XAIBO Configuration Reference

Every agent is defined by an agent.yaml file that declares its LLM connections, cognitive modules, and wiring. This page documents the full schema.

Full Schema Example

llms:
  main-llm:
    model: "kimi-k2.6"
    base_url: "https://relay.public.cloud.xpress.ai/v1"
    api_key: "${XPRESSAI_API_TOKEN}"
    temperature: 0.7
    max_tokens: 8192
    timeout: 600
  memory-llm:
    model: "mercury-2"
    base_url: "https://relay.public.cloud.xpress.ai/v1"
    api_key: "${XPRESSAI_API_TOKEN}"
    temperature: 0.3
    max_tokens: 4096
    timeout: 300
  system2-llm:
    model: "claude-opus-4-6"
    base_url: "https://relay.public.cloud.xpress.ai/v1"
    api_key: "${XPRESSAI_API_TOKEN}"
    temperature: 0.5
    max_tokens: 16384
    timeout: 600

modules:
  pfc:
    class: "modules.pfc.PFCOrchestrator"
    llm: main-llm
    system_prompt_file: "prompts/system.txt"
    max_thoughts: 10
    context_token_budget: 150000
  hippo:
    class: "modules.hippo.MemoryOrchestrator"
    llm: memory-llm
    memory_size: 10
    mid_term_memory_size: 3
    consolidation_interval: 300
  system2:
    class: "modules.system2.System2ThinkingProvider"
    llm: system2-llm
  thalamus:
    class: "modules.thalamus.Thalamus"
  meeseeks:
    class: "modules.meeseeks.MeeseeksProvider"
    llm: system2-llm
    meeseeks:
      - name: research
        description: "Delegate a focused research task."
        system_prompt: "You are a research specialist..."
  response:
    class: "modules.response_tool_provider.ResponseToolProvider"
  desktop:
    class: "modules.desktop.DesktopProvider"
  tool_logger:
    class: "modules.tool_logger.ToolLogger"
    max_result_chars: 30000

entry_point: pfc

Sections

llms

Declares named LLM connections. Each entry defines a model endpoint that modules can reference by key.

Field	Type	Required	Description
model	string	Yes	Model identifier (e.g., kimi-k2.6, mercury-2, claude-opus-4-6)
base_url	string	Yes	API endpoint URL
api_key	string	Yes	API key. Supports ${ENV_VAR} syntax for environment variable substitution
temperature	float	No	Sampling temperature (default varies by model)
max_tokens	int	No	Maximum tokens per response
timeout	int	No	Request timeout in seconds

tip

Use environment variable substitution (${XPRESSAI_API_TOKEN}) for API keys. Never hardcode secrets in agent.yaml.

modules

Declares the cognitive components that make up the agent. Each module has a class (Python import path) and module-specific configuration.

See Cognitive Architecture for detailed documentation of each module.

PFC (Prefrontal Cortex)

Field	Type	Default	Description
class	string	--	modules.pfc.PFCOrchestrator
llm	string	--	Key from llms section
system_prompt_file	string	--	Path to system prompt text file (relative to the agent's directory, e.g., prompts/system.txt)
max_thoughts	int	10	Maximum reasoning iterations per turn
context_token_budget	int	150000	Token budget for context window

Hippo (Memory Orchestrator)

Field	Type	Default	Description
class	string	--	modules.hippo.MemoryOrchestrator
llm	string	--	Key from llms section
memory_size	int	10	Number of recent messages in short-term memory
mid_term_memory_size	int	3	Number of mid-term memory results to retrieve
consolidation_interval	int	300	Seconds between memory consolidation cycles

System 2

Field	Type	Default	Description
class	string	--	modules.system2.System2ThinkingProvider
llm	string	--	Key from llms section (typically a high-capability model)

Thalamus

Field	Type	Default	Description
class	string	--	modules.thalamus.Thalamus

No additional configuration. Acts as a safety validation layer.

Meeseeks

Field	Type	Default	Description
class	string	--	modules.meeseeks.MeeseeksProvider
llm	string	--	Key from llms section
meeseeks	list	--	List of sub-agent definitions

Each entry in meeseeks:

Field	Type	Description
name	string	Tool name exposed to PFC (e.g., research)
description	string	Tool description shown to the LLM
system_prompt	string	System prompt for the sub-agent

Response Tool Provider

Field	Type	Default	Description
class	string	--	modules.response_tool_provider.ResponseToolProvider

Provides the respond tool that the PFC uses to deliver its final answer.

Desktop Provider

Field	Type	Default	Description
class	string	--	modules.desktop.DesktopProvider

Enables computer-use via Claude Sonnet (screenshot-action loop).

Tool Logger

Field	Type	Default	Description
class	string	--	modules.tool_logger.ToolLogger
max_result_chars	int	30000	Maximum characters per tool result before truncation

entry_point

Specifies which module receives incoming messages. Almost always pfc.

entry_point: pfc

The entry point module's handle_message method is called first. It then orchestrates calls to other modules (memory, tools, system 2) as needed.

Validation Notes

• entry_point reference: The entry_point value must match a key defined in the modules section. If it references a nonexistent module, the agent will fail to start with a configuration error.
• llm references in modules: Each module's llm field must match a key defined in the llms section. A missing or misspelled LLM reference will cause a startup error.
• Required modules: The pfc and response modules are required for a functional agent (PFC needs a way to deliver responses). All other modules are optional -- omit system2 if you don't need deep reasoning, omit desktop if the agent doesn't need computer-use, etc. The thalamus module is strongly recommended for safety validation but technically optional.
• system_prompt_file: Paths are resolved relative to the agent's directory (e.g., /data/home/agents/{agentName}/). A missing file will cause an error when the module initializes.