Memory + agentic layer for your LLM

Give your AI a memory.
Change one URL.

Kortexio sits between your app and your LLM as an OpenAI-compatible memory layer. It remembers every session, runs tools when asked, and never touches your existing integration — point your SDK at Kortexio and everything else stays the same. Ollama clients can keep using /api/chat.

Start free — no credit card See how it works Read the quickstart →

your-app · chat.ts

  POST https://api.yourapp.com/chat
- POST https://api.openai.com/v1/chat/completions
+ POST https://api.kortexio.io/v1/chat/completions

request · turn 1

POST https://api.kortexio.io/v1/chat/completions
Authorization: Bearer cmk_live_...
Content-Type: application/json

{"model":"gpt-4o-mini","user":"user-42","messages":[{"role":"user","content":"Remember: KORTEX-PINEAPPLE"}]}

response · turn 2 (same X-Session-Id)

{"choices":[{"message":{"role":"assistant","content":"The secret word is KORTEX-PINEAPPLE."}}]}
# Memory recalled — no code changes required

That's the entire integration. Session memory works on the next turn automatically.

The problem

Your LLM forgets everything the moment the response ends.

Every new message starts from zero. Long conversations get expensive, context gets truncated, and building real memory means standing up a vector database, an ingestion pipeline, and a retrieval layer — before you've shipped a single feature.

Kortexio remembers so you don't have to build it.

Every session gets its own compiled memory, maintained automatically after each turn. No embeddings to manage, no retrieval logic to write — just a persistent, structured memory your model can actually use.

How it works

Three steps. Same contract you already use.

Point your endpoint here

Swap your chat completion URL for Kortexio's OpenAI-compatible /v1 endpoint. Same SDK, same request body — memory is ambient.

Kortexio remembers and acts

Each session builds its own memory. Ingest docs into Global Wiki — with digests, a searchable catalog, and point-in-time facts. Skills and guardrails steer the agent; HITL pauses destructive tools until a human confirms.

Your app gets a smarter answer

The response comes back in the exact format your app already expects. Nothing else in your code changes.

Read the quickstart →

What's included

Everything memory needs, nothing you have to build.

Session Memory

Every conversation gets a structured, evolving wiki — compiled before each response, compacted automatically as it grows. No vector pipeline to own.

Global Wiki

Ingest Jira, Confluence, SQL exports, and pipelines into an app-wide knowledge base. Digests and a wiki catalog keep retrieval cheap — the agent calls wiki_search only when it needs those facts.

Temporal knowledge

Facts are revisioned, not overwritten. Query what was true last quarter with asOf — supersede updates keep history for audits, compliance, and “what changed?” answers.

Skills & guardrails

Platform and per-app policy packs (anti-hallucination, wiki-first, privacy, and your own) shape how the agent behaves — editable in Markdown, inherited cleanly across apps.

Agentic tools + HITL

Sandbox, MCP, and built-in wiki tools on the same chat URL. Destructive actions pause for human confirmation; progress streams so clients can show iteration timelines.

Playground

Try memory, tools, and HITL in the dashboard before you wire production — streaming replies and an agentic timeline so you see every step, not just the final answer.

Bring Your Own Integrations

Connect your own MCP servers — internal tools, ticketing, billing — scoped per app, never shared across tenants. Hosted memory MCP for Cursor and Claude in minutes.

Multiple Apps, One Account

Run a support assistant and a sales agent with completely separate memory, tools, and rules — under a single Kortexio account.

European Infrastructure

Hosted and billed entirely within the EU. No data leaves European infrastructure by default.

Why not just build it yourself

Other memory tools ask you to rewrite your app. Kortexio asks you to change a URL.

Kortexio

Typical vector-store memory

Integration effort

One URL

New SDK, new data pipeline

Setup time

Minutes

Days to weeks

Memory format

Structured session + Global Wiki

Raw embeddings

Knowledge retrieval

On-demand wiki_search + digests

Always-on RAG inject

Point-in-time facts

asOf / supersede revisions

Overwrite or re-embed

Policies

Skills + guardrail packs

Prompt soup in your app

Agentic tools included

Yes — with HITL + playground

Usually a separate product

Infra to maintain

None

Vector DB + ingestion jobs

Pricing

Simple tiers. Concrete limits.

BYOK — use your own LLM key with no markup on tokens. Export your wiki memory as markdown anytime.

Free

€0/month

1,000 / month
Apps1 app
Web searchNot included
MCP servers0 MCP servers
Memory retention30 days
Seats1 seat
SupportCommunity

Start free

Your data stays in Europe. Full stop.

Kortexio runs entirely on European infrastructure, billed through a European payment provider. Nothing about your account, your memory data, or your usage ever routes through infrastructure outside the EU by default.

Kortexio is built for GDPR compliance by design. A Data Processing Agreement is available to every customer.

API keys are hashed, never stored in plaintext
Every admin action is logged and auditable
Destructive agentic actions require explicit human confirmation (HITL)
Global Wiki revisions keep superseded facts for audit and asOf queries
Full data export and deletion available on request

Questions

Frequently asked

Do I need to change my app's code?+

Just the base URL. Point the OpenAI SDK (or any OpenAI-compatible client) at api.kortexio.io/v1. Request and response shapes stay the same — Ollama clients can keep using /api/chat.

What happens to my data if I cancel?+

You can export everything before closing your account. After cancellation, your data is deleted according to our retention policy.

Can I run more than one assistant under one account?+

Yes — each app you create has its own memory, tools, and rules, completely isolated from your other apps.

What is Global Wiki?+

An app-wide knowledge base for ingested docs (Jira, Confluence, SQL, and more). Digests and a wiki catalog keep discovery cheap; your agent searches with wiki_search when session memory alone is not enough — shared across all sessions of that app.

Can I query what was true in the past?+

Yes. Global Wiki documents are revisioned. Updates supersede previous facts by default, and you can query with asOf to see what was valid at a point in time — useful for audits and “what changed?” answers.

What are skills and guardrails?+

Policy packs that shape agent behaviour — platform defaults plus per-app packs you edit in Markdown (anti-hallucination, wiki-first, privacy, and custom rules). Destructive tools still pause for human confirmation (HITL).

Does Kortexio work with Microsoft Copilot?+

Yes — any app you build in Kortexio can be exposed as a Microsoft 365 Copilot agent, discoverable inside your organization's Copilot.

Where is my data hosted?+

Entirely within the EU, on European infrastructure, billed through a European payment provider.

Get in touch

Questions before you start?

Whether you're evaluating Kortexio for a team or need something custom for an enterprise deployment, we're happy to talk.

Prefer email? Reach us directly at hello@kortexio.io

Give your AI a memory.Change one URL.