Republic

[!WARNING] This project is currently "archived/maintained"; the main work was implemented in bub v0.4.x.

Build LLM workflows like normal Python while keeping a full audit trail by default.

Visit https://getrepublic.org for concepts, guides, and API reference.

Republic is a tape-first LLM client: messages, tool calls, tool results, errors, and usage are all recorded as structured data. You can make the workflow explicit first, then decide where intelligence should be added.

Quick Start

pip install republic

from __future__ import annotations

import os

from republic import LLM

api_key = os.getenv("LLM_API_KEY")
if not api_key:
    raise RuntimeError("Set LLM_API_KEY before running this example.")

llm = LLM(model="openrouter:openrouter/free", api_key=api_key)
result = llm.chat("Describe Republic in one sentence.", max_tokens=48)
print(result)

Why It Feels Natural

• Plain Python: The main flow is regular functions and branches, no extra DSL.
• Structured error handling: Errors are explicit and typed, so retry and fallback logic stays deterministic.
• Tools without magic: Supports both automatic and manual tool execution with clear debugging and auditing.
• Tape-first memory: Use anchor/handoff to bound context windows and replay full evidence.
• Event streaming: Subscribe to text deltas, tool calls, tool results, usage, and final state.

Authentication

Republic supports:

• static api_key
• provider maps such as api_key={"openai": "...", "anthropic": "..."}
• dynamic api_key_resolver
• provider-specific OAuth resolvers for OpenAI Codex and GitHub Copilot

Keep the README short and move auth setup into the docs:

• Quickstart: docs/quickstart.md
• Authentication guide: docs/guides/authentication.md
• OpenAI Codex example: examples/06_openai_codex_oauth.py
• GitHub Copilot example: examples/07_github_copilot_oauth.py

Development

make check
make test

See CONTRIBUTING.md for local setup, testing, and release guidance.

License

Apache 2.0

---

This project is derived from lightning-ai/litai and inspired by pydantic/pydantic-ai; we hope you like them too.