Release 0.2.0 — Tape-First, Explicit, and Small

[PsiACE]

This release ships a new, unified design centered on the tape model: an append-only history with explicit anchors and controllable context. The public API is intentionally minimal, the docs are reorganized around the core mental model, and the codebase has been tightened for clarity and maintainability.

Highlights

• Tape-first state with explicit anchors and context selection.
• Minimal public API surface (LLM, Tape, TapeContext, TapeEntry, and tools).
• Clear separation between stateless chat and stateful tapes.
• Docs reorganized into Concepts, Guides, Operations, and Reference.
• New docs verification script to validate code examples against a real provider.

Quick Example

import os

from republic import LLM

llm = LLM(
    model="openrouter:openrouter/free",
    api_key=os.environ["LLM_API_KEY"],
)

print(llm.chat("Give me a one-sentence overview of Republic."))

notes = llm.tape("notes")
print(notes("Remember this for later."))
print(notes.messages())

Upgrade Notes

• Public imports are now intentionally limited at the top level. For advanced components, import from submodules.
• Batch support is provider-dependent; OpenRouter does not support batch operations.
• Tool calling requires tool-capable models; the openrouter/free router may not return tool-call-capable models.

Thanks

Derived from lightning-ai/litai with a pragmatic feel, and reshaped around the tape mental model.

---

This discussion was created from the release Release 0.2.0 — Tape-First, Explicit, and Small.