Skip to main content
HiveHive
FeaturesArchitectureIntegrationsDocsThemesGitHub
Download
Open Source — AGPL-3.0 / MIT
Staron GitHub

Your AI that learns, protects, and works while you sleep.

Free, open-source desktop AI platform. Native Rust performance, multi-agent intelligence, and military-grade privacy.

Download for FreeView on GitHub
brew install PatSul/tap/hiverecommended
curl -fsSL https://hivecode.app/install.sh | bashmacOS / Linux
macOS
Windows
Linux
Hive desktop application — AI-powered development companion with chat, agents, git ops, and 24 live panels
200k+
Lines of Rust
3,700+
Tests
21
Crates
15
AI Providers
26
Live Panels
15
Built-in Skills

Three pillars. One platform.

Hive unifies development, personal productivity, and security into a single native application that no competitor combines.

Development Excellence

Multi-agent swarm with a Queen coordinator and specialized teams. 15 AI providers with capability-aware routing, LanceDB vector memory, universal TOML skills, tool approval gate, and a built-in terminal.

  • Capability-aware AI routing (15 providers)
  • LanceDB vector memory & RAG
  • 15 built-in TOML skills + user-created, cross-model
  • Tool approval gate — diff preview for AI writes
  • A2A protocol, remote control & CLI client
  • TOON encoding — 30-40% token savings in AI prompts
  • Hybrid task pipeline — deterministic AI with validation gates
  • Prompt Library & Code Map panels
  • Context file attachments with token counting
  • Docker, K8s, AWS, Azure, GCP integrations

Personal Assistant

Email triage, calendar management, smart reminders, and document generation in 7 formats. Your AI handles the busywork so you can focus on what matters.

  • Email triage & AI drafting
  • Calendar & daily briefings
  • Smart reminders (cron, event-triggered)
  • Document export (PDF, DOCX, XLSX, +4)

Security First

HiveShield scans every request with 4 layers of protection. PII detection, secrets scanning, vulnerability assessment, and AES-256-GCM encryption. Zero telemetry.

  • 4-layer HiveShield protection
  • PII & secrets detection (11+ types)
  • Memory flush — insights extracted before compaction
  • AES-256-GCM encrypted storage
  • Zero telemetry — fully local
Multi-Agent

A swarm that thinks together

Hive's Queen coordinator assigns tasks to specialized agent teams — Architect, Coder, Reviewer, Tester, and Security — each working in isolated git worktrees. They collaborate, review each other's work, and deliver production-ready code.

  • Hierarchical Queen + HiveMind architecture
  • Teams work in parallel on isolated branches
  • Collective memory shared across all agents
  • Hybrid pipeline — curate context → execute → validate → retry
  • Budget enforcement with daily and per-agent spending limits
  • Approval gates — human-in-the-loop for sensitive operations
  • Autonomous skill acquisition — learns what it doesn't know
QueenCoordinator
ArchitectDesign
CoderImplementation
ReviewerQuality
SecurityShield
Vector Memory

Remembers everything. Forgets nothing.

LanceDB embedded vector database stores durable, searchable memory across conversations. Code is indexed with 50-line overlapping windows. Memories are categorized, importance-scored, and recalled on every query. Embedding via OpenAI or Ollama with automatic fallback.

  • LanceDB columnar vector DB — no external server needed
  • Two-table design: code chunks + durable memories
  • Background code indexer with file hash change detection
  • Memory flush extracts insights before context compaction
  • TOON context encoding — 30-40% fewer tokens per query
chunks table
source_file
content
embedding
start_line
memories table
content
category
importance
embedding
Auto-injected on every message
Remote Control

Control Hive from any device

Built-in Axum web server with real-time WebSocket streaming. Pair your phone via QR code using X25519 key exchange and AES-GCM encryption. No passwords needed. Session state persists via JSONL journal for crash recovery.

  • HiveDaemon with canonical session state
  • WebSocket streaming for chat, agents, and panels
  • QR pairing — X25519 + AES-GCM encrypted
  • REST API: /api/state, /api/chat, /api/panels, /api/agents
SendMessageChat from phone
StreamChunkReal-time tokens
SwitchPanelNavigate panels
AgentStatusMonitor agents
QR PairingX25519 + AES-GCM
A2A Protocol

Interoperable agents. Open standard.

Hive implements the Agent-to-Agent (A2A) protocol for cross-platform agent interoperability. Receive tasks from and delegate to any A2A-compatible agent. Built-in HTTP server with SSE streaming, auto-generated agent cards, and encrypted auth middleware.

  • Full A2A protocol — JSON-RPC over HTTP
  • SSE streaming for real-time task progress
  • Auto-generated agent card with skill advertisements
  • Client discovery cache for efficient repeat communication
DiscoverFetch remote agent card
AuthenticateAPI key + URL validation
DispatchRoute to Hive orchestrator
StreamSSE task progress events
BridgeA2A ↔ Hive type conversion
Universal Skills

One skill set. Every model.

TOML-based skills shared across all 27 AI providers. Each skill declares capability requirements and the SkillExecutor adapts prompts per model at runtime. File-backed persistence in ~/.hive/skills/*.toml with SHA-256 integrity verification.

  • 15 built-in TOML skills (code-review, git-commit, deploy, docker, +11)
  • Cross-model compatibility — Claude, GPT, Gemini, Grok, local models
  • File-backed persistence with SHA-256 integrity hashes
  • Capability requirements: tool_use, extended_thinking, structured_output
help
web-search
code-review
git-commit
generate-docs
test-gen
slack
jira
notion
db
docker
k8s
deploy
browse
index-docs
Approval Gate

You approve every write

AI file write operations are intercepted before execution. A diff preview card shows additions in green and removals in red with context lines. Approve or reject with one click — rejected writes return feedback so the AI adapts its approach.

  • Diff preview with syntax-highlighted additions/removals
  • One-click approve/reject for AI write operations
  • Async blocking — tool loop pauses until you decide
  • Rejected writes feed back to AI for course correction
  • Apply buttons on AI code blocks — one-click file writes
src/main.rs
-println!("debug: {}", value);
+log::info!("processed: {}", value);
+metrics.record(value);
Approve
Reject
AI Routing

15 providers. Dynamic model fetching.

Connect Anthropic, OpenAI, Google, Venice, Groq, OpenRouter, and more. 15 providers total. Hive dynamically fetches available models from each provider's API — no hardcoded lists. The CapabilityRouter ranks models by task type and automatically selects the best model based on capability scores, cost, and your learned preferences.

  • Dynamic model registry — fetches live model lists from provider APIs
  • Claude Opus 4.6, Sonnet 4.5, Gemini with Search Grounding & Code Execution
  • Capability-aware routing by task type
  • TOON encoding — ~30-40% token savings on context injection
Anthropic
OpenAI
Google
xAI
Groq
Venice AI
OpenRouter
Ollama
LM Studio
LiteLLM
HuggingFace
Mistral
Doubao
HiveGateway
Generic Local
HiveShield

4 layers between your data and the cloud

Every request passes through HiveShield before it leaves your machine. PII detection catches 11+ sensitive data types. Secrets scanning finds API keys and tokens. Vulnerability assessment blocks prompt injection and jailbreak attempts. The SecurityGateway validates every command against 14 dangerous patterns, detects SQL injection, and enforces HTTPS and domain allowlists.

  • PII detection (email, SSN, credit cards, +8 more)
  • Secrets scanning with risk levels
  • Prompt injection & jailbreak detection
  • SecurityGateway — command validation, SQL injection detection, URL filtering
Layer 1PII Detection11+ types
Layer 2Secrets ScannerActive
Layer 3Vuln AssessmentActive
Layer 4Access ControlEnforced
Self-Improving

Gets smarter every time you use it

Hive's learning engine tracks what you accept, reject, and edit. It refines model routing, evolves system prompts, and builds a pattern library from your preferred coding style — all locally, with zero telemetry.

  • Outcome tracking (accepted/rejected/edited)
  • Prompt evolution with version history
  • Pattern library from your coding style
  • Self-evaluation every 200 interactions
Week 195% acceptance rateWeek 12
Themes

8 built-in themes. Community voting.

Personalize Hive with 8 handcrafted themes — from Midnight Gold to Solarized and Nord. Browse the theme gallery, preview live, and vote for your favorites. Community ratings help surface the best themes.

  • 8 built-in themes with live preview
  • Community voting & ratings
  • One-click theme switching
  • Settings export/import with AES-256-GCM + Argon2id encryption
Midnight Gold
Ocean Blue
Forest Green
Sunset
Nord
Solarized
Dracula
Rosé Pine
Messaging

Every channel. One inbox.

Hive connects to Slack, Discord, Telegram, Teams, Matrix, WhatsApp, Signal, Google Chat, and iMessage. Triage messages, draft AI-powered replies, and manage all your channels from a single panel.

  • 9 messaging platforms supported
  • WhatsApp, Signal, Google Chat, iMessage (v0.3.3)
  • AI-powered message triage and drafting
  • Background scheduler for automated workflows
Slack
Discord
Telegram
Teams
Matrix
WhatsApp
Signal
Google Chat
iMessage
27 Panels

Everything you need. Nothing you don't.

Chat, Files, Terminal, Agents, Activity, Kanban, Git Ops, Monitor, Costs, Shield, Learning, Assistant — 27 live panels wired to real data. No mock screens. Every panel is functional from day one.

  • Chat with streaming AI responses
  • Full Git Ops (staging, commits, PRs, branches)
  • MCP server + cross-platform UI automation
  • Real-time system monitor & cost tracking
Chat
History
Files
Terminal
Specs
Agents
Activity
Workflows
Channels
Kanban
Monitor
Logs
Costs
Git Ops
Skills
Routing
Models
Learning
Shield
Assistant
Review
Network
Token Launch
Code Map
Prompts
Settings
Help

Connects to everything

40+ integrations across git hosting, infrastructure, cloud platforms, databases, project management, knowledge bases, messaging, blockchain, and AI providers.

GitHubGit
GitLabGit
BitbucketGit
DockerInfra
KubernetesInfra
AWSCloud
AzureCloud
GCPCloud
PostgresDatabase
MySQLDatabase
SQLiteDatabase
JiraPM
LinearPM
AsanaPM
NotionKnowledge
ObsidianKnowledge
SlackMessaging
DiscordMessaging
TelegramMessaging
TeamsMessaging
MatrixMessaging
WhatsAppMessaging
SignalMessaging
Google ChatMessaging
iMessageMessaging
EthereumBlockchain
SolanaBlockchain
PolygonBlockchain
xAIAI
Venice AIAI
MistralAI
DoubaoAI
OllamaAI
LM StudioAI
LanceDBDatabase
GoogleCloud
VercelCloud
SupabaseCloud
CloudflareCloud
Philips HueSmart Home
TelnyxVoice

+ OAuth2 (PKCE), MCP Client/Server, Webhooks, P2P Federation, Plugin System

The complete platform

43 capabilities across AI, development, security, integrations, and platform infrastructure — all in a single native desktop application.

AI & Intelligence

Multi-agent swarm
Queen coordinator + specialized teams (Architect, Coder, Reviewer, Tester, Security)
15 AI providers
Anthropic, OpenAI, Google, xAI, Groq, Venice, OpenRouter, Ollama, LM Studio, and more
Capability-aware routing
Automatic model selection based on task type, complexity, cost, and learned preferences
Tiered context loading
L0/L1/L2 intent classification — simple chat skips RAG, code tasks get full retrieval
Session compression
Automatic context compaction at 80% capacity with fact extraction and memory persistence
Vector memory (LanceDB)
Durable cross-conversation recall with background code indexing and semantic search
Memory lifecycle
Automatic decay, pruning, and deduplication keep memory lean and relevant
Autonomous skill acquisition
Detects knowledge gaps, researches documentation, authors and installs new skills
Self-improvement engine
Tracks outcomes, evolves prompts, learns preferences, builds pattern library

Development Tools

Git worktree isolation
Each agent team works in its own worktree for conflict-free parallel execution
Full Git Ops panel
Staging, commits, PRs, branches, gitflow, LFS — all from the UI
Tool approval gate
Diff preview for AI file writes with one-click approve/reject
Interactive terminal
Built-in shell with streaming output, kill/restart, and command history
Code Map panel
Symbol browser — functions, structs, traits, enums grouped by file with search
15 universal TOML skills
Cross-model skills that work across all providers with capability adaptation
Docker & Kubernetes
Full container lifecycle, pod management, service orchestration
Hybrid task pipeline
Deterministic AI execution: curate context, execute, validate, retry

Context & Memory

Context engine
TF-IDF scoring + semantic search + RAG with token-budgeted greedy packing
TOON encoding
Token-Oriented Object Notation — 30-40% fewer tokens per context injection
Collective memory
SQLite-backed persistent memory with 7 categories and importance scoring
Fact extraction
Extracts preferences, decisions, and code patterns from compacted conversations
MCP context tools
query_memory, search_context, list_memories exposed as MCP tools for external agents
Background code indexer
Walks project files, tracks hashes for incremental re-indexing on change

Security & Privacy

4-layer HiveShield
PII detection, secrets scanning, vulnerability assessment, access control
SecurityGateway
Command validation, SQL injection detection, URL filtering, path traversal protection
Local-first architecture
All data in ~/.hive/ — no telemetry, no analytics, no cloud dependency
Encrypted storage
AES-256-GCM + Argon2id key derivation for API keys and settings export
Works fully offline
Run with local models (Ollama, LM Studio) — no internet required

Integrations

Cloud platforms
AWS, Azure, GCP — compute, storage, functions, monitoring
Databases
PostgreSQL, MySQL, SQLite with query execution and schema introspection
Messaging
Slack, Discord, Telegram, Teams, Matrix, WhatsApp, Signal, Google Chat, iMessage
Project management
Jira, Linear, Asana — issues, tasks, and project tracking
Knowledge bases
Notion pages and databases, Obsidian vault management
Git hosting
GitHub, GitLab, Bitbucket — REST API integrations
MCP client + server
22 built-in tools with UI automation and integration dispatch

Platform & UI

27 live panels
Chat, Files, Terminal, Agents, Kanban, Git Ops, Monitor, Costs, Shield, and more
Native Rust + GPUI
120fps GPU-accelerated rendering, <50MB memory, <1s startup
Remote control
Control from phone/tablet via WebSocket — QR pairing with X25519 + AES-GCM
Terminal CLI client
Full chat TUI, model management, sync, and config from the command line
A2A protocol
Agent-to-Agent interoperability — receive and delegate tasks across platforms
P2P federation
Distributed swarm execution and fleet learning across instances
8 themes + community voting
Midnight Gold, Nord, Dracula, Solarized, and more with custom theme support
Automation workflows
Cron, event, and webhook triggers with 6 action types and visual builder

Native Rust · GPUI · 200k+ lines · 21 crates · Zero telemetry · Works offline

Open Source

Free & Open Source

Hive is completely free and open source. The app and agent runtime are AGPL-3.0; all library crates are MIT-licensed for plugin developers. We believe the best developer tools should be accessible to everyone.

Star on GitHubDownload Latest

Frequently asked questions

Hive is a free, open-source desktop AI platform built entirely in Rust. It combines a multi-agent development environment, LanceDB vector memory that remembers context across conversations, a personal assistant (email, calendar, reminders), remote control from any device, and military-grade security scanning — all in a single native application under 50MB.

Yes, completely free and open source. The app and agent runtime are AGPL-3.0 (copyleft for SaaS), while all library crates are MIT-licensed so plugin developers can build freely. There are no usage limits, no subscriptions, and no hidden costs. You just need your own API keys for cloud AI providers (or use free local models via Ollama).

Hive supports 15 AI providers: Anthropic (Claude), OpenAI (GPT), Google (Gemini), xAI (Grok), Groq, OpenRouter, Venice AI, Mistral, Doubao, HuggingFace, HiveGateway (cloud proxy), and for local/private use: Ollama, LM Studio, LiteLLM, and Generic Local. Hive intelligently routes each request to the best model based on complexity and cost.

Absolutely. Hive is local-first with zero telemetry. All data is stored on your machine in an encrypted SQLite database (AES-256-GCM). HiveShield scans every outbound request for PII, secrets, and sensitive data before anything leaves your machine. You can also run Hive completely offline using local AI models.

Hive runs natively on macOS (Apple Silicon), Windows (x64), and Linux (x64 with Vulkan). It's a compiled Rust binary — no Electron, no web wrapper — so it starts in under 1 second and uses less than 50MB of RAM.

Download the latest release from GitHub for your platform. On macOS, open the .dmg and drag Hive to Applications. On Windows, run the installer. On Linux, extract and run the binary. No dependencies required — everything is bundled.

Yes. Hive includes a built-in remote control system. A background daemon serves a web UI accessible from any device on your network. Pair your phone by scanning a QR code — the connection is secured with X25519 key exchange and AES-GCM encryption. You get real-time chat streaming, agent monitoring, and panel access from your browser.

Hive uses LanceDB, an embedded columnar vector database, to maintain durable memory across conversations. Your codebase is automatically indexed in the background with file-hash change detection. Key insights from conversations are extracted and stored with importance scoring and categories. On every message, relevant code chunks and recalled memories are injected into the AI context — so Hive remembers your preferences, decisions, and project context without any cloud dependency.

HiveHive— Your AI that works while you sleep
GitHubReleasesLicenseSecurity
© 2026 Airglow LLC. All rights reserved. Built with Rust, powered by ambition.