Every capability of the Hive platform, how the pieces fit together, and training videos to get you started.
Everything you need to build, deploy, and monitor multi-agent AI workflows.
Manages the Docker infrastructure that powers the platform. Pull/remove images, deploy/stop core services, view real-time container logs, and control the MCP DevServer.
Run LLMs locally without cloud API keys. Pull the Ollama container, browse models from the Ollama library or HuggingFace, detect GPU, and chat with local models directly.
Securely store API keys for LLM providers and tool services. Encrypted at rest using platform credential stores (DPAPI on Windows, AES-256-GCM on Linux/Mac). Auto-migration encrypts existing plaintext keys on first boot.
Create and manage reusable agent templates. Each agent has a name, description, system prompt, and LLM provider configuration. Templates become live agents when added to a workflow and deployed.
Create reusable prompt skills (templates with categories) that attach to agents. Building blocks for agent behavior — "Summarize Text," "Extract Entities," "Draft Email."
Browse and configure installed plugin tools and MCP (Model Context Protocol) servers available to agents. View tool descriptions and parameter requirements.
Browse and download tool packages from a GitHub-hosted NuGet feed. One-click install adds tools to the platform.
View and manage downloaded tool packages. Each package contains tool functions that agents can use. Configure package-specific parameters.
| Package | Description |
|---|---|
| Agents.Tools.FileIO | File read/write/list/delete with sandboxed workspace |
| Agents.Tools.CodeExec | Shell, .NET, and Git CLI command execution |
| Agents.Tools.Git | Full git operations (21 tools: clone, commit, push, merge, etc.) |
| Agents.Tools.GitHub | GitHub API via Octokit (repos, issues, PRs, releases, branches) |
| Agents.Tools.WebSearch | Web search via SerpAPI + URL fetch |
| Agents.Tools.Evals | LLM-as-judge scoring + Ollama embedding similarity |
| Agents.Tools.Adzuna | Job search via Adzuna API (12+ countries) |
| Agents.Tools.JSearch | Job search via RapidAPI (Indeed, LinkedIn, Glassdoor, ZipRecruiter) |
| Agents.Tools.TheMuse | Job search via The Muse (free, no API key) |
| Agents.Tools.Email | Generic email send (SMTP) and read (IMAP) |
| Agents.Tools.GmailReceive | Read Gmail via IMAP with Google OAuth2 |
| Agents.Tools.GmailSend | Send Gmail with Google OAuth2 |
| Agents.Tools.ImapEmail | IMAP email client |
| Agents.Tools.SmtpEmail | SMTP email client |
The core visual canvas for building AI workflows. Drag agents onto a canvas, connect them with edges defining communication paths, configure per-agent settings, and deploy as Docker containers. Supports Scheduler, Input Gateway, Output Gateway, and A2A Proxy nodes.
Send messages to deployed workflows in real-time. Supports @AgentName routing, per-message quality scores (auto-eval badges), cost tracking, token usage, and session management.
Add time-based triggers to workflows. Fires on a cron schedule, sends a configured message to a connected agent. Trigger history recorded per workflow.
Enforce structured output formats. Three parser types: JSON Schema, Regex Extract, and Markdown Sections. Auto-retries with error feedback if format doesn't match. Format instructions auto-injected into the agent's system prompt.
Detect and redact personally identifiable information from agent responses. Configurable per workflow node — choose which PII categories to filter.
Three-tier quality evaluation. Tier 1: static checks. Tier 2: reference metrics (BLEU, ROUGE, BERTScore). Tier 3: LLM-as-judge. Three modes: Monitor, Gate, Self-Correct. Custom rubrics, eval reports with per-metric badges.
When an agent returns a list of results, automatically rank and score each item. Four criteria: keyword boost, recency decay, semantic relevance, LLM-as-judge. Returns top-K ranked results.
Assign specific tools to each agent node and configure their parameters. API keys pulled from Key Store, workspace paths auto-resolved.
Use any Docker image instead of the built-in agent image. Custom images must expose A2A protocol endpoints. Supports Python, Java, Go, Rust agents. Key declarations pass secrets and config.
Expose a workflow as an external API endpoint. Supports API key and OIDC/JWT auth, input guardrails via Guardian, configurable port for stable URLs, and Cloudflare Tunnel support for public access.
A forward proxy that transparently forwards requests to a remote Hive instance. Other agents call it like a regular peer — they don't know they're talking to a proxy. Guardian safety checks run on every outbound message.
Every LLM interaction logged with token counts, cost, and quality scores. Cost calculated at insertion time using provider pricing. Streams to Dashboard via SSE. Ollama interactions tracked at $0.
An MCP server exposing 51 tools for building workflows entirely through AI conversation. Any MCP-compatible client (Claude Desktop, Cursor, Windsurf) can create agents, skills, workflows, configure nodes, and deploy via natural language.
Single-click desktop installer. On launch: splash screen, Docker auto-detection, image pull, infrastructure deploy, Dashboard opens. Auto-updates via GitHub Releases.
Headless web server on port 8001. No Electron — access via any browser. For remote/team access or Docker deployment. Same codebase, same capabilities.
Short, narrated walkthroughs of every feature. Watch them in order or jump to what you need.
Looking for early access, licensing, or a demo? Drop a line.
Get in touch