Local, source-backed memory for LLM agents.

Link gives Codex, Claude, Cursor, Kiro, VS Code, Copilot, and other MCP clients the same durable memory about you and your work. It stays on your machine as plain Markdown, with sources, backlinks, graph context, review state, and an audit trail you can inspect.

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

Product site · First 10 minutes · Why Link? · Web UI · MCP setup · CLI · Security · Changelog

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.

Run the demo first. It creates a complete local wiki with raw sources, wiki pages, one starter memory, graph data, and query packets ready to inspect.

macOS with Homebrew:

brew install gowtham0992/link/link
link demo
link serve link-demo

Or from source:

git clone https://github.com/gowtham0992/link.git
cd link
python3 link.py demo
python3 link.py serve link-demo

Open:

http://127.0.0.1:3000
http://127.0.0.1:3000/graph

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:

link query "why does Link help agents?" link-demo --budget small
link brief "working on agent memory" link-demo
link benchmark "agent memory" link-demo
link status --validate link-demo

From a source checkout, use python3 link.py ...:

python3 link.py query "why does Link help agents?" link-demo --budget small
python3 link.py brief "working on agent memory" link-demo
python3 link.py benchmark "agent memory" link-demo
python3 link.py status --validate link-demo

The generated demo is the public proof wiki. The repo's root wiki/ directory is only a scaffold for local development and personal testing. Generated content inside wiki/, raw/, and link-demo/ is ignored by git so personal memory is not published by accident.

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

Web UI Review memory, ingest, graph, audits, captures, and explanations in a local viewer.

CLI Script readiness, query packets, briefs, validation, backup, benchmark, and repair.

MCP Let Codex, Claude, Cursor, Kiro, VS Code, Copilot, and other agents recall memory.

Run one installer from the cloned checkout:

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

Then ask your agent:

is Link ready?
brief me from Link before we continue
ingest raw/notes.md into Link
remember that I prefer short release notes
query Link for the release process
what does Link remember about local personal memory?

python3 -m pip install --upgrade link-mcp

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

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

Link has one simple rule:

Sources become wiki knowledge.
Explicit "remember" becomes agent memory.
Queries use both.

The storage model is plain and inspectable:

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

• query_link: an answer-ready packet with relevant memories, pages, graph neighborhood, reasons for selection, budget limits, and follow-up actions.
• memory_brief: a compact pre-work brief with user/project preferences, active context, review warnings, and safe memory-use rules.
• ingest_status: exact next steps for raw files, including source safety, stale ingest detection, validation, and memory proposal guidance.
• remember_memory: durable local memory with duplicate/conflict checks, review state, provenance, and audit logging.
• explain_memory: why a memory exists, what it links to, whether it is ready for recall, and what needs review.

Agents should use Link in this order:

1. link_status to check readiness and safe next actions.
2. starter_prompts when the user asks what to try first.
3. ingest_status before touching raw sources.
4. query_link for compact answer-ready context.
5. memory_brief before longer work.
6. get_graph_summary when graph context is useful but the full graph would be noisy.
7. backup_wiki before broad repair or migration work.
8. validate_wiki after ingest or broad wiki edits.

Full MCP tool list: MCP setup.

Link is local-first:

• No telemetry.
• 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.
• link backup excludes raw/ unless you explicitly pass --include-raw.
• Secret-looking values are detected in raw sources, captures, and release hygiene checks.
• 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 doctor
python3 link.py validate
python3 scripts/check_release_hygiene.py

More detail: Security guide.

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 unittest discover -s tests
python3 scripts/check_release_hygiene.py
python3 scripts/check_runtime_duplication.py
python3 scripts/check_tool_contract.py
git 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.