# atlaso > A memory store that flags its own conflicts before your agent does. Python SDK, zero telemetry, open source. Atlaso is a memory layer for LLM-powered applications. Every `recall()` returns a verdict on whether the data is settled — `is_confident`, `has_disagreement`, `agreement_score`, `conflict_peers` — so agents stop confidently retrieving stale or contradictory facts. The full rendered docs site is at https://www.atlaso.ai/docs. ## How to use atlaso - Install with `pip install atlaso` (Python 3.10+). - Construct one `Memory()` per process, scope per user with `mem.for_user(user_id)`. - Write a fact with `handle.deposit(claim=..., polarity=..., evidence=..., scope=...)`. - Read with `handle.recall(query=..., k=...)` — each result carries `is_confident`, `agreement_score`, and `conflict_peers`. - Treat `has_disagreement=True` as a signal to ask the user, not to silently pick one side. - Run as an MCP server (`atlaso serve --mcp`) to plug into Claude Code, Cursor, Windsurf, Codex, Cline. - Store lives at `./.atlaso/field.db` (SQLite + FTS5). Nothing leaves the machine. ## Start here - [Welcome](https://www.atlaso.ai/docs.md): What atlaso is and the mental model in three minutes. - [Quickstart](https://www.atlaso.ai/docs/quickstart.md): Bind to a user, write a deposit, recall it. Under a minute. - [Concepts: polarity & evidence](https://www.atlaso.ai/docs/concepts/polarity-evidence.md): Field 3.0 in one page. - [Confidence & conflict](https://www.atlaso.ai/docs/concepts/confidence-conflict.md): The conflict gate — atlaso's defining mechanism. ## Getting Started - [Welcome](https://www.atlaso.ai/docs.md): What atlaso is and the mental model in three minutes. - [Installation](https://www.atlaso.ai/docs/installation.md): pip install atlaso. Python 3.10+. Zero telemetry. - [Quickstart](https://www.atlaso.ai/docs/quickstart.md): Bind to a user, write a deposit, recall it. Under a minute. ## Concepts - [Deposits](https://www.atlaso.ai/docs/concepts/deposits.md): Append-only, immutable records of what the agent learned. - [Polarity & evidence](https://www.atlaso.ai/docs/concepts/polarity-evidence.md): Four polarities, four evidence grades — Field 3.0. - [Scope](https://www.atlaso.ai/docs/concepts/scope.md): Six facets that pin a deposit to where it actually applies. - [Confidence & conflict](https://www.atlaso.ai/docs/concepts/confidence-conflict.md): The gate that flags its own disagreements before your agent does. - [Idempotency](https://www.atlaso.ai/docs/concepts/idempotency.md): Bulk imports, Stripe-style 24-hour replay safety. ## Memory API - [Memory](https://www.atlaso.ai/docs/api/memory.md): The synchronous client. Construct once, share across requests. - [AsyncMemory](https://www.atlaso.ai/docs/api/async-memory.md): The async client. Use inside FastAPI handlers and async loops. - [UserHandle](https://www.atlaso.ai/docs/api/user-handle.md): Frozen per-user view returned by for_user(). ## Data Types - [Deposit](https://www.atlaso.ai/docs/types/deposit.md): The immutable evidence record. - [SearchResult](https://www.atlaso.ai/docs/types/search-result.md): One retrieval hit with dispersion-aware diagnostics. - [Diagnostics & FMI](https://www.atlaso.ai/docs/types/diagnostics.md): Field Maturity Index — health of a user's memory. - [All result types](https://www.atlaso.ai/docs/types/results.md): AddResult, RetractResult, AddManyResult, PeekView. ## Errors - [Exception hierarchy](https://www.atlaso.ai/docs/errors.md): Every public exception, when it raises, and what's retried. ## Architecture - [Mental model](https://www.atlaso.ai/docs/architecture.md): How the moving parts fit together. - [The gate](https://www.atlaso.ai/docs/architecture/gate.md): Conflict flagging — the SDK's defining mechanism. - [Retrieval pipeline](https://www.atlaso.ai/docs/architecture/retrieval.md): Query → filter → rank → return, with dispersion math. - [Storage layer](https://www.atlaso.ai/docs/architecture/storage.md): SQLite + FTS5. On-disk schema. Where the file lives. ## Recipes - [LangChain](https://www.atlaso.ai/docs/recipes/langchain.md): Drop atlaso into a LangChain agent loop. - [LlamaIndex](https://www.atlaso.ai/docs/recipes/llamaindex.md): Plug into a LlamaIndex retriever. - [DSPy](https://www.atlaso.ai/docs/recipes/dspy.md): Memory as a DSPy module. - [OpenAI Agents](https://www.atlaso.ai/docs/recipes/openai-agents.md): Use atlaso with the OpenAI Agents SDK. - [CrewAI](https://www.atlaso.ai/docs/recipes/crewai.md): Share memory across CrewAI roles. ## Reference - [Glossary](https://www.atlaso.ai/docs/glossary.md): Every Field 3.0 term defined in one place. - [FAQ](https://www.atlaso.ai/docs/faq.md): Answers to the questions that come up most often. ## Operations - [CLI reference](https://www.atlaso.ai/docs/cli.md): Every atlaso subcommand and flag. - [MCP server](https://www.atlaso.ai/docs/mcp.md): Run atlaso as an MCP server for Claude/Cursor/etc. - [Configuration](https://www.atlaso.ai/docs/configuration.md): Env vars, file paths, transport injection. - [Admin API](https://www.atlaso.ai/docs/admin.md): Cross-tenant operations behind the literal-confirm wall. ## Research - [Research index](https://www.atlaso.ai/research): The papers behind atlaso's design. - [Growing-field substrate](https://www.atlaso.ai/research/growing-field-substrate): Why the field gets sharper with use. - [Adaptive stigmergy gates](https://www.atlaso.ai/research/adaptive-stigmergy-gates): When agents should stop coordinating. ## Optional - [Full concatenated docs](https://www.atlaso.ai/llms-full.txt): Every page above, one blob, machine-friendly. - [Source code on GitHub](https://github.com/imashishkh21/atlaso): MIT-licensed Python SDK + MCP server. - [PyPI package](https://pypi.org/project/atlaso/): `pip install atlaso`. - [Issue tracker](https://github.com/imashishkh21/atlaso/issues): Bug reports and feature requests.