New in Eldric 4.0.0

Agent Builder

AI That Builds AI

The AgentBuildWorker is an autonomous system that designs, creates, tests, and packages AI agents using advanced reasoning LLMs. Describe what you need in natural language, and watch as it architects the perfect agent for your use case.

What Is the Agent Builder?

Reasoning-Powered Design

Powered by state-of-the-art reasoning LLMs like DeepSeek R1 and Qwen QwQ, the Agent Builder doesn't just follow templates—it thinks. It analyzes your requirements, considers trade-offs, and designs optimal agent architectures through genuine reasoning.

Agentic AI Patterns

Implements cutting-edge patterns including ReAct (Reasoning + Acting), Chain-of-Thought prompting, self-reflection, and iterative refinement. The builder critiques its own work and improves until the agent meets quality standards.

Complete Agent Hub

Built agents are stored in the integrated Agent Hub—a marketplace-style registry where agents can be discovered, rated, versioned, and deployed. Think npm or Docker Hub, but for AI agents.

Docker Compilation

Agents with source code are automatically compiled using Docker containers for cross-platform compatibility. Generate .deb, .rpm, .pkg, and .exe packages from a single source, all automated.

The 8-Phase Build Pipeline

When you submit a build request, the Agent Builder executes a sophisticated multi-phase pipeline, with the reasoning LLM making decisions at each step.

1

Analyze

Deep analysis of your requirements, identifying core capabilities and constraints

2

Design

ReAct loop to select optimal tools, permissions, and architecture

3

Prompt

Generate comprehensive system prompt with Chain-of-Thought

4

Dependencies

Determine required packages, MCP servers, and integrations

5

Test

Sandbox validation with automated quality checks

6

Refine

Self-critique and iterative improvement until tests pass

7

Package

Docker-based compilation for multiple platforms

8

Publish

Deploy to Agent Hub with versioning and ratings

What Can It Build?

Code Assistants

Specialized coding agents for specific languages, frameworks, or codebases

Research Agents

Deep research capabilities with web search, academic databases, and synthesis

Data Analysts

Agents that connect to databases, analyze data, and generate insights

Automation Agents

Task automation with Bash, file operations, and system integration

DevToolBox Agents

Git integration, CI/CD monitoring, issue triage, skills, and automations

Domain Experts

Medical, legal, financial, or scientific domain specialists

Conversational Agents

Customer support, onboarding, or interactive assistants

Supported Reasoning Models

The Agent Builder leverages the latest reasoning-capable LLMs that can "think through" problems step by step, producing higher-quality agent designs.

DeepSeek R1

deepseek-r1:70b

Primary reasoning model with extended thinking capabilities

Qwen QwQ

qwq:32b

Alternative reasoning model with strong analytical skills

Qwen 2.5

qwen2.5:32b

Fallback model for faster inference

Qwen Coder

qwen2.5-coder:32b

Specialized for code generation tasks

Quick Start

# Start a build job - describe your agent in natural language curl -X POST http://localhost:8880/api/v1/agent-hub/build \ -H "Content-Type: application/json" \ -d '{ "name": "python-debugger", "description": "An AI agent specialized in debugging Python code. It should analyze stack traces, suggest fixes, explain error messages, and help identify common pitfalls. Focus on clarity and educational explanations.", "platforms": ["linux_deb", "linux_rpm", "macos_pkg"] }' # Response: {"job_id": "build-a1b2c3d4e5f6", "status": "queued"} # Check build progress curl http://localhost:8880/api/v1/agent-hub/builds/build-a1b2c3d4e5f6 # Watch the reasoning trace (see the LLM thinking!) curl http://localhost:8880/api/v1/agent-hub/builds/build-a1b2c3d4e5f6/reasoning # List all agents in the hub curl http://localhost:8880/api/v1/agent-hub/agents

API Reference

Method Endpoint Description
POST /api/v1/agent-hub/build Start a new agent build job
GET /api/v1/agent-hub/builds/:id Get build job status and progress
GET /api/v1/agent-hub/builds/:id/reasoning Get the LLM's reasoning trace
GET /api/v1/agent-hub/builds/:id/actions Get actions taken during build
POST /api/v1/agent-hub/builds/:id/cancel Cancel a running build
GET /api/v1/agent-hub/agents List all agents in the hub
GET /api/v1/agent-hub/agents/:id Get agent details
POST /api/v1/agent-hub/agents/:id/evolve Evolve agent based on feedback
POST /api/v1/agent-hub/agents/:id/fork Fork and modify an existing agent
GET /api/v1/agent-hub/agents/:id/source List agent source files
POST /api/v1/agent-hub/agents/:id/source Upload source code
PUT /api/v1/agent-hub/agents/:id/docker-config Update Docker build configuration
GET /api/v1/agent-hub/builder/status Check AgentBuilder health

Agent Evolution

Feedback-Driven Improvement

Submit user feedback, usage patterns, or performance metrics, and the Agent Builder will analyze them to generate an improved version of your agent. Each evolution is tracked with full version history.

Fork & Customize

Found an agent that's almost perfect? Fork it and apply your own modifications. The builder will incorporate your changes while preserving what works, creating a new agent tailored to your needs.