Portable memory for AI agents.

Open-source, runs locally, and keeps your agent memory on your machine. Searchable, correctable, cross-tool memory with a real dashboard.

Install local Star on GitHub

{
  "mcpServers": {
    "lodis": {
      "command": "npx",
      "args": ["-y", "@sunriselabs/lodis"]
    }
  }
}

Local MCP over stdio. Setup guide includes Codex, Claude Code, Cursor, Windsurf, and other MCP clients.

Lodis dashboard showing memory browser with search, domain filters, and confidence scores

AI memory today is broken.

Today

With Lodis

Today

Your AI builds up knowledge about you over time. You can’t see what it stored.

With Lodis

Browse, search, and edit every memory in a real dashboard.

Today

Teach something in one tool, start over in the next. Context trapped in walled gardens.

With Lodis

One memory layer shared across all your AI tools.

Today

Corrections don’t stick. You can’t trace where it learned something or how sure it was.

With Lodis

Confirm, correct, or flag — with confidence scores and full provenance.

Lodis makes AI memory yours.

Search & Retrieve

Hybrid search finds the right memory even when the wording differs. Context-packed retrieval delivers token-budget-aware results. Filter by domain, confidence, entity type, or just ask.

Search results showing memories with confidence scores and entity types

Correct & Control

Confirm what's right. Correct what's wrong. Split compound memories. Flag mistakes. Your AI learns from your feedback.

"Prefers TypeScript with strict mode enabled"

0.92

ConfirmCorrectSplit

Connect & Understand

Memories support 14 entity types. Agents can classify people, projects, preferences, and snippets, while Lodis handles deterministic links, connection proposals, cached entity profiles, and contradiction review workflows.

person

Sarah Chen

Engineering lead at Acme

0.95

project

Project Atlas

Next-gen search platform

0.88

preference

TypeScript strict mode

Always enabled in all projects

0.99

39 MCP tools. One install.

Everything your agent needs to remember, learn, and forget.

memory_write

Create a memory with dedup detection and permanence tiers

memory_write({ content: "Sarah prefers morning meetings", domain: "work" })

memory_search

Hybrid semantic + keyword search with filters

memory_search({ query: "design decisions", minConfidence: 0.8 })

memory_correct

Store corrected or user-asserted information

memory_correct({ id, correction: "It's Thursdays, not Tuesdays" })

memory_connect

Create typed relationship between memories

memory_connect({ from, to, relationship: "works_at" })

memory_onboard

Guided onboarding: scan tools, interview, seed

memory_onboard({ scan: ["claude", "cursor", "gitconfig"] })

memory_export

Export memories as portable JSON

memory_export({ format: "json" }) → lodis-backup.json

See all 39 tools

Core

memory_writeCreate a memory with dedup detection and permanence tiers

memory_searchHybrid semantic + keyword search with filters

memory_getFetch one or many memories by ID (up to 50)

memory_contextToken-budget-aware context retrieval

memory_rate_contextClose the feedback loop on a prior context retrieval

memory_updateStructural edit — content, detail, domain, or entity fields

memory_removeSoft-delete with reason tracking

memory_remove_bulkBulk soft-delete scoped by domain, entity name, or IDs

memory_bulk_uploadBatch-insert up to 5,000 memories for external imports

Trust

memory_confirmMark a memory as verified and boost its confidence

memory_correctStore corrected or user-asserted information

memory_flag_mistakeDegrade confidence on bad information

memory_pinPin as canonical — decay-immune, high confidence

Graph

memory_connectCreate typed relationship between memories

memory_connect_batchCommit multiple relationship edges in one call

memory_get_connectionsTraverse the knowledge graph

memory_propose_connectionsServer-side candidate selection for the connection loop

memory_splitBreak compound memories into atomic units

Discovery

memory_listBrowse by domain, confidence, or recency

memory_list_domainsList all domains with memory counts

memory_list_entitiesShow extracted entities by type

memory_classifyList untyped memories for agent-side classification

memory_briefingCached entity profiles or evidence for synthesis

Safety

memory_scrubDetect and redact PII from memory content

memory_set_permissionsPer-agent read/write access by domain

memory_archiveArchive for reference — deprioritize and freeze

Progress

memory_write_snippetWrite a validated progress event (shipped/started/stalled/blocked)

memory_query_progressTime-ranged snippet query by domain or goal

memory_progress_summaryRollup: totals by domain, type, and goal for weekly reviews

memory_register_domainRegister a life domain slug for snippet writes

memory_archive_domainArchive a domain so snippet writes are rejected

Onboarding

memory_onboardGuided onboarding: scan tools, interview, seed

memory_importImport from Claude, ChatGPT, Cursor, gitconfig

memory_interviewAgent-driven cleanup and gap-fill

memory_tutorialInteractive chapter-by-chapter tutorial for how Lodis works

Data

memory_exportExport memories as portable JSON

memory_indexIndex external docs for unified search

memory_index_statusCheck staleness of indexed documents

memory_migrateLegacy migration helper for existing deployments

Local-first. No account required.

Default path

AI client -> stdio MCP -> ~/.lodis/lodis.db

Your agent talks to Lodis locally. Search, embeddings, rerank, and the dashboard run on your machine.

SQLite + FTS5 + native vector search in ~/.lodis/lodis.db.

Embeddings run locally via Transformers.js. No API calls for search.

0600 file permissions. OS-level access control.

JSON export. Your data is always portable.

Local dashboard at localhost:3838 for review, correction, and export.

Four permanence tiers: canonical, active, ephemeral (TTL), and archived.

Run it locally. Keep it portable.

Lodis launches as a local MCP process and stores memory in SQLite on your machine.

Local

Recommended · Open source · MIT License

SQLite on your machine - zero infrastructure
Stdio MCP transport - works with any local MCP client
No accounts or API keys required
In-process reranker - no external inference service
Dashboard at localhost:3838

Your data never leaves your machine.

Install local

Your data stays in ~/.lodis/lodis.db and exports as portable JSON.

Get started in 30 seconds.

Paste this JSON into most local MCP clients. Codex users can use the command listed below.

{
  "mcpServers": {
    "lodis": {
      "command": "npx",
      "args": ["-y", "@sunriselabs/lodis"]
    }
  }
}

Codex: codex mcp add lodis -- npx -y @sunriselabs/lodis
Claude Code: ~/.claude.json
Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json
Cursor: ~/.cursor/mcp.json
Windsurf: ~/.codeium/windsurf/mcp_config.json