Sawtooth – An async, multi-tiered memory framework for LLM agents

Sawtooth Memory, a new async hierarchical memory framework for LLM agents, eliminates the blocking latency and data loss of standard memory systems by offloading summarization to background workers. The framework stores user messages in milliseconds and preserves critical facts like UUIDs and names in an immutable ledger, preventing the "Lost in the Middle" hallucination effect. Benchmarks show an 11.3x reduction in main thread latency and 100% fact recall accuracy compared to traditional summary memory.

A high-performance, non-blocking hierarchical memory framework for LLM Agents. Standard LLM memory systems like LangChain's ConversationSummaryMemory process conversation history sequentially on the main application thread. Every time a user sends a message, the entire application freezes while the system waits for an LLM to generate a new historical summary. Furthermore, these summaries suffer from the "Lost in the Middle" hallucination effect, frequently deleting specific UUIDs, names, or rules to save tokens. Sawtooth Memory eliminates this latency and data loss. It immediately stores the user's message and returns control to the application in milliseconds, offloading the heavy summarization to an asynchronous background worker. To prevent hallucinations, it extracts critical facts into an immutable ledger before summarizing. For deep architectural deep-dives, comprehensive API specifications, and advanced lifecycle configurations, please refer to the official documentation: View Detailed Architecture & API Reference DOCUMENTATION.md /HtooTayZa/sawtooth-memory/blob/main/DOCUMENTATION.md Standard Memory Blocking Sawtooth Memory Async ────────────────────────── ─────────────────────── Application Application │ │ ▼ ▼ Save Context ContextManager │ │ ▼ ├───────────────────┐ Instant Return LLM Summarizes ▼ ▼ App freezes for 5-10s Next User Turn Background Worker │ │ ▼ ▼ Next User Turn LLM Summarizes When your agent is ready to respond, Sawtooth stitches together an optimized context payload from distinct layers, ensuring critical facts are never summarized away. Agent Loop │ ▼ ┌─────────────────────┐ │ ContextManager │ │ ┌───────────────┐ │ │ │ L0 System │ │ immutable persona + tool schemas │ │ L2 Archive │ │ compressed narrative memory │ │ L1.5 Entities │ │ exact IDs, paths, UUIDs │ │ L1 Working │ │ recent raw conversation │ └───────────────┘ │ └──────────┬──────────┘ │ ▼ build prompt │ ▼ LLM API By moving compression to the background, Sawtooth achieves massive latency reductions on the main thread while maintaining 100% recall accuracy. Local GPU Benchmark NVIDIA RTX 5060 | Model: phi4-mini | 20-Message Conversation | Performance Metric | Standard Summary Memory | Sawtooth Hierarchical | Architectural Advantage | |---|---|---|---| Main Thread Latency | 64.15 seconds | 5.70 seconds | 11.3x Faster Execution | Final Prompt Payload | 506 tokens | 454 tokens | 10% Lower Token Cost | UUID / Fact Recall | Variable / Hallucinates | 100% Retained | Guaranteed via L1.5 Ledger | For full methodology, cloud comparisons, and reproducibility steps, view our Read the Performance Benchmarks /HtooTayZa/sawtooth-memory/blob/main/BENCHMARKS.md . pip install sawtooth-memory Optional dependencies for cloud providers: pip install langchain-openai langchain-anthropic langchain-google-genai Initialize the ContextManager and let the background worker handle the heavy lifting. Sawtooth is universally compatible with local air-gapped models Ollama and cloud APIs. python import asyncio from sawtooth memory import ContextManager, ContextManagerConfig from sawtooth memory.config import OllamaConfig async def main : config = ContextManagerConfig soft limit tokens=1000, hard limit tokens=2000, ollama=OllamaConfig base url="http://localhost:11434", model="phi4" async with ContextManager system prompt="You are a helpful assistant.", config=config as cm: 1. Instantly ingest messages Main thread is never blocked await cm.add message "user", "My transaction ID is txn 998877 alpha" await cm.add message "assistant", "I have noted your transaction ID." 2. Build the optimized prompt to send to your main LLM prompt = cm.build prompt print prompt if name == " main ": asyncio.run main Sawtooth eliminates the "black-box" of agent memory by providing deterministic audit trails. You can query the memory system to see exactly why a fact was retained in the prompt. python trace = cm.explain prompt import json print json.dumps trace, indent=2 Output: { "system prompt": "You are a helpful assistant.", "l2 summary lineage": "User initiated troubleshooting for router.", "User provided MAC address." , "l1 5 entities": { "key": "user transaction id", "value": "txn 998877 alpha", "origin": "Anchored via L1.5 explicit instruction" } , "l1 active messages": 4, "total tokens": 342 } Sawtooth provides a native SawtoothMemorySaver adapter, acting as a drop-in checkpointer replacement for LangGraph architectures. python from langgraph.graph import StateGraph from sawtooth memory.integrations.langgraph import SawtoothMemorySaver graph builder = StateGraph State ... add nodes and edges ... memory saver = SawtoothMemorySaver cm graph = graph builder.compile checkpointer=memory saver - Phase 1: Core Architecture - L1/L2 Hierarchical Buffer - Asynchronous Background Worker - Local Ollama & Cloud compatibility - Phase 2: Observability & Telemetry - EventBus Subsystem - Explainability Traces - Persistent JSONL Auditing Journal - Performance Benchmarking Harness - Phase 3: Advanced Architectures Up Next - Multi-Agent Memory Pooling Shared contextual state - Semantic Vector L3 Archival Memory RAG integration - Redis/Postgres Adapter for Distributed Deployments We welcome pull requests. See our CONTRIBUTING.md https://www.google.com/search?q=CONTRIBUTING.md for guidelines on how to run the test suite and ensure code quality. This project is licensed under the MIT License - see the LICENSE.md /HtooTayZa/sawtooth-memory/blob/main/LICENSE.md file for details.