Switch models inside one agent session to cut AI costs
Zero is an open-source that lets the model be swapped during the same work session. It connects to more than 25 providers, including OpenAI, Anthropic, Gemini, DeepSeek, Qwen, and Groq, and it can also use through Ollama or LM Studio. The `/model` command changes the model without losing the .
Cheap and fast models handle routine work such as reading files, summaries, , and . A is used only for steps that need stronger reasoning. A is used when code or data should stay on the user’s machine.
This changes the cost pattern because expensive tokens are used only for the smaller part of the workflow that really needs them. Sessions are saved as disk files, the tool is a single Go binary, it has no telemetry, and it includes a that streams JSON for automation.
Key points
- Zero is an open-source with swappable models.
- The `/model` command changes models inside one session while keeping context.
- Cheap models are used for routine work, a for harder reasoning, and a for sensitive code or data.
- The approach reduces cost by limiting expensive tokens to the steps that need them.
- It stores sessions as files, runs as one Go binary, has no telemetry, and supports headless JSON output.