OAIClient (OpenAICompatible) Parameter Guide (v4)

This page explains how parameter settings actually work in Agently v4, based on source behavior in the OpenAI-compatible requester.

OpenAICompatible, OpenAI, and OAIClient are aliases of the same model requester plugin in v4.

1. Where to set parameters

Layer	API	Typical keys	Scope
Global / agent settings	Agently.set_settings("OpenAICompatible", {...})	base_url model request_options stream	Default for following requests
Per-request override	agent.options({...})	temperature top_p max_tokens tools	Current request prompt

Both sources eventually become request body fields.

2. Effective precedence in source logic

In OpenAICompatible.generate_request_data():

1. model is force-written from plugin settings
   plugins.ModelRequester.OpenAICompatible.model > default_model[model_type]
   (agent.options({"model": ...}) is overwritten)
2. stream is force-written from plugin settings
   explicit plugins...stream, or defaults: chat/completions=true, embeddings=false
   (agent.options({"stream": ...}) is overwritten)
3. Other request options
   agent.options({...}) overrides request_options

3. Recommended setup for temperature/top_p

3.1 Global defaults

from agently import Agently

Agently.set_settings("OpenAICompatible", {
  "base_url": "https://api.openai.com/v1",
  "api_key": "YOUR_API_KEY",
  "model": "gpt-4o-mini",
  "request_options": {
    "temperature": 0.2,
    "top_p": 0.9,
    "max_tokens": 800
  }
})

3.2 Per-request overrides

agent = Agently.create_agent()

result = (
  agent
  .input("Explain RAG in one paragraph")
  .options({
    "temperature": 0.7,
    "max_tokens": 300
  })
  .start()
)

4. Two different tools mechanisms

4.1 Agently tooling flow (@agent.tool_func + agent.use_tools)

This is Agently orchestration: tool judging + tool execution + result injection into action_results.

from agently import Agently

agent = Agently.create_agent()

@agent.tool_func
def add(a: int, b: int) -> int:
  return a + b

agent.use_tools(add)
print(agent.input("Use tool to calculate 12+34").start())

4.2 Provider-native OpenAI tools (pass-through)

Put raw tools/tool_choice in request_options or agent.options():

agent.options({
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Get current weather",
        "parameters": {
          "type": "object",
          "properties": {
            "city": { "type": "string" }
          },
          "required": ["city"]
        }
      }
    }
  ],
  "tool_choice": "auto"
})

These two approaches can coexist, but they are not the same mechanism.

5. URL, model type, and message shaping

• full_url overrides base_url + path_mapping[model_type]
• model_type=chat sends messages
• model_type=completions sends prompt
• model_type=embeddings sends input and defaults to non-streaming
• strict_role_orders affects role normalization
• Attachment prompts auto-enable rich_content

6. Auth choices

Supported ways:

• api_key
• auth (api_key / headers / body)

Agently.set_settings("OpenAICompatible", {
  "base_url": "https://api.example.com/v1",
  "auth": {
    "headers": {
      "Authorization": "Bearer xxx",
      "X-Project": "demo"
    }
  }
})

7. Fast troubleshooting

1. temperature not working: use request_options or agent.options() instead of root-level options.
2. stream not working: plugin-level stream overrides request-level value.
3. model not working: model is force-written at plugin level.
4. tools not working: confirm whether you want Agently tool orchestration or provider-native tools.
5. endpoint errors: try full_url first.

8. Debugging

from agently import Agently

Agently.set_settings("debug", True)
# or:
# Agently.set_settings("runtime.show_model_logs", True)

You will see request-stage fields like request_url, request_options, and stream.

9. Source references

• agently/builtins/plugins/ModelRequester/OpenAICompatible.py
• agently/builtins/plugins/PromptGenerator/AgentlyPromptGenerator.py
• agently/builtins/agent_extensions/ToolExtension.py
• agently/core/Agent.py

Related docs

• OpenAI model setup: /en/models/openai
• Model settings overview: /en/model-settings
• Full settings reference: /en/settings