InfinityDB Agent Memory Skill
InfinityDB gives autonomous AI agents persistent, verifiable long-term memory with zero external LLM extraction API costs and sub-millisecond local retrieval on SQLite WAL mode.
Capabilities
Deterministic Fact Extraction (L1):
Extracts atomic facts from dialogue and tool transcripts in <10ms with zero extra LLM token calls.
Causal Episode Tracking (L2):
Builds dependency edges between memory events to understand causal chains.
Dynamic Persona & Conflict Auto-Resolution (L3):
Automatically reconciles outdated user or agent preferences based on chronological provenance.
Enterprise Policy Injection (L4):
Guarantees corporate SOPs, compliance rules, and safety guardrails govern every recall query.
Cryptographic Proofs (L6):
Verifies an immutable SHA-256 hash linking each fact directly back to its raw interaction transcript.
Quick Usage
Python
pip install infinitydb-ai
from infinitydb import InfinityDB
db = InfinityDB("agent_memory.db")
# Ingest dialogue
db.add("Dr. Meyer prefers Rust for distributed systems.", user_id="user_1")
# Sub-5ms hybrid recall
memory = db.recall("what language does the user prefer?", user_id="user_1")
print(memory["results"])
Model Context Protocol (MCP) Server
To run InfinityDB as an MCP server for Claude Desktop and Cursor:
python -m infinitydb.mcp_server
Tools Specification
recall_memory(query: str, user_id: str): Recalls relevant facts and causal relations.add_memory(content: str, user_id: str): Commits interaction to L0-L2 memory.get_persona(user_id: str): Retrieves conflict-resolved preference profile.add_policy(rule: str, domain: str, priority: float): Injects governance constraints.verify_proof(atom_id: str): Validates SHA-256 audit proof.
