Andes LabsAI memory · Knowledge graphv1.0 · Open source · MIT2026

Your AI forgets.
Kernal remembers.

An ambient knowledge graph that turns every conversation, meeting, and email into structured memory. AI memory built like a graph, not a blob. Local-first, MCP-native, SQL-simple, open source.

Built & trusted onAWS Activate PartnerMCP ProtocolMIT Licensed
Live graph · hover a node
Drag · hover · click
PersonOrganizationTopic

The Problem

Smarter models won't save your AI program.
The substrate will.

“Through 2028, more than 50% of AI initiatives will halt — not because the models stop improving, but because the plumbing fails.”

— Gartner · Litan, Ramirez, Powledge · 2025-2026

The bottleneck has moved. It's no longer model intelligence — it's the substrate the model sits in. Unresolved agentic identity. Untracked machine identities. Knowledge that starts at zero every session.

Most teams respond by adding layers: RAG, vector DBs, embedding services, “memory” SaaS, eleven kinds of orchestration. It works at demo time. It fails at year two. That's the 50%.

Kernal is the other path. A structured, persistent, ambient knowledge graph that lives below every agent you run. Same SQLite file on your laptop, your server, and your cloud. Same MCP protocol. Same answers, every session.

Proof of Concept

Now imagine this is
your client portfolio.

Fifty meetings. A hundred and ninety people. Fifteen hundred relationships. Every name, every connection, every promise your team has ever made — structured and queryable by your AI agent.

50+
Transcripts processed
190+
People extracted
1,500+
Relationships mapped
8
Enterprise accounts

Real production data from enterprise consulting engagements.

Graph Intelligence

Ask questions that
search can't answer.

Vector search finds documents. Kernal traverses relationships.

    Philosophy

    Context is a feature.
    Ambient is an architecture.

    The first wave of AI products was context-aware: you tell the model what to know, it responds. ChatGPT memory. Claude projects. RAG-as-feature. The model is the brain; context is something you feed it.

    The next wave — the one that survives Gartner's 50% cull — is ambient. The knowledge graph is there before you ask. Persistent across sessions, agents, products. You don't prompt it. It's resident.

    Ambient is harder. It requires structure, not vibes. Schema, not vector blobs. A database that doesn't forget — and never assumes you'll re-tell it what it should already know.

    Kernal is built ambient-first. Every conversation strengthens the graph. Every agent that connects via MCP inherits the same memory. Better models make Kernal more valuable, not less — because the graph is the durable layer, and the model is the rented one.

    Architecture

    Five layers of intelligence.

    Every entity connects to every other via the relationship graph. The graph is the product.

    L5 · IntentIntent
    Goals and milestones that thread through every decision below.
    “Go legitimate within 5 years.”
    L4 · CommercialCommercial
    Deals, pipeline, stakeholders — the business reality.
    “Las Vegas Casino Acquisition.”
    L3 · IntelligenceIntelligence
    Strategic plans, patterns, and the insights that connect them.
    “Barzini is the real threat, not Tattaglia.”
    L2 · ExecutionExecution
    Actions, tasks, decisions — what actually gets done.
    “Send Clemenza to handle the Vegas situation.”
    L1 · ConversationConversation
    The raw material: meetings, calls, transcripts, emails.
    “Peace Summit with the Five Families.”

    Capabilities

    What Kernal does.

    01 · Ingestion

    Entity Extraction

    Drop in a transcript. Kernal extracts people, organizations, topics, and the relationships between them. No templates. No configuration. Just structure.

    02 · Query

    Graph Search

    Not just keyword matching. Query across relationships: “Who at Nordic Tech has influence over the SAP decision?” Kernal traverses the graph.

    03 · Protocol

    MCP Protocol

    Plug Kernal into Claude, Cursor, or any MCP-compatible tool. Your agent gets structured context about your world — people, relationships, history — in every conversation.

    04 · Prep

    Meeting Prep

    Before any call, Kernal surfaces: who you're meeting, their org chart position, recent interactions, open action items, and strategic context. Automatically.

    Data Sovereignty

    Your graph. Your machine.
    Your models.

    Most AI knowledge tools send your data to someone else's servers. Kernal doesn't. No DPA needed. No data residency concerns. No vendor training on your client intelligence.

    I

    Local processing

    Entity extraction runs via Gemma 4 on your machine. Your transcripts, client names, and deal details never leave your infrastructure.

    II

    Local storage

    Your knowledge graph lives in a SQLite file on your device. No cloud database. No vendor access. Full portability.

    III

    Open protocol

    MCP is an open standard, not a proprietary API. Connect Kernal to Claude, Cursor, or any compatible tool. No vendor lock-in.

    Executive searchManagement consultingLaw firmsRegulated industries

    Built For

    Professional services that
    run on relationships.

    Kernal is designed for people whose work depends on knowing who connects to whom, what was said, and what it means.

    · I ·

    Executive Search

    Map candidate networks, board relationships, and organisational dynamics. Know who knows who before the first call.

    · II ·

    Management Consulting

    Build institutional memory across engagements. Every meeting, every stakeholder, every strategic decision — structured and searchable.

    · III ·

    Strategic Advisory

    Track deal pipelines, stakeholder influence, and client goals across your portfolio. Your AI agent knows the full picture.

    · IV ·

    Law Firms

    Matter context that compounds. Client relationships, precedent connections, and engagement history — locally stored, never shared.

    Get Started

    Use the tool you already use.
    Kernal is the brain underneath.

    Same principle as Outlook or shared drive — a company asset issued on day one, surfaced through whichever AI host the team already trusts.

    A

    Self-hosted (free, open source)

    Install locally with one command. Your machine, your data, zero dependencies.

    $ npx @kernal/mcp

    Then add Kernal as an MCP server in whichever host you already use — Claude, Cursor, Codex, Copilot, ChatGPT.

    B

    Kernal Cloud (managed)

    No install needed. Connect directly from your existing AI tool and start building your graph in seconds. Limited managed-access while the product stabilizes.

    Talk naturally

    Start having conversations. Kernal extracts and connects entities automatically. The graph compounds.

    Read the docs on GitHub

    Community

    Join the builders.

    Book a demo

    See Kernal in action. 30 minutes with the founder to explore how it fits your workflow.

    Book a meeting

    GitHub

    Star the repo, open issues, contribute. The core is fully open source under MIT.

    View Repo

    SubStack

    Technical deep dives on agent architecture, knowledge graphs, and the MCP ecosystem.

    Subscribe

    Build agents
    that remember.

    See how Kernal works with your data. Book a 30-minute demo or start building with the open-source core.

    Relationships
    Graph position
    Tweaks Direction C