Constella
Constella is an AI second brain that indexes your notes, files and connected apps into a self-linking knowledge graph, then answers questions in plain language with citations back to your exact sources. Local-first by default.
What is Constella?
Constella describes itself as an AI second brain: everything you read, write and save is indexed once, then stays reachable by meaning rather than by filename. The homepage is blunt about the difference — this is not keyword search, and the point is to pull an insight in seconds instead of digging for an hour.
The technical claim is graph memory rather than flat embeddings. Notes, papers and chats are indexed into a connected graph, and every answer is traced back to the memo, report or note it came from. Two behaviours follow from that structure, and both are uncommon in this category. Links are drawn automatically as material arrives, so nothing has to be filed by hand. And conflicts are surfaced rather than smoothed over: the product shows where an earlier memo and a later report reach opposite conclusions.
The workspace is an infinite visual canvas. Drop in a question and Constella pulls the connected material, searches the library and the open web, and maps both what is known and what is still missing. Capture runs through several doors: a single / key in the desktop app, a Cmd+Shift+O overlay that appears anywhere and disappears again, the Chrome extension called Sidekick — which clips a page, summarises it as you read and follows you into Google Docs — or the mobile app, by typing, dictation or photo.
Around that sit a dashboard of research agents working from your own context, and Paper Alerts, a daily scan of new scientific literature that emails only the papers touching the research areas you declared. Developers get the same library as an MCP server and as an npm package, so an external agent can call a single recall function and receive grounded, cited context.
Privacy is the second pillar. The knowledge base is written and indexed on your device, recall works offline, and syncing — only if you turn it on — travels through a TLS 1.3 tunnel relayed by the vendor. The free tier is entirely local, needs no account and can run your own models. Constella ships as a web app with no installation, a macOS and Windows desktop client, iOS and Android apps and a Chrome extension, with the desktop client's code published on GitHub.
What it does
- Answer questions about your own corpus, with every claim cited to the exact document it came from
- Index notes, PDFs, memos, local folders and connected apps into a single connected graph
- Link related material automatically by meaning, without folders or tags to maintain
- Flag contradictions where two documents or two engagements reach opposite conclusions
- Capture a thought, a file or a link instantly with the / key or the Cmd+Shift+O overlay
- Clip, summarise and question any web page from the Chrome extension, including inside Google Docs
- Expose the whole library to external agents over MCP or the SDK, and export it to Markdown at will
When to use Constella / When not to
A quick filter to help you decide if Constella is the right fit.
When to use Constella
- Research firms and strategy boutiques whose deliverables depend on recalling what the firm already produced
- Academics, PhD students and postdocs tracking a fast-moving literature, who also get 40% off with a .edu address
- Due-diligence and investment teams working across data rooms, memos, expert calls and sector reports
- Developers building agents who need grounded, cited memory through an MCP endpoint or a TypeScript SDK
- Privacy-conscious knowledge workers who want their note graph indexed on their own machine rather than in a vendor cloud
When not to use Constella
- Buyers who need a documented hosting jurisdiction: no country or region is disclosed for synced data
- Procurement teams requiring formal certification — neither SOC 2 nor ISO 27001 is claimed anywhere on the site
- Non-English speakers: no interface language other than English is announced
- Anyone expecting a store-installed mobile app — no App Store or Google Play listing is linked from the site
- Organisations with a wide application stack: paid tiers cap active integrations at three and six respectively
How to use Constella
A typical end-to-end flow, from setup to results.
- Open the web app in a browser — it needs no installation — or download the desktop client for macOS or Windows 10+
- Start on the free plan: it is entirely local and requires no account at all
- Open Manage sources to add local folders and connect the apps you already work in
- Capture as you go by pressing / in the desktop app to save a note, a file or a link
- Install the Chrome extension, then clip a page or right-click a selection to send it into your vault
- Use the Cmd+Shift+O overlay to capture or recall from any application, then let it disappear
- Type a question into the home ask box and press Enter for a visual recall across everything indexed
- Open Discoveries to browse the concepts, themes and links the graph has built on its own
- Install the mobile app on iPhone, iPad or Android and sign in with the same account to stay in sync
- For agents, copy your personal API key from the Upgrade Your Assistant card or Settings, then point a client at the MCP endpoint with an Authorization Bearer header, or install the SDK and call recall()
Pros & Cons
Pros
- A permanent free plan that needs no account, runs entirely on your machine and accepts your own models
- Answers cite the source document, and conflicting sources are pointed out instead of quietly reconciled
- Links are created automatically by meaning, so there is no filing system left to maintain by hand
- A detailed and specific privacy position: no model training, zero-retention agreements with providers, telemetry opt-in and off by default
- Broad surface coverage — web, macOS and Windows desktop, iOS, Android and a Chrome extension
- Genuinely open to agents, with a documented MCP endpoint, an SDK and working code samples
- No lock-in: full Markdown export, permanent deletion on demand, 40% academic discount and free seats for accredited labs
Cons
- No hosting jurisdiction is disclosed — neither a country nor a region is named for synced data
- No sub-processor list is published and no security certification is claimed, so the processing chain cannot be audited from outside
- End-to-end encryption of the sync tunnel is described as being on the roadmap, which means it is not in place
- The legal documents contradict each other on the minimum age: 13 in the terms, 18 in the privacy policy
- The refund promise diverges too — 30 days on the pricing page against 14 days yearly and 3 days monthly in the terms
- Paid tiers cap active integrations at three and six, with metered AI credits on top of the subscription
- The privacy policy dates from September 2024 and the terms from May 2024, well behind the June 2026 trust page
Pricing & Plans
Constella offers a permanent free plan at USD 0, which requires no account and runs entirely on the user's own device. The lowest paid entry point is the Ultra plan at USD 19 per month, above which the Mega plan is priced at USD 39 per month. Annual billing is advertised at 25% off, and a one-off done-for-you ingestion and setup service is priced at USD 500.
- fully local
- use your own models
- 100% private
- unlimited local note-taking
- unlimited notes
- projects and memory
- up to 300 AI connections
- AI auto-linking
- 3 integrations (Notion
- Gmail
- Slack and others)
- AI research agents over 475M+ papers
- everything in Ultra
- unlimited connections
- auto-linking of thoughts and insights
- 6 integrations
- 25
- 000 AI credits
- instant support and a one-on-one AI second brain coach
- Annual billing — 25% off any paid plan
- Done-for-you ingestion and setup — USD 500 one-off
- live in days with no IT lift
Data, GDPR & hosting
A consolidated view of how Constella handles your data.
GDPR overview
Implementation is claimed but only partly documented. The trust page, updated in June 2026, states that Constella supports GDPR and CCPA data-subject requests and offers a Data Processing Agreement to enterprise and institutional customers, with security reviews, SSO and procurement documentation available on request. The privacy policy lists rights of access, rectification, erasure, restriction, objection and receipt of a copy, exercised by writing to team@constella.app, and refers to marketing consent where the law requires it. Beyond that the record is thin: no EU representative is designated under Article 27, no data protection officer is named, no legal basis for processing is given, no transfer mechanism or standard contractual clauses are described, and no hosting country is disclosed. The privacy policy has not been revised since 15 September 2024.
Who owns the data?
The terms are unambiguous on ownership: you retain every right in the content you create, upload or generate, and Constella App, Inc. claims none of it, including no copyright over what the service produces. The company keeps only the service itself, its features and its code. Personal data is never sold, traded or rented, but the privacy policy permits sharing with service providers, corporate affiliates and any acquirer in a merger or asset sale, plus disclosure on a valid legal request. Private notes, sources and unpublished drafts are never used to train a model — not the vendor's, and not a provider's, since providers are bound by zero-retention agreements.
Reuse rights
No permission is needed to reuse what you put in or get out. You are solely responsible for your own content and warrant that you hold the rights to it, and the publisher explicitly disclaims copyright over anything the service references or produces, so your material and your outputs remain yours to use commercially or otherwise. The exit is built in rather than promised: the full knowledge base exports to Markdown and standard canvas formats at any time, any note, any source or the entire account can be deleted permanently with no soft-delete window, and cancelling a subscription removes nothing — notes and canvas stay on the device. The homepage FAQ states plainly that there is no lock-in.
Data retention & training
Hosting summary
Constella is built local-first: the knowledge base is written and indexed on the user's own device, and search, recall and the links between notes are computed there, which is why recall keeps working without a connection. If device sync is enabled, data travels through what the vendor calls a private tunnel — TLS 1.3 in transit, OS-level secure storage at rest on the device — relayed through the company's infrastructure. Sealing that relay end-to-end is described as a roadmap item, so it is not in place today. The privacy policy adds that server-side data is stored at rest in encrypted buckets, and the Chrome extension page states that synced cloud apps live in the customer's own private bucket. No hosting country or region is named anywhere on the site. What is known is corporate rather than technical: the publisher is incorporated in Delaware and the terms are governed by the law of the State of New York.
Things to keep in mind
Risks and trade-offs to weigh before adopting Constella.
- The tool is designed to ingest everything — notes, files, mail, chat — so the volume of personal or client data concentrated in one place is maximal by construction
- Synced data passes through the vendor's relay and is not end-to-end encrypted, which the trust page acknowledges by listing that work as a roadmap item
- No server-side retention period is published and no sub-processor list exists, so you cannot verify who touches the data or for how long
- Answers are cited but still generated: checking that a citation really supports the claim remains the reader's job
- Handing memory to a tool has a cognitive cost — recall, synthesis and the effort of reformulating in your own words all atrophy when they are outsourced
- Connecting the library to external agents relies on a bearer API key: if that key leaks, the whole knowledge base goes with it
- Using it on confidential client material, as the firm-oriented pitch invites, transfers a real contractual and professional responsibility onto the user
Setup & Integrations
Technical difficulty
Low for everyday use. The web app runs in a browser with no installation, the free plan needs no account, capture is a single / key, and sources are added from a Manage sources screen; the Chrome extension installs in one click and the mobile app only needs the same sign-in. Connecting external agents is a step up: it requires copying a personal API key, running a command-line instruction or editing a client configuration file, and the SDK assumes npm and TypeScript. Firms can skip all of it with the USD 500 done-for-you ingestion service.
Deployment
Integrations
Behind Constella
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Constella.
Frequently asked questions
What exactly is Constella?
Is there a free version?
Is my content used to train AI models?
Where is my data stored?
Can I get my data back out?
Is there an API?
Which platforms does it run on?
Are there discounts for students or academics?
What is the minimum age?
How much setup does it take?
Should you pick Constella?
Constella is one of the more clearly differentiated products in a crowded note-and-recall category. Its distinguishing feature is not search but structure: material is linked automatically by meaning, every answer is traced back to the document that supports it, and the system says out loud when two of your own sources disagree — a behaviour almost nothing else in this space offers. Around that core it stacks a genuinely broad surface, from a no-install web app to macOS and Windows clients, mobile apps, a Chrome clipper, an MCP endpoint and an SDK, so the same library can be reached from a canvas or from a coding agent.
The privacy position is unusually specific for a young company: local-first indexing, no model training, zero-retention agreements with providers, telemetry off by default, a permanent free tier that needs no account and can run your own models, and an export path that is documented rather than promised.
The reservations are documentary rather than functional. No hosting jurisdiction is named for synced data, no sub-processor list is published and no security certification is claimed, which makes an institutional review hard to complete from the public site alone. End-to-end encryption of the sync relay is still on the roadmap. And the legal pages contradict each other in three places — minimum age, refund window and postal address — while both predate the June 2026 trust page by well over a year.
The practical read: an excellent fit for researchers, consultants, analysts and agent builders who accumulate more material than they can remember, and who value citation and contradiction detection over polish. Buyers with a formal security questionnaire should get the hosting, sub-processor and retention answers in writing before committing, since the site does not provide them.
- Choosing a selection results in a full page refresh.
- Opens in a new window.