Swarmvault
SwarmVault is a local-first knowledge compiler that turns your files into a provenance-tracked wiki and knowledge graph you actually own.
SwarmVault is the free, local-first option we'd actually install for a knowledge base that outlives the current model vendor — provenance on every claim, schema-guided compiles, and lint that catches contradictions before they calcify. The setup tax is real: you write a schema, learn a dozen subcommands, and manage the graph yourself. Bring your own OpenAI, Anthropic, Gemini, or Ollama key only when you want richer synthesis; the heuristic workflow runs offline with no keys at all.
Verified 4d ago · liveness 60/100 · cite: rightaichoice.com/tools/swarmvault
- Developers who want a local-first knowledge graph with provenance on every claim
- Researchers compiling arXiv, DOI, and article URLs into a durable, queryable wiki
- Knowledge workers who want a private RAG knowledge base that never leaves their machine
- Claude Code, Cursor, or Codex users who need persistent agent memory and a task ledger over MCP
- Non-technical users who want a GUI-only tool with no command-line or schema work
- Teams that need real-time collaboration or managed cloud sync out of the box
- Anyone who just wants a chat box over a handful of PDFs
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip SwarmVault if you want a GUI-only tool with no command-line steps or just a chat box over a handful of PDFs — the compounding value here starts only after you write a swarmvault.schema.md and learn the CLI.
The tool itself is MIT-licensed, but if you add OpenAI, Anthropic, Gemini, or another provider for richer synthesis, that usage is billed by that provider, not by SwarmVault.
SwarmVault is published as an MIT-licensed open-source project on npm (@swarmvaultai/cli), so the software cost is $0 and the real spend is the model provider you choose to bring — or nothing at all if you stay on the built-in offline heuristic workflow. That puts it in a different cost category from hosted chat-with-docs and cloud RAG subscriptions where you pay per seat or per document every month, and it suits individual developers, researchers, and small teams running long-horizon knowledge
In short
Swarmvault — SwarmVault is a local-first knowledge compiler that turns your files into a provenance-tracked wiki and knowledge graph you actually own. Best for Developers who want a local-first knowledge graph with provenance on every claim, Researchers compiling arXiv, DOI, and article URLs into a durable, queryable wiki, Knowledge workers who want a private RAG knowledge base that never leaves their machine. Free to use.
What's new in Swarmvault
Checked 4 days agoAcross the latest 2 updates: 2 feature updates.
Agent context packs and task ledger added
Introduced swarmvault context build for bounded context packs and swarmvault task, a ledger recording agent task goals, decisions, graph evidence, changed paths, and outcomes.
Agent-ready auto-install rules for Claude, Cursor, Codex
Auto-install rules for Claude Code, Cursor, and Codex let you expose the vault directly through MCP to compatible agents.
Viability Score
How well maintained and how widely used is Swarmvault? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Local-first knowledge compiler: raw sources, Markdown wiki, and graph state as inspectable local files
- Schema-guided compilation via swarmvault.schema.md for naming rules, categories, and grounding expectations
- Four-stage workflow: Shape, Ingest, Compile, Compound
- Immutable content-hashed source manifests for every ingested file or URL
- Knowledge graph with provenance — every edge traces to a specific source and claim
- God nodes and communities with plain-English graph report pages
- Anti-drift linting flags stale sources, conflicting claims, missing citations, and coverage gaps
- Deep lint attaches external evidence to flagged claims
- CLI covers init, quickstart, ingest, add, inbox import, compile, review, query, chat, retrieval, explore, lint, watch,
- Query and chat output formats: markdown, report, slides, chart, image
- Agent context packs via swarmvault context build, saved to state/context-packs/
- Agent task ledger via swarmvault task start|update|finish|resume
- Auto-install rules put the vault in front of Claude Code, Cursor, and Codex via MCP
- Heuristic offline workflow runs with no API keys; model providers optional for richer synthesis
- Desktop viewer reads the same graph.json, report.json, approvals, and candidates the CLI writes
About Swarmvault
SwarmVault is an open-source, MIT-licensed knowledge compiler built on Andrej Karpathy's LLM Wiki pattern — and it's aimed squarely at developers, researchers, and anyone running a long-horizon knowledge project. Point it at a code repo, a docs folder, arXiv and DOI links, books, decks, screenshots, or plain files and it extracts the text, creates immutable content-hashed source manifests, then compiles everything into a knowledge graph plus an auto-generated Markdown wiki. The pitch is a persistent artifact, not ephemeral RAG: unlike a retrieval pass that evaporates between sessions, the vault gets richer with every source you add. The workflow runs in four stages. Shape defines your swarmvault.schema.md so naming rules, categories, and grounding expectations are explicit. Ingest builds immutable manifests with content hashes. Compile analyzes sources locally, applies the schema, and can optionally borrow model providers for richer synthesis while building a graph where every edge traces to a source and claim. Compound is where you ask questions, save useful answers, and expose the vault to agents over MCP. Everything is CLI-first — init, quickstart, ingest, compile, query, chat, tasks, retrieval, explore, lint, watch, mcp, graph serve — and the raw sources, wiki, and graph state all live as inspectable local files (raw/, wiki/, graph.json, report.json, state/). You can run fully offline with the built-in heuristic workflow and no API keys; OpenAI, Anthropic, Gemini, Ollama, or any OpenAI-compatible endpoint is optional and only needed for richer synthesis. Code files get parsed on your machine via tree-sitter and stay local. Recent releases pushed it further into agent territory: context packs via swarmvault context build and a task ledger via swarmvault task that records goals, decisions, graph evidence, changed paths, and outcomes. Anti-drift linting flags stale sources, conflicting claims, missing citations, and coverage gaps. If you want a private RAG
Behind the Verdict
Reach for SwarmVault when the knowledge itself is the asset. A thesis, a book-length research thread, a codebase you're onboarding into, a memex you intend to still be querying in two years — this is the shape of project it was built for. The four-stage Shape → Ingest → Compile → Compound loop is the reason to pick it over a folder of PDFs and a chat window: the schema forces you to declare how sources should be named, categorized, and grounded before the compiler touches them, and the immutable content-hashed manifests mean you can always trace a claim back to a specific version of a specific source. The provenance story is the part that holds up in practice. Every edge in the graph points at a source, deep lint can attach external evidence to flagged claims, and the anti-drift checks surface stale sources and contradictions instead of letting them quietly contaminate downstream answers. In our experience that's the failure mode that kills long-running knowledge bases, and SwarmVault is one of the few tools that treats it as a first-class problem. Pass on it if you want managed cloud sync, real-time collaboration, or a mobile app. This is CLI-first and local-first by design; the desktop viewer reads the same graph.json, report.json, approvals, and candidates the CLI writes, which is genuinely useful for browsing freshness and review queues on one screen, but it isn't a collaborative workspace. Non-technical users will bounce off the schema and the subcommand surface. Teams that need shared access today should look at hosted options like NotebookLM or a managed RAG service instead. The closest comparison inside the open-source world is a hand-rolled LlamaIndex or LangChain pipeline over your own vector store. SwarmVault wins on structure and durability — you get a
Researching Swarmvault? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Swarmvault actually fits — and what changes day-one when you adopt it.
Run npm install -g @swarmvaultai/cli, then swarmvault quickstart ./your-repo to build a vault from the repo tree, followed by swarmvault next, swarmvault query "What are the key concepts?", and swarmvault graph serve to inspect the graph report.
Outcome: Module pages are compiled with local import resolution, and the graph report plus benchmark artifacts are clickable in the installed viewer reading the same graph.json the CLI wrote.
Write swarmvault.schema.md with domain naming and grounding rules, add arXiv, DOI, and article URLs with swarmvault add, then compile into source pages with normalized metadata and run lint across the vault.
Outcome: Contradictions across sources surface in lint output and graph report pages, and the thesis evolves as an inspectable Markdown wiki rather than a discarded chat transcript.
Point the agent at the vault over MCP using the auto-install rules, then use swarmvault context build to assemble a bounded pack of pages, graph evidence, citations, and explicit omissions, and swarmvault task start|update|finish to log goals and decisions.
Outcome: The agent picks up the same context pack and task history in the next session instead of relearning the project, with context packs saved in state/context-packs/ and task records under state/memory/tasks/.
Use Cases
- Compile a code repository into module pages, resolve local imports, and inspect graph reports plus benchmark artifacts.
- Ingest arXiv, DOI, and article URLs with swarmvault add and build durable source pages with normalized metadata.
- Build a knowledge graph from research papers where lint surfaces contradictions across sources and an evolving thesis forms.
- Stage compile changes, inspect approval bundles, and promote candidates instead of trusting silent rewrites.
- Create a durable memory store for Claude Code, Cursor, or Codex via MCP context packs and task ledgers.
- Build a book companion wiki chapter by chapter with character, theme, and plot pages that compound.
- Maintain a personal memex of journals, health logs, podcast notes, and goals cross-referenced into a structured vault.
- Run fully offline with the built-in heuristic workflow and no API keys, then add a provider for richer synthesis.
Models Under the Hood
as of 2026-10-17
Limitations
- SwarmVault is CLI-first with an optional local desktop viewer, so the core loop runs in a terminal: init, ingest, compile, query, chat, lint, watch, mcp, graph serve, and roughly a dozen more subcommands.
- Raw sources, the LLM-generated Markdown wiki, and graph state all live as inspectable local files (raw/, wiki/, graph.json, report.json, state/), so the artifact is single-machine rather than cloud-synced or real-time collaborative.
- It is built on Karpathy's LLM Wiki pattern and is schema-guided: the wiki is only as good as swarmvault.schema.md and your raw/ folder, and projects with no appetite for writing and maintaining a schema will stall at the first step.
- The built-in heuristic path works locally and offline with no provider setup or API keys; external model providers are optional upgrades, and code files are always parsed on your machine and stay local.
as of 2026-10-04
Verification history
We have re-verified Swarmvault 9 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 9 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Swarmvault tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Open Source (MIT)
$0
Ideal for
Individual developers, researchers, and small long-horizon projects that want a local-first knowledge graph without a software subscription.
What this tier adds
Starting tier — MIT-licensed CLI and desktop viewer at $0; the built-in offline heuristic workflow needs no API keys, and paid model providers are optional.
Where the pricing makes sense
The company stage and team size where Swarmvault's pricing actually pencils out — and where peers do it cheaper.
SwarmVault is published as an MIT-licensed open-source project on npm (@swarmvaultai/cli), so the software cost is $0 and the real spend is the model provider you choose to bring — or nothing at all if you stay on the built-in offline heuristic workflow. That puts it in a different cost category from hosted chat-with-docs and cloud RAG subscriptions where you pay per seat or per document every month, and it suits individual developers, researchers, and small teams running long-horizon knowledge
Setup time & first value
How long it actually takes to get something useful out of Swarmvault — broken out by persona, not the marketing-page minute.
Solo developer: minutes to first compile — install globally with npm install -g @swarmvaultai/cli (Node >= 24) and run swarmvault quickstart ./your-repo. Researcher or book-companion builder: budget an afternoon, since the real work is writing swarmvault.schema.md before your first compile. Agent-memory user: fastest path if the vault already exists — apply the auto-install rules for Claude Code,
Switching to or from Swarmvault
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From chat-with-docs tools: move your source PDFs and files into raw/ and let the compile step build a durable wiki instead of a transcript.
- →From a personal Obsidian vault: SwarmVault writes Markdown and ships an optional .obsidian/ workspace config, so your notes can sit alongside compiled wiki pages.
- →From a manual notes folder: run swarmvault inbox import to pull loose files into the vault as immutable source manifests.
- →From scattered arXiv and DOI bookmarks: use swarmvault add to ingest them as source pages with normalized metadata.
- →From agent chat history: replace pasted context with swarmvault context build packs and a swarmvault task ledger under state/memory/tasks/.
- ↗To another LLM or agent tool: swarmvault export ai writes llms.txt, JSON-LD graph data, and per-page siblings so the compiled wiki can be consumed without running a server.
- ↗To a static site or docs generator: the wiki is Markdown under wiki/, so pages can be published directly.
- ↗To git-based workflows: ingest, compile, or query auto-commit wiki/ plus state/ changes with --commit.
- ↗To a shared repository: your vault is a folder of local files (raw/, wiki/, graph.json, report.json, state/) that can be versioned like code.
- ↗To an Obsidian-based workflow: open the vault with the optional .obsidian/ workspace config already in the project structure.
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Swarmvault”, and we withheld 6: 6 could not be judged, because “Swarmvault” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Swarmvault.
Official links
Tools that pair well with Swarmvault
Common stack mates teams adopt alongside Swarmvault, with the specific reason each pairing earns its keep.
remio
Local-first AI agent that turns your files, meetings, and research into finished decks, sheets, and reports.
MyMemo AI
MyMemo AI turns your saved links, videos, PDFs and voice notes into a searchable knowledge base you chat with.
GraphRAG
Open-source Microsoft Research RAG that builds a knowledge graph and community summaries from your documents, then queries them in four modes.
Featured Head-to-Head Comparisons
Swarmvault vs Praktika
Choose Praktika if you want on-demand speaking practice with AI tutors; choose Swarmvault if you need a private, local-first knowledge graph for research and development. They serve entirely different needs—language learning vs. personal wiki/RAG. No news overrides static facts.
Swarmvault vs Surge Ai
Choose Swarmvault if you're a developer wanting a free, private knowledge graph with offline RAG. Choose Surge AI if you need expert human feedback for training or evaluating frontier models, backed by new rigorous benchmarks and enterprise trust.
Alternatives to Swarmvault
View all
Frequently Asked Questions
Used Swarmvault? Help shape our editorial sentiment research.