Skip to main content

DeepSeekAgent

What it is​

A minimal agent factory and IntentAgent subclass that wires a local DeepSeek R1 8B chat model (via Ollama) into the naas_abi_core intent-based agent framework, with a reasoning/math/science-oriented system prompt and a predefined set of intent triggers.

Public API​

  • create_agent(agent_shared_state: Optional[AgentSharedState] = None, agent_configuration: Optional[AgentConfiguration] = None) -> IntentAgent

    • Builds and returns a configured DeepSeekAgent:
      • Loads the DeepSeek model from naas_abi_marketplace.ai.deepseek.models.deepseek_r1_8b.
      • Registers a list of Intent entries that route to the "call_model" target.
      • Applies default AgentConfiguration(system_prompt=SYSTEM_PROMPT) if none provided.
      • Applies default AgentSharedState(thread_id="0") if none provided.
      • Uses no tools, no sub-agents, and memory=None.

  • class DeepSeekAgent(IntentAgent)

    • Empty subclass of IntentAgent (no additional behavior defined here).

Module-level constants​

  • NAME: "DeepSeek"
  • DESCRIPTION: "Local DeepSeek R1 8B model via Ollama - advanced reasoning, mathematics, and problem-solving"
  • AVATAR_URL: URL string for an avatar image
  • SYSTEM_PROMPT: multi-section prompt describing reasoning/math/science focus and response style
  • SUGGESTIONS: empty list

Configuration/Dependencies​

  • Depends on naas_abi_core.services.agent.IntentAgent for:
    • IntentAgent, AgentConfiguration, AgentSharedState, Intent, IntentType
  • Depends on a model definition:
    • naas_abi_marketplace.ai.deepseek.models.deepseek_r1_8b (imports model)
  • Default configuration applied by create_agent:
    • AgentConfiguration(system_prompt=SYSTEM_PROMPT)
    • AgentSharedState(thread_id="0")

Usage​

from naas_abi_marketplace.ai.deepseek.agents.DeepSeekAgent import create_agent

agent = create_agent()

# The returned object is an IntentAgent (specifically DeepSeekAgent)
print(type(agent).__name__)
print(agent.name)

Caveats​

  • This module does not define runtime execution methods; it only constructs the agent object and intent mappings. Actual message handling depends on IntentAgent implementation in naas_abi_core.
  • Tools, sub-agents, and memory are explicitly set to empty/None in create_agent.