Agentic Loop
Teleton implements a Think-Act-Observe reasoning loop that enables autonomous multi-step task completion.
How It Works
When a message arrives, the agent enters an iterative reasoning cycle:
- Think. Analyze the user's request, context, and available tools to determine the best course of action.
- Act. Select and execute the most appropriate tool from the 114 available options.
- Observe. Process the tool's output and decide whether the task is complete or requires additional steps.
This cycle repeats until the task is complete or the iteration limit is reached.
Iteration Limits
The iteration limit is configurable (default: 5). This prevents infinite loops while allowing complex multi-step tasks.
config.yaml
agent:
maxIterations: 5 # Maximum tool calls per message
debounceMs: 2000 # Wait time for group messagesTool RAG
With 114 tools available, Teleton uses semantic tool selection (Tool RAG) to efficiently find the right tool for each task:
- Tool descriptions are embedded as vectors
- User intent is matched against tool embeddings
- Top-k relevant tools are presented to the LLM
- Reduces token usage and improves accuracy
config.yaml
agent:
enableToolRag: true # Enable semantic tool selection