Unabyss
Unabyss is a universal context layer for AI tools. It connects Gmail, Slack, Notion, GitHub and twenty other sources, structures them into one living profile, then serves it to Claude, ChatGPT, Cursor and any MCP client.
What is Unabyss?
Unabyss calls itself the universal context layer for AI tools, and the description is accurate: it is neither the memory built into one assistant nor a Markdown file you maintain by hand, but a separate store of who you are that any AI client can read. The problem it targets is stated in two halves on the homepage. Your AI tools do not talk to each other, so whatever you worked out in ChatGPT is invisible to Claude and to Cursor. And your real context lives in Slack, Gmail, Notion, Calendar and GitHub, where none of them can see it. The mechanism is a one-time connection followed by continuous extraction. You authorise your sources through OAuth, and the platform pulls, structures and tags the material by topic, source and sensitivity into three layers: Identity for public facts, Profile for professional history and working style, and Mind for private notes that are opt-in only. That structured picture is then served over the Model Context Protocol from a public JSON-RPC endpoint, so Claude, Cursor, ChatGPT, Grok, VS Code and a dozen other clients pull fresh context at the start of each session rather than being told it again. Two design choices set it apart from a static file. The flow runs continuously, so the context never becomes a snapshot, and it is bidirectional: what you do inside connected agents feeds back in. When two sources disagree, the conflict is flagged rather than silently resolved. If MCP is not an option, twenty-plus one-click Markdown exports and a REST API cover the rest. Beyond the individual product there are two extensions: a Team plan where each colleague connects their own accounts and nothing is pooled, and Company Brain, an enterprise knowledge base built by agents in around forty-eight hours and served through a single MCP endpoint. The publisher, OneType Prosta Spółka Akcyjna, is a small Warsaw company. The product is young and moving fast: a Product Hunt launch in late May 2026, version 1.16.0 by 11 August 2026, and a public changelog running back to v0.3.0.
What it does
- Pull your working context out of Gmail, Slack, Notion, GitHub, Calendar and twenty other apps
- Turn that raw material into one structured profile, tagged by topic, source and sensitivity
- Serve the right slice of it to Claude, ChatGPT, Cursor or any MCP client, on demand
- Decide app by app and file by file what each AI tool is allowed to read
- Export the whole context as Markdown and paste it into any tool that accepts text
- Search your own context semantically and chat with it to correct or complete it
- Run the same model across a team, where each seat keeps its own private context
When to use Unabyss / When not to
A quick filter to help you decide if Unabyss is the right fit.
When to use Unabyss
- People who split their day across several AI assistants and are tired of briefing each one from scratch
- Founders who want every AI they touch to stay current on strategy, customers and what actually shipped
- Developers and technical builders whose agents need past decisions, coding conventions and repository state
- Agencies juggling several client accounts, where each client's context has to stay cleanly separated
- Go-to-market teams whose pipeline, account notes and call recordings live in five different tools
When not to use Unabyss
- Anyone who lives inside a single assistant: the vendor itself concedes that built-in memory is enough there
- Engineering teams shopping for a memory SDK to embed in their own product, a need the FAQ redirects to Mem0
- People whose situation barely changes and who are well served by a hand-written context file
- Anyone under eighteen, since the terms and the privacy policy both set that as a hard floor
- Users who need a mobile app or a browser extension today, neither of which has shipped
How to use Unabyss
A typical end-to-end flow, from setup to results.
- Create a free account at app.unabyss.com/register; no credit card is requested to start
- Connect at least two sources in one click each, through OAuth or an API key where OAuth is unavailable
- Let the extraction run: the site advertises a usable context in under ninety seconds
- Review the structured profile and adjust what sits in the Identity, Profile and Mind layers
- Add the official Unabyss connector from the Claude directory, or generate a static unby_mcp_ token for another client
- Point your client at the bare origin https://mcp.unabyss.com, never at a path underneath it, or authorisation fails
- Open a new chat and send the setup message so the assistant finishes onboarding for you
- Set permissions app by app and file by file, and revoke any of them instantly if needed
- If you would rather skip MCP, generate a one-click Markdown export and paste it into any AI tool
- For a team, buy at least three seats, then invite colleagues by email within the seven-day invitation window
Pros & Cons
Pros
- One context shared by every AI tool, instead of a separate memory locked inside each one
- Stays current on its own from live sources, and feeds back what you do inside connected agents
- Fast to adopt: a one-click official connector in the Claude directory and a context ready in about ninety seconds
- Wide, verifiable integration coverage, with thirteen MCP clients and twenty-five source applications listed
- Fine-grained permissions by app, by file and by sensitivity layer, backed by an audit log
- Repeated, explicit commitment that customer context is never used to train any model
- EU-based publisher, GDPR directly applicable, and a subprocessor list published by name
Cons
- Very young product: the domain was registered in March 2026 and the first archived capture dates from June 2026
- SOC 2 Type II is described as in progress, so no certification report can be relied on today
- No mobile app at all, and the browser extension is still announced rather than shipped
- The free plan exists in the terms but is nowhere to be seen in the pricing grid on the site
- Adopting it means granting read access to email, drive, internal chat and code repositories at once
- Once context leaves over MCP, the vendor disclaims responsibility for what the receiving tool does with it
- The managed PostgreSQL database may sit outside the EEA, despite the EU-hosted claim in the marketing FAQ
Pricing & Plans
A permanent free plan is provided for in the terms and conditions, with limited features and usage caps, although it does not appear in the pricing grid displayed on the site. Every paid plan opens with a seven-day free trial. The cheapest paid entry point is the Pro plan at USD 13.00 per month on annual billing, or USD 15.00 per month without commitment.
- Free — a no-cost tier defined in the terms
- with limited features and usage caps
- not shown in the pricing grid
- up to 3 connected agents over MCP
- up to 20 connected accounts
- premium usage tier
- seven-day free trial
- unlimited agents
- unlimited connected accounts
- multi-account support for Google apps
- unlimited usage
- early access to new features
- full Max access on every seat
- shared workspace
- email invitations
- central seat and invoice management
- minimum three seats
- Company Brain — enterprise offer for organisations
- no public price
- arranged through a booked demo
Data, GDPR & hosting
A consolidated view of how Unabyss handles your data.
GDPR overview
GDPR implementation is detailed and concrete. The controller is established in Poland, so the regulation applies directly and no Article 27 representative is needed. No DPO is appointed, which the policy justifies under Article 37; the privacy contact is privacy@unabyss.com. Each processing purpose carries a stated legal basis under Article 6, and the eight data subject rights are spelled out with a one-month response deadline. Subprocessors are named individually, with notice before a new one is engaged. Transfers outside the EEA rely on adequacy decisions, the EU-US Data Privacy Framework or Standard Contractual Clauses. The supervisory authority is the Polish UODO. The policy took effect on 31 July 2026. SOC 2 Type II is described as in progress, not obtained.
Who owns the data?
Under the terms, all Context Data imported, uploaded or created stays the property of the user, and AI Output generated from it belongs to the user too once fees are paid. Unabyss holds only a non-exclusive, worldwide, royalty-free licence, limited to hosting, displaying, transmitting and processing that material in order to run the service; it expressly excludes training any AI model and terminates automatically when the account is deleted. OneType Prosta Spółka Akcyjna is the data controller. On Team plans the employing organisation is normally the controller and Unabyss acts as processor, but administrators still get no view into a member's context.
Reuse rights
Users keep full reuse rights over their own material: context files and AI Output can be exported as Markdown at any time, moved between models and reused without asking permission, and portability is restated as a GDPR right in the privacy policy. On the vendor side, use is bounded to running the service, analysing, classifying and enriching context, powering Context Chat and exports, and distributing context over MCP on the user's instruction. Connections are read-only by design, so Unabyss imports but never posts, edits or deletes inside a connected tool. Personal data is not sold or rented, and neither Unabyss nor its AI subprocessors may train on it. Once context reaches a third-party AI client the user chose, that client's own terms govern what happens next.
Data retention & training
Hosting summary
Application infrastructure and object storage run on servers operated by Contabo within the European Economic Area, while the production PostgreSQL database is managed by Neon and may be provisioned in a region outside the EEA. Edge delivery, CDN and the signed-URL proxy run on Cloudflare. Regular backups are held on S3-compatible storage under the same protections as production. Data is encrypted in transit and at rest, with short-lived JWT access tokens and rotating refresh tokens. Several subprocessors, including OpenAI, Anthropic, ElevenLabs, Stripe, Neon, Pipedream, Chatwoot and Google, may process data in the United States or other non-EEA jurisdictions, covered by adequacy decisions, the EU-US Data Privacy Framework or Standard Contractual Clauses. Depending on the plan, context can instead be forwarded to an external destination configured by the user and never persistently stored by the vendor. Note that the marketing FAQ states EU-hosted data more categorically than the privacy policy does.
Things to keep in mind
Risks and trade-offs to weigh before adopting Unabyss.
- Connecting Unabyss means granting read access to email, drive, internal chat and code repositories at once, which concentrates a great deal of sensitive material in a single place
- Once context is delivered to a third-party AI client over MCP, the publisher disclaims responsibility for how that client stores or uses it
- SOC 2 Type II is advertised as in progress, so security claims rest on the vendor's own description rather than on an audited report
- The managed database may be provisioned outside the EEA even though the marketing FAQ promises EU-hosted data, a nuance worth checking before committing regulated data
- There is no guarantee of permanent storage, and the publisher explicitly advises keeping your own copies of anything critical
- The list of restricted jurisdictions announced in the terms is not actually published, so you cannot verify in advance whether your country is eligible
- Handing your working memory to an outside layer makes it easy to stop noticing what you actually know, and the company behind it has barely a year of visible activity
Setup & Integrations
Technical difficulty
Low for most users. Creating an account and connecting sources takes one click each, and the publisher advertises a usable context in under ninety seconds. Claude users add an official connector from the directory with nothing to paste. Other clients need a static token or an OAuth flow, which is straightforward but assumes familiarity with MCP configuration, and the endpoint must be entered as a bare origin or authorisation fails. No code is required for normal use. Anyone uncomfortable with MCP can fall back on a one-click Markdown export.
Deployment
Integrations
Supported languages
Behind Unabyss
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Unabyss.
Frequently asked questions
What exactly is Unabyss?
How is it different from ChatGPT Memory or Claude Memory?
Which sources can I connect?
Do I have to use MCP?
Is my data used to train AI models?
Where is my data hosted?
What does it cost and is there a free option?
Is there a minimum age?
Should you pick Unabyss?
Unabyss answers a problem that has become real for anyone running more than one AI assistant: the context you build in one tool is invisible to the next, and the material that would actually inform a good answer is scattered across email, chat, documents and repositories. The response is coherent. Sources are connected once and read continuously, the extracted material is structured and tagged rather than dumped, and the result is served over an open protocol to whatever client you happen to be using. The execution matches the pitch: a documented MCP endpoint, an official connector in the Claude directory, twenty-five source integrations verifiable on the integrations page, permissions that go down to the individual file, and an audit trail. The privacy position is unusually clear for a young company, with a named subprocessor list, an EU controller and a training exclusion written into the licence itself. The reservations are mostly about maturity. This is a product measured in months, not years: the domain dates from March 2026, SOC 2 Type II is still in audit, there is no mobile app, and the browser extension remains an announcement. The pricing story has a loose end, since a permanent free plan is defined in the terms but never appears on the pricing grid. Above all, the value proposition requires handing read access to email, internal chat, drive and code at once, which is a large trust surface to grant a company with a short track record. Worth a serious trial if you genuinely split your work across several AI tools and want to own the context rather than rent it from one vendor. Less compelling if you live inside a single assistant, where the vendor itself admits built-in memory does the job.
- Choosing a selection results in a full page refresh.
- Opens in a new window.