Skip to content

v0.6.0 - Open Hive 🐝

Choose a tag to compare

View all tags
@TimothyZhang7 TimothyZhang7 released this 27 Feb 04:52
· 773 commits to main since this release
v0.6.0
6a8286d
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Release Notes

Release Date: February 26, 2026
Tag: v0.6.0

Open Hive

v0.6.0 is the biggest release in Hive's history. The framework breaks out of the terminal with a full web workspace -- a React SPA backed by a new FastAPI HTTP server where you can run agents, chat, visualize graphs, and manage credentials from the browser. Under the hood, a new queen/worker/judge multi-agent runtime enables agents that spawn, monitor, and coordinate other agents. The credential system gets a ground-up rewrite with local encrypted storage, health checks, and mid-session installation. And OpenAI Codex joins as a first-class LLM provider with OAuth subscription support.

Highlights

Web Workspace

A complete browser-based workspace replaces the TUI as the primary interface for running agents. Built with React, TypeScript, Vite, shadcn/ui, and Tailwind CSS.

  • Home page -- browse agents and start sessions with an initial prompt
Screenshot 2026-02-26 at 8 54 36 PM
  • Workspace -- interactive agent graph visualization (ReactFlow), live chat panel, node detail inspector, and streaming log pane
Screenshot 2026-02-26 at 8 54 12 PM
  • Credentials modal -- add, validate, and revoke credentials mid-session without restarting
  • Real-time streaming -- SSE-based event streaming for live execution updates
  • Markdown rendering -- rich message formatting in the chat panel
  • Tab persistence -- multi-session support with persistent workspace tabs

HTTP API Server

A FastAPI server at core/framework/server/ provides the backend for the web workspace and exposes a full REST + SSE API:

  • Session lifecycle -- create, list, resume, cancel, reconnect sessions
  • Execution control -- start, stop, send input, cancel workers
  • Graph inspection -- load and query agent graph definitions
  • Credential management -- validate, install, and revoke credentials at runtime
  • SSE event streaming -- real-time execution events for frontend consumption
  • Log queries -- structured runtime log access

Full API documentation in core/framework/server/README.md and docs/server-cli-arch.md.

Queen / Worker / Judge Multi-Agent Runtime

A new orchestration layer enables hierarchical multi-agent coordination:

  • Queen lifecycle tools -- spawn, monitor, stop, and cancel worker agents with escalation ticket support
  • Worker health monitoring -- stall detection, iteration tracking, and automatic cleanup of stale sessions
  • Concurrent judge -- event-bus-integrated evaluation that runs alongside agent execution
  • Session Manager -- replaces the old Agent Manager with full concurrent session support, reconnection, and session isolation
  • 3-layer resume prompts -- robust checkpoint recovery across session restarts
  • Trigger node visualization -- schedule info displayed when clicking trigger nodes in the graph

Credential System v3

The credential subsystem is rewritten again with local-first encrypted storage:

  • Local credential registry (credentials/local/) -- encrypted storage with namespace support ({credential_name}/{alias})
  • Health check framework -- live API validation for stored credentials (Brevo, PostgreSQL, and more)
  • Mid-session credential management -- install missing credentials and resync MCP servers without restarting
  • Deferred validation -- credentials validated on use rather than on load, with dismissable error banners
  • Revoke and re-validate -- full lifecycle management in both TUI and web UI

OpenAI Codex LLM Provider

Codex joins as a supported LLM provider with full OAuth integration:

  • OAuth consent flow -- browser pop-out for subscription authentication
  • Streaming support -- through LiteLLM with proper tool call handling
  • Weak model fixes -- skip auto-block when models output text instead of calling tools, remove implementation hints from judge feedback

What's New

Architecture & Runtime

  • Web frontend -- Full React + TypeScript SPA with agent workspace, graph visualization, chat, and credential management. (core/frontend/)
  • HTTP API server -- FastAPI backend with route modules for sessions, execution, graphs, credentials, events, and logs. (core/framework/server/)
  • Session Manager -- Replaces Agent Manager with concurrent session support, reconnection, and isolation. (core/framework/server/session_manager.py)
  • Queen lifecycle tools -- 6 tools for spawning, monitoring, stopping, and canceling worker agents with escalation support. (core/framework/tools/queen_lifecycle_tools.py)
  • Worker monitoring tools -- Health checks, stall detection, iteration tracking, and stale session cleanup. (core/framework/tools/worker_monitoring_tools.py)
  • Concurrent judge runtime -- Event-bus-integrated judge that evaluates agent execution in parallel. (core/framework/monitoring/judge.py)
  • Codex OAuth provider -- OpenAI Codex subscription support with browser-based OAuth consent flow. (core/codex_oauth.py)
  • LiteLLM overhaul -- Improved tool call handling, stop reason logging, weak model support, deprecated unused functions.
  • Refactored system prompt narratives -- Cleaner prompt composition with worker path support.
  • Event bus enhancements -- Structured logging, execution_id tracking, iteration-based message IDs.

TUI Improvements

  • Get-started menu -- guided entry point for new users
  • Ctrl+J newline fallback -- alternative newline input in chat (@Ttian18)
  • Local credential add screen -- TUI modal for installing credentials locally
  • Agent picker refinements -- improved loading and selection flow
  • Quickstart prompt -- restart terminal prompt after quickstart setup

New Tool Integrations

Tool Description Contributor
Wikipedia Search and retrieve Wikipedia article summaries @nafiyad
arXiv Search papers and download PDFs from arXiv @ishaannk
PostgreSQL Read-only SQL queries via MCP @SANTHAN-KUMAR
Brevo Email marketing API integration Core team
Competitive Intelligence Full agent template for market analysis @nafiyad

New Agent Template

  • Competitive Intelligence Agent -- Market analysis agent with web scraping, competitor tracking, and reporting. Ships as both template and ready-to-run export. (@nafiyad)

Bug Fixes

  • Fix cancel worker logic for clean shutdown
  • Fix erroneous skip judge logic that bypassed evaluation
  • Fix stalled agent stop tools for proper cleanup
  • Fix agent loading after graph changes
  • Fix credentials modal blocking incorrectly
  • Fix session reconnect and iteration-based message IDs
  • Fix faster input_requested event delivery
  • Fix fake goal prompt injection vulnerability
  • Fix deferred credential validation with dismissable error banners
  • Fix missing MCP reference in credential setup
  • Fix clean up stale active sessions on worker load
  • Fix credential loading and popup flow
  • Fix session ID issues and duplicative starter API
  • Fix isolate session loading between concurrent sessions
  • Fix Codex tool call usage for streaming responses
  • Fix asyncio.run crash in GraphBuilder (@vincentjiang777)
  • Fix retired claude-3-5-haiku model reference with claude-haiku-4-5 (@kart1ka)
  • Fix zero-config local LLM support and AgentRunner crash (@vakrahul)
  • Fix Shift+Enter newline behavior in chat input (@Ttian18)
  • Fix web scrape robots.txt support and networkidle wait
  • Fix SuccessCriterion type field and evaluation guard (@Rudra2637)
  • Reduce subprocess spawning in quickstart scripts (@mubarakar95)

Documentation

  • Fix README formatting and links (@JamieJiHeonKim)
  • Fix Roadmap Mermaid diagram rendering for GitHub (@sabasiddique1)
  • Security tools README documentation
  • Server CLI architecture documentation (docs/server-cli-arch.md)
  • Worker health monitoring documentation (docs/worker-health-monitoring.md)
  • HTTP API README (core/framework/server/README.md)

Breaking Changes

  • Session Manager replaces Agent Manager -- The old agent_manager.py API surface is superseded by session_manager.py. Scripts using AgentRuntime.start_agent() should migrate to the session-based API.
  • Credential namespacing -- Credentials are now stored under {credential_name}/{alias} paths. Existing credentials in ~/.hive/credentials will need to be re-stored.
  • gpt-nano removed -- No longer available as an LLM option.

Community Contributors

A huge thank you to everyone who contributed to this release:

  • Nafiyad Adane (@nafiyad) -- Wikipedia search tool and Competitive Intelligence agent template
  • @ishaannk -- arXiv search and download tools
  • @SANTHAN-KUMAR -- PostgreSQL read-only MCP tool
  • Rudra (@Rudra2637) -- SuccessCriterion type field and evaluation guard
  • Rahul Vakiti (@vakrahul) -- Zero-config local LLM support and AgentRunner crash fix
  • Kartik Saini (@kart1ka) -- Retired Haiku model replacement
  • @Ttian18 -- Shift+Enter newline fallback in chat input
  • @mubarakar95 -- Reduced subprocess spawning in quickstart scripts (Windows)
  • Jamie Kim (@JamieJiHeonKim) -- README formatting and link fixes
  • Saba Siddique (@sabasiddique1) -- Roadmap Mermaid diagram fix
  • Vincent Jiang (@vincentjiang777) -- asyncio.run crash fix in GraphBuilder
  • Junaid (@juni2003) -- README org link fixes
  • Richard Tang (@RichardTang-Aden) -- Codex subscription OAuth, credential system, LLM token sourcing

Upgrading

git pull origin main
uv sync

# For the web frontend
cd core/frontend
npm install
npm run build

What's Next

  • GCU (General Compute Use) -- resource tracking and usage-based billing- Worker Bee GIT repository -- git based version control system for hive worker bee agents
  • Cost visibility -- detailed runtime log of LLM costs per node and per session- Remote agent deployment -- run agents as long-lived services with managed infrastructure

Contributors

vincentjiang777, kart1ka, and 11 other contributors
Assets 2
Loading