{"slug": "show-hn-loom-memory-that-follows-the-code-you-edit", "title": "Show HN: Loom – Memory that follows the code you edit", "summary": "ClickHouse released Loom, an open-source, self-hosted MCP server that gives coding agents such as Claude Code, Codex and Cursor long-term shared memory, a repository code index and document search with citations. Loom stores agent context and index data in ClickHouse, can use Postgres at first boot for live memory and knowledge state, and supports namespaces, roles and scoped tokens in multi-user deployments. The project reports lower model cost for coding agents as a measured result.", "body_md": "**Shared memory, document search and code intelligence for AI agents.**\n\n  For Claude Code, Codex, Cursor and your own agents.\n\n  \n  English · [简体中文](https://github.com/ClickHouse/loom/blob/main/i18n/zh-CN/README.md) · [日本語](https://github.com/ClickHouse/loom/blob/main/i18n/ja/README.md) · [한국어](https://github.com/ClickHouse/loom/blob/main/i18n/ko/README.md) · [Português (Brasil)](https://github.com/ClickHouse/loom/blob/main/i18n/pt-BR/README.md) · [Español](https://github.com/ClickHouse/loom/blob/main/i18n/es/README.md)\n\n[What Loom is for](#what-loom-is-for) ·\n  [Benchmarks](#measured-results) ·\n  [Get started](#get-started) ·\n  [Capabilities](#why-loom) ·\n  [Integrations](#integrations) ·\n  [Documentation](https://clickhouse.github.io/loom/)\n\nLoom is an open-source, self-hosted MCP server that gives your agents long-term, shared\nmemory: what one agent learns in a session, the next agent or tool can find.\nIn an authenticated team deployment, each agent uses a token scoped to namespaces and\nread or write roles. [Check shared memory and denied access](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/connect-a-coding-agent.md#share-a-team-namespace-with-separate-credentials).\n\nAgents connect through MCP, HTTP, Python or TypeScript. Self-host with Docker or\n[on Kubernetes with the Helm chart](https://github.com/ClickHouse/loom/blob/main/docs/operations/kubernetes.md); agent context and\nindex data live in [ClickHouse](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/self-hosted-clickhouse.md), and a\ndeployment can choose [Postgres at first boot](https://github.com/ClickHouse/loom/blob/main/docs/operations/postgres.md) for live\nmemory and knowledge state. The dashboard brings memories, repositories, tasks and runs\ninto one place.\n\n- **Long-term memory shared across your coding agents** , with plugins that load context\nat session start and capture the session as it runs.[Agent memory](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/agent-memory.md) ·[Connect a coding agent](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/connect-a-coding-agent.md)\n- **Memory for an agent application** , per user, from the Python or TypeScript\nSDK.[Your first agent](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/first-agent.md)\n- **A multi-user MCP memory server** with namespaces, roles and tokens.[Multi-user mode](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/quickstart.md#start-in-multi-user-mode) ·[Namespaces and sharing](https://github.com/ClickHouse/loom/blob/main/docs/features/namespaces-and-sharing.md)\n- **A code index of your repositories** that coding agents query for definitions,\ncallers and change impact.[Code index](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/code-intelligence.md) ·[Language support and limits](https://github.com/ClickHouse/loom/blob/main/docs/features/code-index/overview.md#language-support)\n- **Memories linked to the code they are about** , which follow it through refactors, are\nflagged when it changes and appear when an agent reads or edits it.[Memory linked to code](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/memory-linked-to-code.md)\n- **Lower model cost for coding agents** , as[measured](#measured-results) below.[Coding-agent cost](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/coding-agent-cost.md)\n- **Document search with citations (RAG)** , including PDF,`.docx` and`.pptx` files the\nextractor service converts.[Document search](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/document-search.md)\n- **Shared context for multi-agent workflows** : shared memory, task inboxes and\ndecisions.[Multi-agent](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/multi-agent.md)\n- **Memory, recall, audit and erasure for autonomous agents.**[Autonomous agents](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/autonomous-agents.md)\n- **A record of agent work** , with searchable conversations (when recording is on),\nsession briefs to resume from, runs with their goals, outcomes and model costs when\nsupplied, and PR and tracker history.[Runs and PR history](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/runs-and-pr-history.md)\n\n**Memory for a software factory.** Loom gives implementation and review agents shared\nlessons, issue context and source provenance across sessions. Scope each principal's\naccess with tokens, await required writes and task handoffs in your application, and\nrecheck code-linked lessons against the current source before review. The same service\nsupports multi-agent workflows and autonomous bots; your orchestration runs the agents,\ntests and approvals. Offering a tool does not guarantee the model uses it.\n[Multi-agent workflow](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/multi-agent.md) ·\n[Autonomous-agent memory](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/autonomous-agents.md)\n\nEvery use case, and when another tool fits better: [use cases](https://github.com/ClickHouse/loom/blob/main/docs/use-cases/index.md).\n\nRead [Choosing memory for a software factory](https://github.com/ClickHouse/loom/discussions/939)\nfor memory architecture choices, data ownership and operating responsibilities.\n\nLoom is not an in-process library for one agent's short-lived memory, nor an agent\nframework. [When Loom does not fit](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/overview.md#when-it-does-not-fit)\n\n| Evaluation | With Loom | Without Loom | Efficiency gain | Scope | \n|---|---|---|---|---|\n| Coding-agent answer quality | **0.84** | 0.78 | **20% lower** model cost | 24 questions × 3 paired runs | \n| Coding-agent file localization | **96.4%** | 91.4% | **21% lower** model cost | 36 questions × 10 paired runs | \n| Memory vs full-history answers | **94%** | 90% | **91% fewer** reader prompt tokens | 100 paired LongMemEval-S questions | \n\nBoth coding setups use Claude Code (`claude-sonnet-5`) on the ClickHouse codebase;\ncost is the whole session's model bill, excluding Loom's server-side model calls, hosting\nand indexing. Answer quality\nmeasures coverage of a verified answer's key points, with a contradiction penalty.\nFile localization counts whether the right file is among the up to three files the agent is\nasked to name at the end of its answer.\n\nThe memory row compares retrieved memories with full conversation history using\nthe same `gpt-5` reader. Token savings exclude memory construction; the 94% vs 90%\naccuracy difference is descriptive, not statistically established.\n\n**Separately, Loom's `memory.ask` configuration scored 96.6% answer accuracy on all\n500 LongMemEval-S questions.** This is a product evaluation with its own configured\nanswer models, not the fixed-reader comparison above.\n\nResults are specific to the tested workloads.\n\n[Protocols and reproducible artifacts](https://github.com/ClickHouse/loom/blob/main/docs/benchmarks/results.md) ·\n[detailed results](#benchmark-details) ·\n[evaluate on your workload](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/evaluate.md)\n\n- **Personal use—the default:** no login, localhost-only access and a`dev/local` namespace. One person can connect multiple agents.\n- **Team use:** authenticated accounts and namespace access controls.[Start in multi-user mode from the first boot](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/quickstart.md#start-in-multi-user-mode) .\n\nBoth modes support the same Loom features, including configured LLMs and embeddings.\nSingle-user is not a reduced-feature edition. You can also\n[switch an existing installation without deleting its data](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/quickstart.md#2-open-the-dashboard);\ngrant new accounts access to the existing namespaces.\n\nTo try it with nothing installed, open it in a codespace:\n\n[(](https://codespaces.new/ClickHouse/loom?quickstart=1)[what it runs and who can reach it](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/quickstart.md#try-it-in-github-codespaces))\n\nFor the default personal installation, with Docker installed (on Windows, in a WSL 2\nterminal where `docker` works: Docker Desktop with its WSL integration on, or Docker Engine\ninstalled in the distribution):\n\n```\ncurl -fsSL https://github.com/ClickHouse/loom/releases/latest/download/install.sh -o loom-install.sh && sh loom-install.sh && rm loom-install.sh\n```\n\nRun it from your home directory, or another\n[directory only you can write to](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/quickstart.md#fast-path-released-image-no-clone).\nThe installer starts Loom and ClickHouse, applies migrations and prints the dashboard URL:\n**[http://localhost:7777/](http://localhost:7777/)**; on a new install it then checks that a memory saves and reads\nback. On a terminal it also offers to connect the coding agents it finds (Claude Code, Codex,\nVS Code) and, on a new install, opens the dashboard. The image includes the server,\nworkers, dashboard and Python, TypeScript and Go compiler indexers, plus the C++ one on\n`amd64`; TypeScript dependencies are prepared automatically. [Quickstart and pinned releases](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/quickstart.md)\n\nThe installation lives in `~/.loom` (`LOOM_INSTALL_DIR` chooses another directory): the\nCompose file and the startup settings in `~/.loom/.env`. Running the install command again\nupgrades in place and keeps your `.env` settings.\n\nKeep single-user mode localhost-only. For shared access, follow the\n[team deployment guide](https://github.com/ClickHouse/loom/blob/main/docs/operations/deployment.md#identity-modes), including TLS.\n\n**The stock installation boots without keys or network.** Storage, the dashboard,\nstructural code indexing and semantic search work, on a small local embedding model the\nimage carries; the fake LLM gives no AI-generated answers.\n\n| Component | Configure for real use | Without configuration | \n|---|---|---|\n| Embeddings | Provider, embedding model, dimensions and API key. | `local` (bge-small on CPU): semantic retrieval, weaker than a hosted model. | \n| LLM | Provider, model and API key. | `fake-echo-1` : no meaningful extraction, summaries or answers. | \n\nOpen **Settings → Models & keys**, choose both providers and models, enter their\nkeys, then click **Test & apply**. No Docker restart is needed; changing the embedding\nspace rebuilds the existing index. The dashboard's embedding choices are hosted\nproviders: to keep `local` and add an LLM, set the LLM lines in `~/.loom/.env` instead\n([Models and keys](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/models-and-keys.md#claude-for-the-llm)).\n\n**One OpenAI API key can serve both roles.** Claude needs a separate embedding provider.\nOpenAI, Gemini and OpenAI-compatible embedding endpoints are supported. Your agent's\nCLI login or subscription does not supply Loom's provider keys.\n\nPrefer environment configuration? Edit `~/.loom/.env`, then recreate the server:\n\n```\ndocker compose --project-directory \"$HOME/.loom\" up -d --force-recreate loom\n```\n\nA key alone does not change the pinned models (the local embedder and the fake LLM); set\nthe provider and model too.\nSaved dashboard settings take precedence over `.env`. Follow\n[models and API keys](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/models-and-keys.md) for provider recipes\nand health checks.\n\n**Everything else is in one file.** [`loom.example.yaml`](https://github.com/ClickHouse/loom/blob/main/loom.example.yaml) lists\nevery setting with its default and what it does, by area: the ClickHouse connection,\ntimeouts and budgets, rate limits, workers, and the `*_enabled` switches that turn\nfeatures on and off. Set any of them as `LOOM_<NAME>` in `~/.loom/.env` and recreate the\nserver; a deployment that mounts a config file can set them in a copy of this file\ninstead ([how settings are read](https://github.com/ClickHouse/loom/blob/main/docs/operations/deployment.md#how-settings-are-read)). One marked\n`live` also changes while Loom runs: `loom settings set <name> <value>`, or Admin → Deployment\nsettings. `loom settings list --changed` shows what differs from the defaults and\nwhere each value came from ([runtime settings](https://github.com/ClickHouse/loom/blob/main/docs/operations/runtime-settings.md)).\n\nIf the installer connected your agent, skip the commands below. On the machine running a personal installation, the plugins need no token:\n\n```\n# Claude Code\nclaude plugin marketplace add ClickHouse/loom\nclaude plugin install loom@loom\n\n# Codex\ncodex plugin marketplace add ClickHouse/loom\ncodex plugin add loom@loom\n```\n\nStart a new session; the first one connects to `http://localhost:7777`. In Codex, run\n`codex` interactively once and approve the plugin's hooks; until then only the tools\nwork. Tell the agent something worth keeping, then ask about it in the next session.\n\nCursor lists Loom as a [plugin](https://cursor.com/marketplace/clickhouse/clickhouse-loom)\nand an [MCP server](https://cursor.com/marketplace/mcp/loom).\nFor a team marketplace on a Teams or Enterprise plan, a team admin can also import this\nrepository; its [Cursor plugin](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/cursor.md#plugin) installs\nthe MCP entry and Loom's skills. On the machine running a\npersonal installation, the button below adds the [local MCP\nentry](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/cursor.md#mcp-tools), which needs no token:\n\nFor VS Code agent mode on the machine running a personal installation, the buttons below\nadd the [local MCP entry](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/vscode.md#local-single-user-server),\nwhich needs no token:\n\nFor a team server, pass `--config server_url=… --config token=…` to Claude Code, or\npoint Codex at it as described in\n[connect a coding agent](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/connect-a-coding-agent.md). Create\ntokens in **Settings → Tokens**. Cursor and other MCP clients, and\n[the SDK](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/first-agent.md), are covered there too.\n\n- **Memory that carries forward.** Share decisions and findings across agents and\nsessions. Each fact records who wrote it, version history and conflict review make\ncorrections traceable, and a namespace can be rolled back to any point in time.[Memory](https://github.com/ClickHouse/loom/blob/main/docs/concepts/memory.md)\n- **Code intelligence with evidence.** Find definitions, callers, change impact and\nrelevant tests; connect code to commits, pull requests and issues. Use source\nlocations, completeness signals and per-tier freshness to check a result.[Code index](https://github.com/ClickHouse/loom/blob/main/docs/features/code-index/overview.md)\n- **Grounded retrieval.** Search permitted memories, documents, conversations and\ntracker records. Answers cite their sources; low-confidence retrieval can abstain.[Retrieval](https://github.com/ClickHouse/loom/blob/main/docs/concepts/retrieval-and-ranking.md)\n- **Multi-agent handoffs.** Assign work through durable task inboxes with leases,\nretries, review rules and outcomes. Your agents execute the workflow; Loom\npreserves the handoffs.[Task queues](https://github.com/ClickHouse/loom/blob/main/docs/features/tasks-and-decisions.md)\n- **Control over agent data.** Manage namespace access, provenance, retention, audit\nrecords and verified erasure. Context remains queryable in ClickHouse.[Security](https://github.com/ClickHouse/loom/blob/main/docs/architecture/security.md)\n\n| Integration | What it connects | \n|---|---|\n| [Claude Code](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/claude-code.md) ,[Codex](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/codex.md) ,[Cursor](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/cursor.md) and[other MCP clients](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/mcp-clients.md) ([all integrations](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/index.md) ) | Memory and code intelligence in your existing agent workflow. | \n| [Python](https://github.com/ClickHouse/loom/blob/main/docs/reference/python-sdk.md) ,[TypeScript](https://github.com/ClickHouse/loom/blob/main/docs/reference/typescript-sdk.md) ,[Rust](https://github.com/ClickHouse/loom/blob/main/sdk/rust/README.md) ,[Go](https://github.com/ClickHouse/loom/blob/main/sdk/go/README.md) ,[.NET](https://github.com/ClickHouse/loom/blob/main/sdk/dotnet/README.md) and[HTTP](https://github.com/ClickHouse/loom/blob/main/docs/reference/operations.md) | Your own agents and applications. | \n| [Langfuse](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/langfuse.md) | Request traces, model costs, retrieval scores and prompt-version attribution. | \n| [ClickStack / HyperDX](https://github.com/ClickHouse/loom/blob/main/docs/features/integrations/clickstack.md) | OpenTelemetry traces, logs and metrics in ClickHouse. | \n\nLangfuse and ClickStack are optional integrations, not bundled services in the Docker stack.\n\n| Evaluation | Result | \n|---|---|\n| LongMemEval-S evidence retrieval (500 questions) | **98.6%** of questions with at least one supporting session in the first 30 memories. | \n| LongMemEval-S fixed `gpt-5` reader (500 questions) | **92.0%** answer accuracy from the top 200 memories. | \n| LoCoMo (all 10 conversations) | **89.6%** answer accuracy;**90.8%** correct abstentions on adversarial questions. | \n| MemoryAgentBench fact-question sample (500 questions) | **99.0%** SubEM;**95.7%** token-overlap F1 on generated answers. | \n| Conceptual code search (36 questions) | Gold-file recall **91.2%** at 5 /**97.2%** at 10; MRR**0.91** . | \n| SWE-bench Verified file localization (200 issues) | Issues with a changed file in the first 10: **82%** vs**57%** for BM25;**76.5%** in the first 5. | \n| Tracker duplicate detection (104 pairs) | Recorded older duplicate ranked first **87.5%** of the time. | \n\nThese measure different tasks, not one aggregate score. SWE-bench measures\nfile localization, not issue resolution. The [results ledger](https://github.com/ClickHouse/loom/blob/main/docs/benchmarks/results.md)\nincludes scoring definitions, small-sample evaluations, weaker and negative findings,\nand the full protocols, models and artifacts.\n\n[Methodology](https://github.com/ClickHouse/loom/blob/main/docs/benchmarks/methodology.md) ·\n[coverage and known gaps](https://github.com/ClickHouse/loom/blob/main/docs/benchmarks/coverage.md) ·\n[run the benchmarks](https://github.com/ClickHouse/loom/blob/main/docs/benchmarks/running.md)\n\nFor shared agent context, code intelligence and permissioned retrieval, start with the\n[documentation](https://clickhouse.github.io/loom/). For deployment requirements, see the\n[prerequisites](https://github.com/ClickHouse/loom/blob/main/docs/getting-started/overview.md#prerequisites).\n\n[Deployment](https://github.com/ClickHouse/loom/blob/main/docs/operations/deployment.md) · [FAQ](https://github.com/ClickHouse/loom/blob/main/docs/faq.md) ·\n[Contributing](https://github.com/ClickHouse/loom/blob/main/CONTRIBUTING.md) · [Security reporting](https://github.com/ClickHouse/loom/blob/main/SECURITY.md) ·\n[`llms.txt`](https://clickhouse.github.io/loom/llms.txt) ·\n`llms-full.txt`", "url": "https://wpnews.pro/news/show-hn-loom-memory-that-follows-the-code-you-edit", "canonical_source": "https://github.com/ClickHouse/loom", "published_at": "2026-11-22 13:17:34+00:00", "updated_at": "2026-11-22 13:42:06.099547+00:00", "lang": "en", "topics": ["ai-agents", "agent-protocols", "ai-tools", "developer-tools"], "entities": ["ClickHouse", "Loom", "Claude Code", "Codex", "Cursor", "Postgres"], "also_reported_by": [], "takeaway": "Loom is a self-hosted MCP server that shares long-term memory across coding agents, links memories to the code they describe so they follow refactors, and adds a code index plus cited document search, with ClickHouse storing context and index data and Postgres optional at first boot.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/show-hn-loom-memory-that-follows-the-code-you-edit", "markdown": "https://wpnews.pro/news/show-hn-loom-memory-that-follows-the-code-you-edit.md", "text": "https://wpnews.pro/news/show-hn-loom-memory-that-follows-the-code-you-edit.txt", "jsonld": "https://wpnews.pro/news/show-hn-loom-memory-that-follows-the-code-you-edit.jsonld"}}