Skip to main content

mini-swe-agent

Before you start

You need a Divyam account and an API key. If you don't have them yet, follow Getting Started first.

Replace the shortened divyam-sw-v1-... shown in examples with your complete API key.

The examples use divyam-as/gpt-6-sol. For another model, select its current ID in your harness or copy it from Models and pricing, and check its context limit and supported features.

1. Install mini-swe-agent​

Follow the official installation instructions.

2. Add the Divyam model​

divyam-as/

Model ID: divyam-as/gpt-6-sol. Context window: 1,050,000 tokens. Models and pricing

Model configuration scope

The divyam.yaml file configures this run's model through LiteLLM's OpenAI Chat Completions adapter. The openai/ prefix selects that adapter. Loading the file selects Divyam for this run without updating Mini's global model setting or registering a separate named provider.

Save this as divyam.yaml in your project folder. Replace divyam-sw-v1-... with your Divyam key:

model:
  model_name: openai/divyam-as/gpt-6-sol
  cost_tracking: ignore_errors
  model_kwargs:
    api_base: https://api.preview.divyam.ai/v1
    api_key: divyam-sw-v1-...
    temperature: 1
    max_tokens: 8192
Cost estimates

ignore_errors lets Mini use a custom model that is absent from LiteLLM's price list. Check charges in the Divyam console.

3. Start mini-swe-agent​

From the folder containing divyam.yaml, start the interactive CLI:

mini -c mini.yaml -c divyam.yaml

Keep mini.yaml first to retain the default agent configuration. On first run, leave the global model and key prompts blank because divyam.yaml supplies them.

In mini-swe-agent's composer
Hello, what are your capabilities?
Paste this message, then send it.
Feature limits

This setup uses the default LiteLLM workflow for local Bash commands and text-file edits. Images, MCP tools and alternate environments are outside this guide.

Check your requests​

Open the Divyam console and select Logs. Your requests appear there within a few seconds, newest first, with their status, token counts and latency. Usage and Credits & Billing catch up within a few minutes.

Local cost estimates

Some clients display zero when model prices are missing. Check Credits & Billing for Divyam charges. Local dollar limits depend on the client's price and usage data.

For filters, cached-token counts and the requested-versus-routed view, see Logs and usage.