that supports iterative, multi-turn execution with tool calls, judge-based acceptance, and client-facing interaction
• Emit bus events for runtime observability
• Add graph validation for client-facing nodes
• Soft-fail on schema mismatch during context handoff (no more hard failures)

🖥️ Interactive TUI Dashboard

• Add interactive TUI dashboard for agent execution with 3-pane layout (logs/graph + chat)

• Implement selectable logging, interactive ChatREPL, and thread-safe event handling

• Screenshot feature, header polish, keybinding updates

• Lazy widget loading, Horizontal/Vertical layout fixes

• Integrate agent builder with TUI

• 

💬 ClientIO Gateway

• Implement ClientIO gateway for client-facing node I/O routing
• Client-facing nodes can now request and receive user input at runtime

🐙 GitHub Tool

• Add GitHub tool for repository and issue management
• Security and integration fixes from PR feedback

💼 Slack Tool Integration

• Add Slack bot integration with 45+ tools for multipurpose integration
• Includes CRM support capabilities

🔑 Credential Store

• Provider-based credential store (aden provider credential store by provider)
• Support non-OAuth key setup in credential workflows
• Quickstart credential store integration

📦 Migration to uv

• Migrate from pip to uv for package management
• Consolidate workspace to uv monorepo
• Migrate all CI jobs from pip to uv
• Check for litellm import in both CORE_PYTHON and TOOLS_PYTHON environments

🛠️ Other Features

• Tool truncation for handling large tool outputs
• Inject runtime datetime into LLM system prompts
• Add sample agent folder structure and examples
• Add message when LLM key is not available
• Edit bot prompt to decide on technical size of issues
• Update skills and agent builder tools; bump pinned ruff version

---

🐛 Bug Fixes

• ON_FAILURE edge routing: Follow ON_FAILURE edges when a node fails after max retries
• Malformed JSON tool arguments: Handle malformed JSON tool arguments safely in LiteLLMProvider
• Quickstart compatibility: Fix quickstart.sh compatibility and provider selection issues
• Silent exit fix: Resolve silent exit when selecting non-Anthropic LLM provider
• Robust compaction logic: Fix conversation compaction edge cases
• Loop prevention: Prevent infinite loops in feedback edges
• Tool pruning logic: Fix incorrect tool pruning behavior
• Text delta granularity: Fix text delta granularity and tool limit problems
• Tool call results: Fix formulation of tool call results
• Max retry reset: Reset max retry counter to 0 for event loop nodes
• Graph validation: Fix graph validation logic
• MCP exports directory: Handle missing exports directory in test generation tools
• Bash version support: Fix bash version compatibility

---

🏗️ Chores & CI

• Consolidate workspace to uv monorepo
• Migrate remaining CI jobs from pip to uv
• Clean up use of setup-python in CI
• Windows lint fixes
• Various lint and formatting fixes
• Update .gitignore and remove local claude settings
• Update issue templates

---

📖 Documentation

• Add Windows compatibility warning
• Update architecture diagram source path in README

---

👏 Contributors

Thanks to all contributors for this release:

• @mubarakar95 — Interactive TUI dashboard (3-pane layout, ChatREPL, selectable logging, screenshot feature, lazy widget loading)
• @levxn — Slack bot integration with 45+ tools including CRM support
• @lakshitaa-chellaramani — GitHub tool for repository and issue management
• @Acid-OP — ON_FAILURE edge routing fix after max retries
• @Siddharth2624 — Malformed JSON tool argument handling in LiteLLMProvider
• @Antiarin — Runtime datetime injection into LLM system prompts
• @kuldeepgaur02 — Fix silent exit when selecting non-Anthropic LLM provider
• @Anjali Yadav — Fix missing exports directory in MCP test generation tools
• @Hundao — Migrate remaining CI jobs from pip to uv
• @ranjithkumar9343 — Windows compatibility warning documentation
• @yogesh Sakharam Diwate — Architecture diagram path update in README

Release 0.3.2

Before the upcoming major event loop node release (v0.4.0), we wanted to ship a release focused on developer experience — making it easier to get started, stay set up, and build with confidence. This release packs 203 commits from 29 contributors with new integrations, a revamped quickstart, hardened reliability, and expanded platform support.

---

What's New

Email Service Tool — @Hundao

A full-featured email integration powered by Resend, with support for CC/BCC recipients. Agents can now send emails as part of their workflows without external glue code.

Credential Store Auto Sync

The credential store now syncs automatically with the Aden provider, replacing the old credential manager. Credentials stay up to date without manual intervention.

Quickstart Overhaul with uv — @shivamhwp

The quickstart.sh has been largely rewritten (700+ lines changed) to use uv as the package manager. It now auto-installs uv, picks a compatible Python (>=3.11), and creates separate virtual environments for core and tools. The result is a significantly faster and more reliable onboarding experience.

Native PowerShell Setup for Windows

A first-class PowerShell setup script for Windows users — no more WSL requirement for getting started on Windows.

Execution Quality Tracking

The executor now exposes retry metrics, giving visibility into node execution quality and retry behavior across agent runs.

HubSpot Integration & Advanced Scraper

New HubSpot tool integration and an improved web scraper for richer data extraction capabilities.

Root Hive CLI Entry Point

A new root hive CLI command that eliminates the need to set PYTHONPATH manually — cleaner developer experience out of the box.

---

Fixes & Improvements

• Security: Removed debug print statements that leaked API key to stdout
• ConcurrentStorage: Fixed race condition and cache invalidation bug — @krish341360
• Agent Builder: Made exports atomic to prevent partial writes
• CSV Tool: Handle csv_write with no parent directory; validate non-negative limit/offset in csv_read
• PDF Tool: Surface truncation warning when exceeding max_pages
• Graph Validation: Added validation for client-facing nodes
• LLM JSON Extraction: Validate extraction output to prevent empty/fabricated data
• Tool Registry: Handle invalid JSON returned by tools
• MCP: Load MCP servers correctly; fix session disconnect cleanup
• Setup Script: Creates missing exports/ directory; fix Python/pip version detection
• CLI: Fix KeyError 'steps' in cmd_list function
• Async Runtime: Add is_running() and is_closed() checks to _run_async() to prevent deadlock
• Windows: Fix Unicode encoding issues
• CI: Add Windows runner to GitHub Actions; add tool tests in CI; ruff formatting fixes
• Dependencies: Add missing litellm to tools dependencies

---

Documentation

• Clarified who Hive is for and when to use it
• Updated roadmap and architecture mermaid chart
• Added Hindi (हिंदी) README translation
• Tools/integration contribution guide
• Rewrote configuration docs to reflect actual Python framework architecture
• Improved Q&A and feature descriptions
• QuickStart improvements: MCP config, Python version, venv docs
• Added .python-version for automatic Python version detection
• Removed duplicate entries and fixed doc links

---

Special Thanks

This release would not have been possible without the contributions of our growing community. A heartfelt thank you to every contributor who submitted code, reported issues, reviewed PRs, and helped shape Hive into what it is today.

• Hundao (@Hundao) — email service tool with Resend provider, CI fixes
• Muzzaiyyan Hussain (@MuzzaiyyanHussain) — atomic agent builder exports, Hindi README translation
• Nicklaus FW (@NicklausFW) — CSV write fix, execution quality tracking
• KrishShharma (@krish341360) — ConcurrentStorage race condition fix
• Pravin Mishra (@mishrapravin114) — PDF read max pages truncation
• Sourab Singh Bora (@Sourabsb) — API key leak fix, async runtime deadlock fix, MCP disconnect cleanup
• Shivam (@shivamhwp) — QuickStart improvements
• Chandradithya Janaswami (@JVSCHANDRADITHYA) — CLI crash fix for missing exports directory
• Harsh Kishorani (@HarshKishorani) — native PowerShell setup for Windows
• Dhruv Pokhriyal (@DhruvPokhriyal) — CSV read validation
• Anshu Bhatt (@Anshu-bhatt) — .python-version auto-detection
• Rocky Sahu (@Rockysahu704) — documentation improvements
• Anshumaan Saraf (@Ansh0928) — Edge Protocol docstring cleanup
• Tiago Sasaki (@tjsasakifln) — root Hive CLI entry point, configuration docs rewrite
• Aman (@Aman030304) — logging refactor, Windows CI runner
• Devanshu Magiawala (@magiawala) — MCP example import fix
• Suhani Jindal (@suhanijindal) — Python 3.13 classifier
• trinh31201 (@trinh31201) — LLM JSON extraction validation
• Shivraj (@Shivraj12) — tool registry invalid JSON handling
• Mrunal Shah (@mrunalnshah) — setup script Python/pip version detection fix
• Siddharth Varshney (@Siddharth2624) — output sanitization docs
• Harsh Makwana (@hrshmakwana) — setup script exports directory fix
• Ayush Pandey (@ap-dev-github) — Windows Unicode encoding fix
• Rahul Vakiti (@vakrahul) — style fixes
• dithzz (@dithzz) — CLI KeyError fix
• RussellLuo (@RussellLuo) — MCP server loading fix

Thank you all for being part of the Hive journey. Onward!

What's next

We're planning for a major release 0.4.0 with the event loop node and other features in the coming days. It is going to increase the worker agent capability a lot. Learn more in #2500

There is also a TUI for worker agent coming soon for testing and using the worker agent(credit: @mubarakar95). Share us feedback if you are interested in it here #1644!

v0.3.1: Cursor IDE Support

Summary

Add Cursor IDE support for existing Claude Code skills and MCP servers.

Changes

• Created .cursor/skills/ directory with symlinks to all 5 existing skills:
  • agent-workflow
  • building-agents-core
  • building-agents-construction
  • building-agents-patterns
  • testing-agent
• Added .cursor/mcp.json with MCP server configuration (same as .mcp.json)

Why symlinks for skills?

• Single source of truth - updates to .claude/skills/ are reflected in both IDEs
• No duplication or sync issues
• Cursor automatically loads skills from .cursor/skills/, .claude/skills/, and .codex/skills/

MCP Configuration

Cursor requires .cursor/mcp.json for project-level MCP servers. This enables:

• agent-builder - Agent building MCP server
• tools - Hive tools MCP server

Setup in Cursor

1. Enable MCP: Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P) and run MCP: Enable
2. Restart Cursor to load the MCP servers from .cursor/mcp.json
3. Skills: Type / in Agent chat and search for the skill name

v0.3.0

Release v0.3.0

A major milestone for the Hive framework — 218 commits from 40+ contributors spanning new execution models, provider flexibility, developer tooling, and hardened reliability.

---

What's New

Parallel Execution Framework

A first-class fanout/fanin model for running nodes concurrently. Define parallel branches in your graph and let the framework handle orchestration, synchronization, and result aggregation.

Pydantic Model Validation for LLM Outputs

LLM responses can now be validated against Pydantic models, giving you structured, typed outputs with automatic error feedback and retry.

Provider-Agnostic LLMJudge

The LLMJudge has been decoupled from Anthropic and now supports OpenAI and any LiteLLM-compatible provider. Configurable at the judge level with api_key and api_base support.

CSV Tools with DuckDB SQL

A new CSV toolkit powered by DuckDB lets agents query, filter, and aggregate CSV data using SQL — no Pandas required.

Framework Guardrails & Node I/O Cleaner

Built-in guardrails for agent execution and an I/O cleaner that sanitizes node inputs/outputs to prevent hallucination propagation.

Multi-Provider Web Search

The web_search tool now supports multiple search providers, making it easier to swap backends without changing agent logic.

Mock Mode LLM Provider

A MockLLMProvider for deterministic testing of agent graphs without making real API calls.

---

Fixes & Improvements

• ConcurrentStorage: Fixed race condition in stop() causing data loss; patched file lock leak with WeakValueDictionary
• Graph retry: Exponential backoff for node retries; restored node.max_retries configuration
• Execution stream: Fixed memory leak in runtime execution streams
• Web scraping: Added robots.txt compliance and Content-Type validation
• Type safety: Corrected callable → Callable annotations across the codebase
• Hallucination detection: Improved code-indicator checking across full strings
• Worker node: Added JSON parsing failure logging
• Safe evaluation: Fixed unsanitized expression evaluation vulnerability
• Trigger & wait: Safe dictionary access in trigger_and_wait()
• Semaphore: Avoided private _value attribute access

---

Documentation

• Architecture README and documentation link fixes
• DeepSeek models support documentation
• Windows/WSL environment guidance
• Updated tool names and descriptions in README
• Skills directory structure documentation

---

Special Thanks

This release would not have been possible without the contributions of our growing community. A heartfelt thank you to every contributor who submitted code, reported issues, reviewed PRs, and helped shape Hive into what it is today.

• yumosx — consistent contributions across multiple areas
• Rahul Vakiti (@vakrahul) — graph retry with exponential backoff
• Sriharsha Kilaru (@HarshaKilaru) — runtime fixes and stream improvements
• Tanuja Nair (@TanujaNair03) — provider-agnostic LLMJudge refactor and prompt templates
• Kira714 — framework guardrails and testing
• Uttam Kumar — node I/O cleaner
• RanxinLi — concurrent framework entrypoints
• Hundao — CSV tools with DuckDB support
• Tahir Yamin — ConcurrentStorage file lock leak fix
• Nihal Morshed — contributions and fixes
• LunaStev — contributions and improvements
• Himanshu Chauhan (@Himanshu-ABES) — Pydantic model validation for LLM outputs
• Fernando Mano — contributions and fixes
• Chrishabh2002 — contributions and improvements
• adionit7 — validate exports fix and docs corrections
• Pradyumn Tendulkar (@pradyten) — configurable LLM provider for LLMJudge
• not-anas-ali — type annotation fixes
• saboor2632 — worker node JSON logging
• savan patel (@savankansagara1) — MockLLMProvider for mock mode
• gaurav (@gaurav-code098) — web scrape Content-Type validation
• Vishal (@vishalharkal15) — ConcurrentStorage race condition fix
• Samkit Shah — contributions
• RussellLuo — contributions
• Arush Wadhawan — contributions
• Aysun Itai — contributions
• AryanyAI — MCP script logging refactor
• AryanRevolutionizingWorld — contributions
• ayush123-bit — contributions
• dhakalrabin — contributions
• guillermop2002 — contributions
• koushith — contributions
• Kotapati Venkata Sai Charan — contributions
• mithileshk — contributions
• Shamanth-8 — contributions
• subhampanja28 — contributions
• Viacheslav Borisov — contributions
• vrijmetse — contributions

Thank you all for being part of the Hive journey. Onward to v0.4.0!

---

Full Changelog: v0.2.0...v0.3.0

Release v0.2.0 - Aden Agent Framework

What's New

Aden Agent Framework is a Python-based system for building goal-driven, self-improving AI agents that can plan, execute, and adapt their behavior based on real-world outcomes.

Core Features

Goal-Driven Architecture

• Define agent goals with success criteria and constraints
• Automatic validation and testing of agent behavior
• Built-in evaluation framework with constraint and success tests

Flexible Execution Models

• Graph-based execution with dynamic routing
• Worker-Judge loop for adaptive planning
• Human-in-the-loop approval system
• Pause/resume support for long-running workflows

Multi-LLM Support

• LiteLLM integration supporting 100+ LLM providers
• OpenAI, Anthropic, Google Gemini, Mistral, Groq, and local models
• Unified interface with provider-specific optimizations

MCP Tools Integration

• 19 built-in tools for file operations, search, and APIs
• Model Context Protocol (MCP) server support
• Custom tool registration and discovery

Testing & Evaluation

• Goal-based evaluation tests
• Parallel test execution
• Automatic test generation from goals
• Debug tools with categorized failure analysis

Infrastructure Improvements

• Migrated CI/CD from Node.js to Python workflows
• Added ruff linting to maintain code quality
• All 107 tests passing with zero linting errors
• Updated documentation and templates for Python development

Agent Building

Build agents using:

• Interactive CLI (`aden run`)
• Python API with GraphExecutor
• MCP server tools for Claude Desktop integration
• Agent-to-agent orchestration

For documentation, examples, and getting started guides, visit:
https://github.com/adenhq/hive

Using Release Artifact Directly

pip install https://github.com/TimothyZhang7/hive/releases/download/v0.2.0/framework-0.2.0-py3-none-any.whl

v0.1.0 - Initial Release

Beeline v0.1.0 - Initial Open-Source Release

We're excited to announce the first public release of Beeline, an AI agent observability and control platform.

Features

• React Frontend (Honeycomb) - Modern TypeScript + Vite frontend
• Node.js Backend (Hive) - Express + TypeScript API server
• MCP Server - Model Context Protocol support for AI agent integration
• Docker Support - Docker Compose configuration for easy deployment
• Centralized Configuration - Single config.yaml for all settings

Getting Started

git clone https://github.com/adenhq/beeline.git
cd beeline
cp config.yaml.example config.yaml
npm run setup
docker compose up
Documentation

Links
Documentation
Discord Community
Roadmap