Capabilities
Agentron provides local AI agent orchestration, workflow automation, and multi-agent design in a single, self-hosted platform: Level 1 (ReAct + tools) for chat/RAG, Level 2 (static multi-agent) for workflows, Level 4 (recursive/self-building) for heap.
Feature matrix
| Feature | Description | Primary actions |
|---|---|---|
| Agents | Create, edit, delete node and code agents | create_agent, update_agent, delete_agent, get_agent |
| Workflows | Multi-agent orchestration in graphs | create_workflow, update_workflow, add_workflow_edges, get_workflow |
| Tools | Native, HTTP, MCP tools | create_tool, update_tool, list_tools, get_tool |
| Runs | Execute workflows/agents, inspect results | list_runs, get_run |
| LLM Providers | Configure OpenAI, Anthropic, Ollama, etc. | list_llm_providers (config via Settings) |
| Agentron (Chat) | Natural-language chat that creates/edits via tool calls | Built-in; see Agentron (Chat) |
| Sandboxes | Container-engine sandboxes for code execution (Container engine setup) | create_sandbox, execute_code |
| Custom Functions | JavaScript/Python/TypeScript as tools | create_custom_function |
| Knowledge / RAG | Document ingestion and retrieval | UI-based; agents can use RAG collections |
| Feedback | Rate agent outputs for learning | UI-based; stored for prompt refinement |
| Files | Upload context files | list_files |
| Remote Servers | SSH tunnel to remote LLMs (e.g. Ollama) | test_remote_connection, save_remote_server |
What users can do
- Design agents: Node graphs or code-based agents
- Build workflows: Connect agents, set max rounds
- Add tools: HTTP, MCP, or custom code
- Run and debug: Execute workflows, inspect runs
- Chat with Agentron: Natural-language commands; see Agentron (Chat)
- Rate and improve: Feedback on assistant responses
Limits and constraints
- Node agents require
llmConfigIdwhen using an LLM node - Standard tools (
std-*) can only update inputSchema/outputSchema - Workflows need
maxRoundswhen using looping edges to avoid infinite runs - Diagnose before fix: Always use get_run, get_workflow, get_agent, get_tool before applying updates
Last updated on