Link

[Link]

Link

Local memory for AI agents.

Link gives Codex, Claude, Cursor, Kiro, VS Code, Copilot, Antigravity, and other local agents the same source-backed memory, stored locally as Markdown.

Website · How it works · Memory · Tools & CLI · Setup · Docs · MCP Registry · PyPI · Homebrew

[GitHub stars] [CI] [MCP Registry] [PyPI] [PyPI Downloads] [MCP Toplist rank]

[Link demo: a preference said once in an agent session is captured automatically, approved by you, and recalled in a brand-new terminal the next day — from a plain Markdown file]

Quick start

brew install gowtham0992/link/link   # or: pip install link-mcplnk setup                            # wires every agent you have

Then work normally. Tell an agent "remember that we deploy on Tuesdays", approve it once, and every other agent recalls it in a new session tomorrow.

lnk status                                   # is it wired up and being used?lnk import claude-code                       # bring existing memory inbrew install --cask gowtham0992/link/linkbar # macOS menu bar app

Memory lives in plain Markdown you can open, and nothing becomes durable without your approval.

What Is Link?

Link is an open-source memory layer for local AI agents. Raw sources become an inspectable Markdown wiki. Explicit "remember this" requests become reviewable memories. Agents retrieve compact, source-backed context through the CLI, MCP, official skills, or the local viewer without dumping the whole wiki into a chat window.

The wiki is the storage layer. The product is durable memory that stays on your machine, remains readable in plain files, and can be shared across multiple agents instead of locked inside one vendor profile.

[lnk recall finds a memory saved in completely different words — matched by meaning, not keywords]

Ask in your own words; Link matches by meaning, not keywords. All local, all plain files.

How It Works

Link gives agents five simple moves:

1. Capture notes, transcripts, docs, screenshots, and project context in raw/.
2. Structure source-backed pages under wiki/.
3. Remember explicit preferences, decisions, facts, and project context as reviewable memory.
4. Retrieve compact query packets through the CLI, MCP, official skills, or the local web viewer.
5. Hold what you approved: rules are enforced before a tool call, memories about code are checked against the repository, and a receipt shows what each session was given.

Most agent sessions start from zero. You re-explain preferences, repo decisions, project constraints, and why something matters. Link turns that repeated context into local memory agents can query.

Link follows Andrej Karpathy's LLM Wiki pattern: keep knowledge outside the chat window, make claims inspectable, and let context compound over time.

Why Link Is Different

Every other agent-memory system stores memory as embeddings in a vector database or as an LLM-extracted graph. Link made five architectural commitments those designs cannot bolt on:

1. Memory you can read. Every memory is a plain Markdown file — open it, grep it, git-diff it. If Link disappeared tomorrow, your memory is still yours.
2. Review-gated writes. Agents propose; you approve. Even the automatic session hooks capture proposals, never facts.
3. No LLM in the memory layer. Ingestion and recall are deterministic — nothing can hallucinate a fact into your memory, because there is no model in the write path.
4. Provably local. CI blocks outbound network code in the runtime, and the optional semantic models load offline-only after one explicit setup.
5. Memory that holds when the agent acts. A memory you reviewed can carry a rule (ask command: git push --force, deny write: migrations/**) that is checked before the agent's tool call, with no model and under a millisecond. Memories about code are re-read against your repository and flagged when the code moved on. Memories surface when the agent opens the file they describe, lnk compile writes reviewed memory into the files every agent already reads, and a receipt proves what reached each session.

And the claims are measured, not asserted — see the benchmarks below. Named comparisons against Mem0/OpenMemory, Zep/Graphiti, and Letta: Why Link?

Benchmarks

Plain files with no LLM in the memory layer, measured against the systems that have one everywhere:

Every number ships with its config, judge model, caveats, and the experiments that lost — including LongMemEval end-to-end, where we re-judged both sides under the neutral Hunyuan 3 referee: mem0's GPT-5 answers score 91.0%, Link's budget-model answers 80.6%. Their published number holds up, and the gap tracks the answering model, not the memory layer — that's what the 99.4% evidence-retrieval row above isolates. Full methodology and reproduction steps: benchmarks/RESULTS.md.

Memory You Can Take Anywhere — And Prove Is Working

Three questions every memory system should answer, and how Link answers them with mechanisms instead of promises:

"Is it on all my machines?" lnk sync moves reviewed memory through a git remote you control — a private GitHub repo, a homelab bare repo — with no server and no account. Secrets are scanned before anything leaves, conflicts become review items instead of git markers, and private captures never sync. lnk team-sync runs a shared team brain on the same rails.

"What did I believe back then?" Ask in plain language: lnk recall "where does local data live in March" returns what was true then, reconstructed from the dated files and their supersede lineage — while the same question without the date returns today's truth. Deterministic: a regex and a calendar, no model. Point-in-time accuracy from plain language: 1.00 on the hygiene benchmark's revised facts, identical to asking with an ISO date.

"Is my agent actually using this?" The question most memory systems cannot answer about themselves. Link records retrievals locally — session briefs pushed to agents, recalls they chose to make — and lnk wins answers with counts, while lnk digest reports the weekly story including memories that have never been retrieved. The ledger stores which memory was read and when, never what you asked; it never syncs, and LINK_USAGE=off disables it. And memory reaches every agent, not just the hooked ones: the first MCP tool response of a session carries the brief, whatever tool was called.

First Run, In Detail

The quick start above is the whole install. This is what each step does.

brew install gowtham0992/link/linklnk proof     # see the promise (~1 second, no setup)lnk setup     # wire every agent you have — workspace, MCP, session hooks, one command

Already have memory scattered across tools? Bring it home as reviewable proposals — nothing is auto-accepted:

lnk import claude-code   # or: cursor, codex, file --file chatgpt.txt

lnk setup detects the agents installed on your machine — Claude Code, Codex, Cursor, Windsurf, Zed, Kiro, Gemini CLI — and wires them all at once; Link is agent-agnostic by design (one MCP server, one workspace, every agent reads the same memory). It is idempotent: after any brew upgrade, run it again and everything — workspace runtime, MCP provisioning, hooks — is refreshed. Prefer to wire one agent explicitly? lnk onboard --agent claude-code --write --hooks still does exactly that.

lnk proof creates a throwaway workspace, writes one reviewed memory, and recalls it through the same path the CLI, skills, and MCP use — the core promise (one local memory, reusable by different agents, no cloud profile) in one second:

Cross-agent memory continuity worksMemory: created and reviewed: Cross-agent Link proofRecall: found through the same bounded recall path used by CLI, skills, and MCP.Result: proof passed

lnk onboard --agent claude-code --write --hooks creates ~/link, provisions the MCP runtime, and wires the agent. Swap claude-code for codex, cursor, kiro, copilot, antigravity, windsurf, zed or vscode. Session hooks - a brief at session start, proposals captured at session end - exist for Claude Code, Cursor and Codex (Codex has no session-end event, so it gets the start brief only). The other agents get memory through MCP, where the first tool response of a session carries the same brief. Drop --write to preview without touching anything, or drop --agent to just create the workspace. lnk disconnect <agent> --write takes Link back out: its MCP entry and hooks, nothing else.

The installed command is lnk because link is already a POSIX/macOS system utility. From a source checkout, use python3 link.py ... instead.

Want the UI, graph, and source pages first? lnk try && lnk serve link-demo. Windows, source checkout, MCP-only, and skill-first paths are in the First 10 Minutes guide.

Or seed your current repo as a separate step so the first real recall is not empty:

cd /path/to/your/projectlnk seed . ~/linklnk query "what is this project about?" ~/link --budget small

lnk seed reads allowlisted project files such as README.md, AGENTS.md, CLAUDE.md, .cursorrules, and editor rule files, blocks secret-looking values, writes a source-backed project page, and rebuilds the graph. It does not create durable memories; agents should still use reviewed memory proposals for preferences and decisions.

The Homebrew formula is maintained in the public gowtham0992/homebrew-link tap.

Open:

http://127.0.0.1:3000http://127.0.0.1:3000/onboardhttp://127.0.0.1:3000/graphhttp://127.0.0.1:3000/health

Use /onboard when you want the same first-run checklist in the local UI: readiness, project context seeding, first memory, agent wiring, and starter prompts. The web viewer is for local use only. It binds to 127.0.0.1, has no user accounts or authentication, and should not be exposed to the internet unless you add your own auth layer.

Try the value loop:

lnk start link-demo --task "working on agent memory"lnk query "why does Link help agents?" link-demo --budget smalllnk brief "working on agent memory" link-demolnk benchmark "agent memory" link-demolnk health link-demo

lnk benchmark reports both performance and value evidence: cache/search/query timings, graph payload shape, and an estimate of how much broad wiki context the bounded Link packet avoided sending to an agent.

The /health page mirrors the readiness loop in the browser: validation state, interrupted writes, memory review status, and copyable repair commands. The viewer stays document-first — common paths in the top nav, deeper tools under more, and a contents outline plus graph-related links on structured pages.

The generated demo is the public proof wiki. Generated content inside wiki/, raw/, and link-demo/ is ignored by git so personal memory is not published by accident.

LinkBar — the menu bar app (macOS)

Link's memory, ambient. LinkBar puts the review gate in your menu bar: a global palette (⌥⌘M) to recall or remember from any app, native notifications with one-tap Accept when a session capture lands, a live pulse while agents are writing, a browser over every memory file, and a status dashboard that flags memories naming files your repository no longer has — all running on the same reviewed lnk commands as the CLI. Pick the workspace in Settings; it defaults to ~/link.

[LinkBar cycling through its tabs: the review inbox showing the rule a memory will enforce before you approve it and the last session's receipt, the memory browser, and settings]

brew install --cask gowtham0992/link/linkbar

Unsigned on purpose (no Apple fee inflating anything): the cask strips the quarantine flag on install, so it opens like any app. Building from source instead: cd apps/LinkBar && bash Scripts/bundle.sh --install.

Killer Demo: One Memory, Two Agents

This is the moment Link is built for:

1. In one agent, say:

   text

   remember that I prefer local, source-backed memory for AI agents

2. In another agent connected to the same ~/link workspace, say:

   text

   start with Link before we continuewhat does Link remember about local agent memory?

3. The second agent should recall the reviewed memory from local Markdown instead of asking you to repeat yourself.

For a clean automated version of the same idea, run:

lnk proof

Ways To Use Link

Pick the surface that matches how you work. They all read and write the same local Markdown wiki.

These surfaces are independent. lnk serve / serve.py is only the local web viewer. CLI commands, official skills, and MCP tools read the same wiki/ files directly, so Claude, Codex, Kiro, Cursor, or another agent can use Link even when the web viewer is not running.

[Link local console tour: home, memory dashboard, health, and graph]

The local web viewer: browse source-backed memory and explore the knowledge graph — all on 127.0.0.1, no accounts, no backend.

Prefer skills instead of MCP? Link ships small, lazy-loadable CLI skills under skills/. They let an agent use lnk health, lnk query, lnk ingest-status, lnk session-end, and lnk remember directly, without MCP setup or a running web viewer.

skills/link-health/SKILL.mdskills/link-retrieve/SKILL.mdskills/link-ingest/SKILL.mdskills/link-memory/SKILL.md

Full guide: Link Skills.

Install For Your Agent

Run one installer from the cloned checkout:

bash integrations/codex/install.shbash integrations/kiro/install.shbash integrations/claude-code/install.shbash integrations/cursor/install.shbash integrations/copilot/install.shbash integrations/vscode/install.shbash integrations/antigravity/install.sh

Installers create or update ~/link, install or upgrade link-mcp, write lightweight agent instructions, and preserve existing wiki data on reinstall. Use --project when a repo needs separate project memory.

On Windows, use the matching PowerShell installer:

.\integrations\codex\install.ps1.\integrations\kiro\install.ps1.\integrations\claude-code\install.ps1.\integrations\cursor\install.ps1.\integrations\copilot\install.ps1.\integrations\vscode\install.ps1.\integrations\antigravity\install.ps1

Then ask your agent:

is Link ready?start with Link before we continueseed this project into Linkingest raw/notes.md into Linkremember that I prefer short release notesquery Link for the release processwhat does Link remember about local personal memory?end this session with Link memory proposals

For CLI-first agents or Link skills, use the same startup loop directly:

lnk seed . ~/linklnk start ~/link --task "working on Link release"lnk session-end session-notes.md ~/link --limit 3

If you want one guided setup for a real workspace and an agent, use lnk onboard --agent AGENT. If your agent already has instructions and you only need MCP wiring, use the lower-level connection helper. Both preview the exact config first; add --write when you want Link to update the agent config file.

lnk onboard --agent codexlnk onboard --agent codex --writelnk connect codex ~/linklnk connect codex ~/link --writelnk connect kiro ~/link --writelnk verify-mcp ~/link

For agents with session-hook support — Claude Code, Codex, and Cursor — add --hooks (works with lnk onboard too) to make the memory loop automatic: the brief is injected at session start and proposal-only notes are captured at session end, so memory no longer depends on the agent remembering to call Link. Empty sessions and duplicate end events are skipped, and when the backlog builds up the brief nudges the agent to offer a read-only lnk consolidate pass. On Claude Code the hooks also run before shell and file tool calls: a reviewed memory's enforced rules are checked there, and memories anchored to the file being read or edited are surfaced at that moment; on Cursor the shell-command check runs the same way. lnk receipt shows what actually arrived in each session. Durable memory still requires your approval. Codex and Cursor hook support is new (wired to their documented schemas — report issues).

lnk connect claude-code ~/link --hooks --writelnk connect codex ~/link --hooks --write    # session-start brief (Codex has no session-end event)lnk connect cursor ~/link --hooks --writelnk consolidate ~/link                      # read-only backlog plan, apply only with approval

Optional: hybrid semantic recall (still fully local)

Lexical recall is always the default and the fallback. Paraphrase matching is opt-in: after the two setup commands below, "how should I structure my pull requests" finds a memory saved about commit style. Until then, recall matches on shared words, and a miss tells you how to turn paraphrase matching on. Installing the optional semantic extra adds a small local static-embedding model. Recall never touches the network: the model loads offline-only after a one-time explicit setup, embeddings live in plain JSON under .link-cache/, similarity runs in-process with no vector database, and semantic-only matches carry capped confidence labels so agents verify before trusting them.

pip install "link-mcp[semantic]"          # fast tier: tiny static model, instant loadpip install "link-mcp[semantic-quality]"  # quality tier: contextual model, best recalllnk semantic ~/link --setup   # one-time model fetch, with your approvallnk semantic ~/link           # status: lexical only vs hybrid, active tierpython3 -m link_mcp --semantic-setup --wiki ~/link/wiki   # MCP-only installs

Measured, not asserted: on the bundled 1,176-case benchmark, the quality tier lifts token-overlap hit@1 from 0.671 to 0.789 and finds half of the pure paraphrases (zero token overlap) in the top 5, where lexical recall finds 6%, at ~11 ms per recall with no service or vector database. On the third-party LoCoMo retrieval track (1,536 evidence-annotated questions over 5,882 conversation turns), hybrid recall lifts any-evidence hit@10 from 0.716 to 0.779, and the opt-in rerank tier to 0.826 with hit@1 0.453. Without the reranker, lexical recall is the stronger top-5 on LoCoMo; RESULTS.md says why. Full methodology, honest limitations, and reproduction steps: benchmarks/RESULTS.md.

python3 -m pip install --upgrade link-mcppython3 -m link_mcp --version

{  "mcpServers": {    "link": {      "command": "python3",      "args": ["-m", "link_mcp", "--wiki", "~/link/wiki", "--surface", "slim"]    }  }}

python3 -m venv ~/.link-mcp-venv~/.link-mcp-venv/bin/python -m pip install --upgrade pip link-mcp

Obsidian users can import an existing vault into raw/ for agent ingest, or open ~/link/wiki directly as a vault for editing Link pages:

lnk init ~/linklnk import-obsidian ~/Documents/ObsidianVault ~/link

See the Obsidian guide for the import, edit, and validation loop.

Storage Model

Under the hood, Link separates source-backed knowledge from durable agent memory:

1. Drop raw notes, transcripts, articles, and project context into raw/.
2. Agents compile those sources into inspectable pages under wiki/.
3. Explicit "remember" requests become reviewable memory pages.
4. Queries retrieve compact agent context from both the wiki and memory layer.

[Link architecture: raw sources become wiki knowledge, explicit remembers become reviewed memory, and agents retrieve compact context]

The storage model is plain and inspectable:

If a raw file was already ingested and later edited, lnk ingest-status marks it as stale and tells your agent to refresh the existing source page instead of creating a duplicate.

Structured exports with a supported schema can use deterministic ingestion. Planning is read-only; applying stages the complete result, rebuilds the graph, validates it, and records output ownership before changing the live wiki:

lnk ingest raw/chezmoi-docs/export.jsonl ~/link \  --adapter chezmoi-docs-graph-v1 \  --exclude "Reference / Release history"
lnk ingest raw/chezmoi-docs/export.jsonl ~/link \  --adapter chezmoi-docs-graph-v1 \  --exclude "Reference / Release history" \  --apply

Later runs update only adapter-owned outputs. Manual edits become conflicts. --replace-unmanaged is required for a first migration over existing pages, and --prune is required to retire managed pages no longer in the plan. Arbitrary notes and articles continue through agent-authored ingest because their useful structure cannot be inferred safely from a file format alone.

What Agents Get

When an agent uses Link through the recommended MCP surface, it gets six model-facing tools. CLI and skill workflows call the same core behavior through lnk.

• status: readiness, schema state, validation, interrupted writes, and safe next actions.
• recall: the one read path for startup briefs, answer-ready query packets, wiki search, graph context, token budgets, and follow-up actions. Every recalled memory carries a confidence label (strong, moderate, weak) and a match field (lexical, semantic, hybrid when the optional local semantic tier is installed), so agents verify weak or paraphrase matches with the user instead of trusting them.
• remember: durable local memory only after explicit user approval, with duplicate/conflict checks, provenance, review state, visibility, optional enforce rules that act on tool calls once you review the memory, optional review_after, and optional expires_at.
• ingest: exact next steps for raw files, source safety, stale ingest detection, validation, and rebuild checks.
• review: memory inbox, profile, audit, log, explain, archive, restore, forget, and lifecycle review workflows — plus review(action="consolidate"), a read-only backlog plan applied only with per-action user approval.
• admin: the escape hatch for backup, migrate, validate, graph export, pages, captures, rebuilds, compatibility actions, and advanced updates.

The stable agent-facing loop is documented at Link Memory Contract: readiness first, bounded recall, explicit memory writes, audit tools, and sharing semantics.

Use review_after for time-sensitive preferences or decisions. When that date arrives, the memory reappears in Link's review inbox so an agent can ask the user to confirm, update, archive, or forget it instead of trusting stale context. Use expires_at for temporary context that should automatically leave default recall after a date; Link keeps the Markdown page inspectable and asks the user to update, archive, or delete it. Use visibility to separate where a memory applies from who should see it: private stays personal, project is intended for a project workspace, and team means the user explicitly approved sharing it with a team.

For team handoff or security review, lnk compliance-export --output audit.json writes a redacted JSON packet with readiness, validation, memory review status, operation markers, and recent audit log entries. Raw source contents and memory bodies are not included.

For day-to-day auditability, lnk memory-log ~/link shows what Link recently remembered, updated, reviewed, archived, restored, forgot, or accepted from raw captures.

For recovery, lnk backup ~/link creates a local archive and lnk restore-backup <archive> ~/link previews what would be restored. Passing --confirm replaces local files after creating a safety backup when possible; raw/ is still excluded unless --include-raw is explicit. If a multi-file write is interrupted, lnk operations ~/link shows the marker and any rollback snapshot; lnk operations ~/link --recover <marker> --confirm restores the snapshot after you review it.

For local proof of value, lnk wins ~/link shows reusable memories, reviewed memory, provenance, project continuity, freshness guardrails, and copyable prompts without tracking user behavior.

For Git-backed team memory, lnk team-sync ~/link checks whether the workspace is ready to share reviewed wiki/ pages while keeping raw/, caches, backups, local MCP Python markers, and wiki/log.md private by default. The audit log is local because it has a single-machine hash chain; merging multiple users' logs would create false tamper alarms. Team sync also blocks "ready" status when the memory inbox is not clear or active visibility: private memories would be included by a broad git add wiki.

lnk team-sync ~/link --remote [email protected]:team/link-memory.git

For a teammate, reviewer, or another agent, lnk share resolves a page, memory, title, alias, or search phrase into a local viewer URL:

lnk share "Prefer local memory" ~/link

For a static, read-only review packet, lnk snapshot exports rendered wiki HTML without raw/, captures, operation markers, live MCP state, or memory pages by default. --include-memories exports only non-private memories; use --include-private-memories only for a personal archive or an explicitly approved review. It blocks export if wiki pages contain secret-looking values unless you explicitly override it.

lnk snapshot ~/link --output link-snapshotlnk snapshot ~/link --output link-snapshot --include-memories --forcelnk snapshot ~/link --output personal-snapshot --include-memories --include-private-memories --force

Agent Contract

For MCP clients, agents should use Link in this order:

1. status to check readiness and safe next actions.
2. recall with an empty query once at the first substantive turn of a session.
3. recall(query, budget="micro"|"small") before broad file reads or asking the user to repeat durable context.
4. ingest before touching raw sources and after source edits for validation/rebuild checks.
5. remember only when the user explicitly asks Link to remember something or approves a proposed memory.
6. review for memory inbox, profile, audit, log, explain, archive, restore, and forget workflows.
7. admin for backup, migration, graph export, captures, rebuilds, compatibility actions, and advanced maintenance.

Full MCP tool list: MCP setup.

Privacy And Safety

Link itself is local-first:

• No telemetry in the installed CLI, MCP server, local web UI, or wiki runtime.
• No hosted backend.
• No external API calls from serve.py or link-mcp.
• Raw sources and generated wiki pages are ignored by git by default.
• lnk backup excludes raw/ unless you explicitly pass --include-raw.
• Secret-looking API keys, provider tokens, JWTs, registry credentials, and private key blocks are detected in raw sources, captures, and release hygiene checks. lnk validate and lnk doctor also fail if secret-looking values are found inside wiki pages before they can be served through the local UI or returned through agent context.
• Optional semantic recall stays local: models load offline-only at recall time (only the explicit lnk semantic --setup may fetch a model, once), and embeddings live in plain JSON under .link-cache/.
• Automatic session hooks store proposal-only notes; transcript extraction skips tool calls and outputs, and no durable memory is written without review.
• The local web server binds to 127.0.0.1 and is not meant to be exposed to the internet without additional auth.

Before sharing a repo, demo, or wiki:

python3 link.py doctorpython3 link.py validatepython3 scripts/check_release_hygiene.py

More detail: Security guide.

Documentation

Contributing

Contributions should come through pull requests targeting main. The develop branch is a maintainer integration branch for larger release work before it is proposed to main.

Before opening a PR:

python3 -m ruff check .python3 -m pytest testspython3 scripts/check_release_hygiene.pypython3 scripts/check_runtime_duplication.pypython3 scripts/check_tool_contract.pygit diff --check

Full contributor guide: Contributing.

Do not include personal wiki data, raw sources, registry tokens, .env files, or local MCP credentials in a PR.

If Link helps your agents remember better, star it on GitHub so more people can find it.