[Bug]: agent provider overriding not working

[dmatora]

Version

Command-line (Python) version

Operating System

MacOS

What happened?

1. pointed default agent to groq
2. tried setting importer and code monkey to ollama (tried different spelling)
3. tried to import project

ollama didn't get a single request

[hqnicolas]

same here....
I have 30 servers and
this project only uses just the :

"agent": {
"default": {
"provider":

[mercury64]

Is this still a problem?
I run Ollama locally on a Mac, works as expected.

[hqnicolas]

I run Ollama locally on a Mac, works as expected.

did you use different images, from different servers?

[mercury64]

I run Ollama locally on a Mac, works as expected.

did you use different images, from different servers?

Images?

Locally served Ollama, internal network, many different models.

[hqnicolas]

many different models.

set a 1b model as "agent": {
"default": {
"provider":

and also set a 14b + model as code monkey
you will understend
it will be 1b for code monkey to

[mercury64]

Yes, works as expected:
Log file:

2025-01-15 12:58:43,123 DEBUG [core.db.setup] Running database migrations for sqlite:///pythagora.db (config: ~/Programming/gpt-pilot/core/db/alembic.ini)
2025-01-15 12:58:43,130 DEBUG [core.ui.console] Starting console UI
2025-01-15 12:58:43,419 DEBUG [core.llm.base] Calling ollama model hf.co/LoupGarou/deepseek-coder-6.7b-instruct-pythagora-v3-gguf:Q6_K (temp=0.5), prompt length: 0.1 KB
2025-01-15 12:58:43,419 DEBUG [core.llm.ollama_client] ** MESSAGES SENT **
2025-01-15 12:58:43,419 DEBUG [core.llm.ollama_client] [{'role': 'user', 'content': "This is a connection test. If you can see this, please respond only with 'START' and nothing else."}]
2025-01-15 12:58:43,429 DEBUG [core.llm.base] Calling ollama model llama3.3 (temp=0.5), prompt length: 0.1 KB
2025-01-15 12:58:43,429 DEBUG [core.llm.ollama_client] ** MESSAGES SENT **
2025-01-15 12:58:43,429 DEBUG [core.llm.ollama_client] [{'role': 'user', 'content': "This is a connection test. If you can see this, please respond only with 'START' and nothing else."}]
2025-01-15 12:58:43,436 DEBUG [core.llm.base] Calling ollama model codellama (temp=0.5), prompt length: 0.1 KB
2025-01-15 12:58:43,436 DEBUG [core.llm.ollama_client] ** MESSAGES SENT **
2025-01-15 12:58:43,436 DEBUG [core.llm.ollama_client] [{'role': 'user', 'content': "This is a connection test. If you can see this, please respond only with 'START' and nothing else."}]
2025-01-15 12:58:43,735 DEBUG [core.llm.ollama_client] *** OLLAMA RESPONSE ***
2025-01-15 12:58:43,735 DEBUG [core.llm.ollama_client] START
2025-01-15 12:58:43,735 DEBUG [core.llm.base] Total ollama response time 0.30s, 43 prompt tokens, 3 completion tokens used
2025-01-15 12:58:43,735 INFO [core.cli.main] API check for ollama codellama succeeded.
2025-01-15 12:58:54,434 DEBUG [core.llm.ollama_client] *** OLLAMA RESPONSE ***
2025-01-15 12:58:54,434 DEBUG [core.llm.ollama_client] START
2025-01-15 12:58:54,434 DEBUG [core.llm.base] Total ollama response time 11.02s, 36 prompt tokens, 2 completion tokens used
2025-01-15 12:58:54,434 INFO [core.cli.main] API check for ollama hf.co/LoupGarou/deepseek-coder-6.7b-instruct-pythagora-v3-gguf:Q6_K succeeded.
2025-01-15 12:59:09,853 DEBUG [core.llm.ollama_client] *** OLLAMA RESPONSE ***
2025-01-15 12:59:09,854 DEBUG [core.llm.ollama_client] START
2025-01-15 12:59:09,854 DEBUG [core.llm.base] Total ollama response time 26.42s, 33 prompt tokens, 2 completion tokens used
2025-01-15 12:59:09,854 INFO [core.cli.main] API check for ollama llama3.3 succeeded.

Default and Importer and CodeMonkey model overides, sent 3 START messages, and received 3 back.
API check for each model succeeded.

[mercury64]

TO override agents in config, you must use the class name of the agent that is to be overridden.

class CodeMonkey(BaseAgent):
    agent_type = "code-monkey"
    display_name = "Code Monkey"

...
class Importer(BaseAgent):
    agent_type = "importer"
    display_name = "Project Analyist"

Config:

  "agent": {
    "default": {
      "provider": "ollama",
      "model": "hf.co/LoupGarou/deepseek-coder-6.7b-instruct-pythagora-v3-gguf:Q6_K",
      "temperature": 0.5
    },
    "CodeMonkey": {
      "provider": "ollama",
      "model": "llama3.3",
      "temperature": 0.5
    },
    "Importer": {
      "provider": "ollama",
      "model": "codellama",
      "temperature": 0.5
    }
  },

As that is how the BaseAgent loads the config, from what I can tell in the code.