Unabyss

Overview

Unabyss is a personal-context layer for people who use ChatGPT, Claude, Cursor, and other agents every day and are tired of re-typing the same background — who they are, what they're working on, who they talked to last week, what their voice sounds like — into each session. It pulls that context from the tools they already use (LinkedIn, Notion, Gmail, Slack, GitHub, X, calendars, meeting recorders) and exposes it to any agent that speaks Model Context Protocol.

The framing the team uses is "self-updating context layer." Rather than asking a user to maintain persona.md and voice.md files by hand, Unabyss extracts the data once, structures it into those files automatically, and refreshes them as the source tools change. Permissions are granular: each MCP host (Claude, Cursor, OpenClaw, Perplexity, etc.) gets a scoped token deciding what slice of the context it can read.

The target user is the founder/operator/builder running half a dozen AI agents daily and watching the same intro context get pasted into every fresh chat. Unabyss starts free with $5 in signup credits, no card required. After those credits, users connect a card and pay as they go; no monthly or annual tier prices are published.

Key Features

• Auto-extraction from connected source apps — Pulls structured context from LinkedIn, Notion, Gmail, Slack, GitHub, Obsidian, X/Twitter, Google Drive, Google Calendar, OneNote, and meeting recorders such as Fathom, Fireflies, tl;dv, and Granola. The official launch copy says extraction can complete in under 90 seconds, while the public website lists these apps plus "+ and more" rather than a precise source count.
• MCP server for every major agent — Generates a scoped MCP token (una_live_8f2b3c…d91a) that drops into Claude, Cursor, Claude Code, Codex, Gemini, Perplexity, OpenClaw, OpenCode, VS Code, and ChatGPT. The same context surfaces in every agent without copy-paste.
• Granular per-host permissions — Each agent gets a different access scope. Exclude private items, hide confidential notes, and segment by topic or confidence level so a public ChatGPT session never sees what a private Claude session can.
• Structured persona files — Aggregates raw source data into persona.md and voice.md style files that LLMs read cleanly. No more bespoke prompt templates per agent — the same canonical persona feeds every model.
• One-click exports — Pre-built exports for investor updates, meeting prep, and similar recurring artefacts. Pulls the most recent context window without manual selection.
• Continuous sync — Source tools are re-polled automatically so the context stays current. A LinkedIn role change or a new Notion doc flows into the layer without re-onboarding.

How to Get Started

1. Sign up at unabyss.com — Email or Google sign-in; no payment method requested.
2. Connect one or two sources — LinkedIn, Notion, and Gmail are common first picks. Auto-extraction starts in roughly 90 seconds.
3. Review the generated context — Inspect the auto-built persona.md / voice.md; redact or hide anything sensitive before sharing it with agents.
4. Generate an MCP token per agent — Each token (una_live_...) is scoped to a specific host. Copy on creation; the platform won't show the secret again.
5. Plug the token into your agent — Add the MCP server to Claude, Cursor, Codex, or any compatible host; the agent then reads the chosen slice of context whenever it starts a session.

How It Compares

• vs. ChatGPT Memory / Claude Memory — Built-in memory features live inside one product and don't follow the user across agents. Unabyss is the cross-agent layer; the same persona shows up in Claude, ChatGPT, and Cursor without re-teaching each one.
• vs. Mem0 / Letta / dedicated memory frameworks — Those are developer SDKs aimed at app builders adding memory to their own products. Unabyss is end-user-facing — the user, not a developer, is the integrator.
• vs. hand-written MCP servers — Engineers can roll their own MCP server reading from Notion or GitHub; Unabyss bundles 20+ connectors plus persona structuring out of the box, trading some control for time-to-first-context.
• vs. Cassidy / Glean / general AI assistants — Those are search-and-summarize layers across team knowledge. Unabyss is narrower and personal — it's about your context as a person, not the team's collective knowledge.

The honest read: Unabyss is best when the user is the central hub of context (founder, solo operator, consultant) and the same persona/voice needs to follow them across many agents. For team-wide knowledge or product-embedded memory, the comparison shifts.

Pricing & Plans

Unabyss starts free with $5 in signup credits. Sign-up requires no payment method and unlocks every feature, every integration, and MCP token generation; after the included credits, usage is pay-as-you-go.

What's not yet public:

• The exact credit cost per operation for source ingestion, exports, MCP queries, stores, and context chat
• Any non-credit usage caps on source extraction frequency, token count, or number of MCP hosts
• An enterprise tier with admin controls, team-shared context, SSO, or audit logging

Treat the $5-credit and pay-as-you-go posture as launch-phase. Teams relying on this in production should ask the team directly about unit pricing, long-term billing, and SLAs before depending on it for revenue-critical workflows.

Best For

• Founders running their own sales, content, and product across Notion, Slack, and Gmail who want one persona surfaced to every agent
• Solo operators and consultants whose work product depends on a consistent voice across many drafts and many models
• Builders evaluating multiple AI productivity agents who need each one to read the same up-to-date context
• Power users of ChatGPT, Claude, and Cursor who experience daily friction re-onboarding each new chat
• Early adopters comfortable trusting a beta service with read access to personal accounts

FAQ