LLM-backed transition function for agents. Maintains conversation history in agent state, processes inbox messages as user turns, and invokes a level 3 grid operation for LLM calls. Supports tool call loops: when the LLM requests tool calls, executes them as grid operations and feeds results back until a text response is produced. Built-in tools: complete_task, fail_task (added dynamically when tasks are pending). All other tools dispatch via the grid.
| Name: | llmagent |
| Class: | covia.adapter.agent.LLMAgentAdapter |
This adapter provides 1 MCP tool(s) for AI agent integration:
| Tool Name | Description |
|---|---|
| llmagent_chat | LLM-backed transition function for agents. The default and recommended choice for plain persona-style chat agents (customer support, character agents, summarisers, etc.) — for hierarchical goal decomposition with subgoal/complete/fail/compact tools, use v/ops/goaltree/chat instead. Maintains conversation history in agent state (state.config for llmOperation/model/systemPrompt/url, state.history for conversation). Invokes the configured level 3 grid operation for LLM calls — credentials are resolved by the level 3 operation from the user's secret store. |