Effortless AI Workflows
Transform natural-language requests into modular AI pipelines with live tracking and full auditability.
Turning Your Words into a Plan
From natural language to structured workflows
Our AI-powered system transforms your requests into clear, actionable steps, orchestrating a team of specialized agents to get the job done.
Translates NL into workflows
The HiveProcessor is the brain that takes a HiveInput (your user's prompt plus optional team/user context), stores an initial "hive_request" to the database, and then calls out to AI (via OpenAI_generate_response) to parse that prompt into a workflow:
- A UUID workflow ID
- A set of discrete tasks, each assigned to a specific agent
- Explicit dependencies between tasks
Orchestrates asynchronous task execution
Your TaskManager (a thread-safe singleton) polls the database for new "pending" tasks and optionally listens on PostgreSQL channels for notifications. It claims tasks whose dependencies are satisfied, and hands them off—via the HiveProcessor and HiveCommunicator—to the appropriate agent implementation.
Provides a unified agent framework
At the core is BaseAgent, an abstract class that loads configuration, logging, OpenAI credentials, and a HiveDB connection. It defines standard methods for input validation, progress reporting, error handling, and saving results back to the database.
Ships with a library of built-in agents
URLAnalyzerAgent
Fetches and parses web pages for structured insights
BlogAgent
Composes multi-section blog posts
AnkhAgent
Executes code snippets securely
LinkedInMonitorAgent
Tracks LinkedIn profiles and engagement
Manages agent metadata & access control
The HiveRegistry lazily loads all agent metadata from the database, enforces scope (global/team/user), and instantiates agent classes on demand. This lets you add or update agents without restarting the server.
Persists everything in a central database
All requests, AI responses, agent requests and results, error logs, and status updates are stored in a Supabase/PostgreSQL backend via HiveDB. This ensures durability and lets you query or replay workflows later.
Turning Your Words into a Plan
Discovering Available "Helpers"
Before doing anything, the system checks which intelligent "helpers" (we call them agents) are on hand. Each helper knows how to do one thing really well—like analyzing a webpage, translating text, or drafting a blog post.
Asking the AI to Sketch a Recipe
Next, the system hands your request (plus a quick list of those helpers) to a smart language model. It asks: "Given these helpers and what they can do, what are the individual steps needed to satisfy this request? And which helpers should run each step?" The AI then returns a simple recipe: a list of steps, which helper handles each one, and which steps need to finish before others can start.
Cleaning Up & Scheduling
That recipe is double-checked for sense (no missing steps, no impossible dependencies) and then each step is saved off as a mini-job in a queue. At this point, you have a clear, ordered "to-do" list of tasks that exactly matches your original request.
How the Helpers ("Agents") Get Chained Together
Waiting in Line
Each helper sits idle until all the steps it depends on are done. For example, a "summarize article" helper won't start until the "fetch webpage" step finishes.
Doing Its Job
When its turn comes, a helper grabs whatever output it needs from earlier steps, does its thing (e.g. runs an AI call to summarize, translate, or generate content), and then hands its result back to the system.
Triggering the Next Step
As soon as one helper finishes, the system checks: "Which pending helpers were waiting on that result?" Any whose prerequisites are now met get kicked off immediately. This keeps everything flowing without unnecessary delays.
Why This Matters
Modularity
Each agent is a plug-and-play specialist. You can add new helpers or improve existing ones without rewriting the entire pipeline.
Clarity
Even complex requests become a clear sequence of steps. You always know which part is running right now and what's next.
Real-Time Updates
As each step completes, progress can be streamed back to users in a UI—no more guessing when a long process will finish.
Audit & Replay
Because every step and its AI call are logged, you can review exactly what happened, or even rerun parts of a workflow with tweaks.
Example pages
Core Capabilities
Key features of Hive's AI orchestration platform.
Hive transforms your natural-language requests into modular AI workflows, combining specialized agents with live progress updates and full auditability.
- Natural-Language → Structured Workflow
- Convert plain-English requests into ordered, dependency-aware task plans automatically.
- Rich Agent Library & Chaining
- Leverage a plug-and-play suite of AI agents that seamlessly hand off outputs to one another.
- Real-Time Progress
- Stream live task updates via WebSockets so you always know what's happening.
- Audit & Replay
- Review every AI prompt, response, and task result to ensure full traceability and replayability.
Commonly asked questions
If your question is not on the list, feel free to get in touch here or open up an issue on GitHub.
Hive converts your free-form requests into structured, multi-step AI workflows, choreographing specialized agents to handle each task seamlessly.