A coding-agent CLI that works the same way regardless of which model is answering - Claude, GPT, or an open-weight model like Qwen, DeepSeek, GLM, or Llama running locally via Ollama, vLLM, LM Studio, or any other OpenAI-compatible server.

Why

Most agent CLIs lean on a provider's native function-calling API and assume near-perfect adherence to its exact tool-call JSON schema. Open-weight models frequently emit malformed, incomplete, or off-format tool calls against that API, so the whole agent loop breaks.

Polyglot treats tool invocation as a text-parsing problem instead: tools are described to the model in the system prompt, and a fault-tolerant streaming parser extracts and repairs tool calls from the model's raw output - trailing commas, single quotes, near-miss tool names, models that default to OpenAI-style {"name":..., "arguments":...} JSON instead of the taught envelope, all handled the same way. The same parser and executor run underneath every provider, so behavior doesn't silently diverge between "well-behaved" and "flaky" models.

Install

npm install -g @usepolyglot/cli

Or build from source - see the install guide.

Quick start

Point it at a local model (Ollama):

export POLYGLOT_PROVIDER=openai-compatible
export POLYGLOT_MODEL=qwen2.5-coder                # any model you've pulled
export POLYGLOT_BASE_URL=http://localhost:11434/v1 # default if unset
polyglot

Or Claude:

export POLYGLOT_PROVIDER=anthropic
export POLYGLOT_MODEL=claude-sonnet-4-5
export ANTHROPIC_API_KEY=sk-ant-...
polyglot

Documentation

Full documentation is at usepolyglot.dev/docs - first-run walkthroughs, the settings/env reference, permission modes, web search, MCP, scripting, running offline, and the model/server compatibility matrix.

Contributing

See CONTRIBUTING.md - dev setup, code style, the changeset/release flow, and what kinds of contributions are most useful right now (fixture tests from real messy model output, Windows testing, and additional OpenAI-compatible provider quirks especially).

Security

See SECURITY.md to report a vulnerability. How Polyglot handles your data is documented at usepolyglot.dev/docs/concepts/data-handling.

License

Functional Source License, Version 1.1, ALv2 Future License - see LICENSE.

Source is fully visible; you can read it, modify it, self-host it, and use it for your own internal purposes freely. The one thing it restricts is launching a competing commercial product or service built on this code. Two years after each version is published, that version automatically converts to the Apache License, Version 2.0.