Skip to main content

GemmaAgent

What it is​

A thin IntentAgent wrapper that configures a local Gemma3 4B chat model (via the marketplace model module) with a predefined system prompt and a set of conversation/intention triggers.

Public API​

  • Constants

    • NAME: Agent display name ("Gemma").
    • DESCRIPTION: Short description of the agent/model.
    • AVATAR_URL: Avatar image URL.
    • SYSTEM_PROMPT: Default system prompt describing behavior/style.
    • SUGGESTIONS: Empty list (no suggestions defined).

  • Functions

    • create_agent(agent_shared_state: Optional[AgentSharedState] = None, agent_configuration: Optional[AgentConfiguration] = None) -> IntentAgent
      • Builds and returns a configured GemmaAgent.
      • Loads the chat model from naas_abi_marketplace.ai.gemma.models.gemma3_4b.
      • Registers a set of Intent entries mapping various phrases to intent_target="call_model".
      • If not provided:
        • Creates AgentConfiguration(system_prompt=SYSTEM_PROMPT)
        • Creates AgentSharedState(thread_id="0")

  • Classes

    • GemmaAgent(IntentAgent)
      • No additional behavior beyond IntentAgent (empty subclass).

Configuration/Dependencies​

  • Depends on core agent framework types:
    • IntentAgent, AgentConfiguration, AgentSharedState, Intent, IntentType from naas_abi_core.services.agent.IntentAgent.
  • Depends on the Gemma model module:
    • naas_abi_marketplace.ai.gemma.models.gemma3_4b (used as model.model when passed to chat_model).
  • Tools/agents/memory:
    • tools = [], agents = [], memory = None (no extensions configured in this file).

Usage​

from naas_abi_marketplace.ai.gemma.agents.GemmaAgent import create_agent

agent = create_agent()
# Use `agent` according to the IntentAgent interface provided by naas_abi_core.

Caveats​

  • This module only constructs the agent; actual runtime behavior (e.g., how call_model is invoked, how intents are matched, and how the model is executed via Ollama) is handled by IntentAgent and the imported model module.