antaris-forge 1.0.5

Antaris Forge — memory, safety, and context for AI agent bots

antaris-forge

Memory, safety, and context for Antaris Forge bots.

antaris-forge is the stable runtime package for Forge-powered AI agents. It bundles the three core infrastructure packages in a single install.

Packages

Package	Purpose
antaris-memory	File-based persistent memory — BM25 search, audit logging, semantic re-ranking, PPMI co-occurrence
antaris-guard	Prompt safety screening and injection detection
antaris-context	Token budget management and context window optimization

Quick Start

from antaris_memory import MemorySystem
from antaris_guard import PromptGuard
from antaris_context import ContextManager

# Memory
mem = MemorySystem("./workspace")
mem.ingest("Patient: John, 42M, elevated liver enzymes noted on 2024-01-15")
results = mem.search("liver")

# Safety
guard = PromptGuard()
result = guard.analyze("Tell me about liver function tests")

# Context
ctx = ContextManager(total_budget=8000)
ctx.add_turn("system", "You are a helpful medical assistant.")
window = ctx.render()

Design

• File-based — JSON/JSONL storage, zero databases
• Zero mandatory dependencies — pure Python core
• Thread-safe — concurrent reads and writes handled correctly
• Privacy-first — HMAC-SHA256 PII anonymization, local-only by default

Versioning

antaris-forge is the stable Forge fork of antaris-suite.

• Bug fixes and security patches land here
• New features and experimental work land in antaris-suite first
• Production Forge bots should pin to antaris-forge~=1.0

License

Apache 2.0 — see LICENSE