antaris-spark 1.0.1

Antaris Spark — memory, safety, and context for AI agent bots (500 memory tier)

antaris-spark

Memory, safety, and context for AI agent bots — Spark tier (500 memories)

antaris-spark is the entry-level tier of the Antaris bot runtime — bundling persistent memory, safety screening, and context optimization in a single install. Capped at 500 memories for lightweight deployments.

What's Included

• antaris-memory — BM25 + co-occurrence semantic search, audit logging, memory decay, recovery
• antaris-guard — Prompt injection detection, PII redaction, rate limiting, compliance templates
• antaris-context — Token budget management, context window optimization, compression strategies

Tier Limits

Tier	Memory Cap	Package
Spark	500 memories	antaris-spark
Flame	5,000 memories	antaris-flame
Forge	25,000 memories	antaris-forge

Upgrade path: pip install antaris-flame or pip install antaris-forge — drop-in replacement, no code changes.

Installation

pip install antaris-spark

# With optional semantic reranking (MiniLM, ~22MB model):
pip install antaris-spark[semantic]

Quick Start

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

# Memory — capped at 500 entries by default
mem = MemorySystem("./workspace")
mem.load()
mem.ingest("User prefers concise answers", source="conversation")
results = mem.search("user preferences")

# Guard
guard = PromptGuard()
if not guard.is_safe(user_input):
    return  # block before reaching model

# Context
ctx = ContextManager(total_budget=8000)
ctx.set_memory_client(mem)
ctx.add_content("conversation", messages)
ctx.optimize_context()

GitHub

https://github.com/Antaris-Analytics-LLC/antaris-spark

License

Apache 2.0 — see LICENSE