01 — the catalog

Everything you need, one click away

Add apps, models, skills, tools, knowledge bases, agents and GitHub repos straight from the catalog. Extella installs, connects and saves the setup so next time you never start from scratch. If something is missing, just describe the task and the system assembles the workflow.

GitHub

Paste a link or pick from the catalog. Extella deploys the repo and gets it ready for work.

Hugging Face & libraries

Connect models, datasets and libraries with no manual setup. Everything is saved for future runs.

Agents, tools, APIs

Pick something ready-made from the catalog, or describe the task and Extella wires up the right components itself.

02 — how it works

Three steps. No setup ever again

/ 01

Describe the task

Say in plain words what result you need. No prompt engineering, diagrams or manual setup.

/ 02

Extella executes

The system works with real files, data, APIs and tools. The output is not advice in a chat — it is the finished task: a file, a report, an action or a working process.

/ 03

Becomes a tool

A successful solution is saved to your library. Run it again, improve it, hand it to the team, or build new processes on top of it.

03 — under the hood

Four layers everything rests on

rules/behavioral layer

How agents act

Rules define how an agent behaves: working style, constraints, roles, safety rules and preferences. Say it once and the rule holds in future sessions. And when you correct an agent, the system can save that as a new rule.

concepts/semantic memory

What agents remember

Concepts store knowledge, not chat history: your processes, documents, clients, decisions and preferences. The right context is pulled in by meaning, automatically — so you never explain the same thing twice.

experts/execution layer

What agents can execute

Experts are isolated execution modules for code, models, libraries and external tools. One expert can transcribe audio with local Whisper, another process a spreadsheet, a third wire the result into an API. Each runs on its own, with no dependency conflicts.

kv store/secrets

Where keys and credentials live

API keys, tokens and configuration are stored separately and encrypted on your device. They never appear in plain text in code, chat or model memory. An agent gets only the access a specific task requires.

04 — agents & teams

One request. A team of specialists inside

Build agents for different roles: analyst, developer, editor, researcher, data operator. Each has its own knowledge, rules, tools and model. A master agent understands the task, hands parts of the work to the right specialists and assembles the result.

Isolation is built into the architecture

Every agent has its own system prompt, model, tools and memory. Contexts split across three levels: Global, Team and Agent. Shared rules stay shared, team knowledge stays within its group, and an agent's personal memory never mixes with other tasks.

Its own model for every agent

OpenAI, Anthropic, Google, xAI, Qwen, MiniMax, your own endpoint, or a local model via Ollama. Each agent can have its own provider, temperature and reasoning mode. Different models work side by side in one team.

Every agent with its own context

Five agents in parallel means five separate context windows. One reads a long document, another works a spreadsheet, a third calls an API. Contexts never collide, so complex tasks split across agents without overload.

05 — cspl runtime

CSPL — Container Specific Programming Language — Extella's own execution layer.

Describe the “what”. The runtime knows the “how”

CSPL separates the intent of a task from the mechanics of execution. The model writes a short declarative description; a deterministic handler turns it into exact code. Improve the handler once — and every task of that class gets better, without rewriting a single line.

Up to 15× fewer tokens on repeated tasks, and not a single hallucinated step along the way.

extella · runtime

task: weekly-client-report handler: weekly_report · deps resolved pulling data · 4 sources ✓ report generated · 12 pages ✓ delivered to 6 recipients tokens used: 1/15 of chat baseline

before · the standard approach

Every expert knew both the “what” and the “how”

An expert carried both the goal and the technical mechanics. On complex tasks the LLM re-guessed structure, syntax and execution steps every single time. That burned tokens and gave unstable results.

with CSPL · Custom Scripting / Processing Layer

The model declares the “what”. The handler executes the “how”

An expert describes the desired result in a compact format such as JSON or a DSL. The handler reads that description and executes it a proven way: generating code, calling libraries, running tools or chaining other experts.

06 — scenarios

Built once. Run it again

Every pipeline you build stays in Extella as a ready-made skill. Repeat it, improve it, and use it as the base for new tasks.

knowledge work

Research without manual assembly

Extella gathers sources, removes duplicates, prepares briefs and reports. With every run it gets better at your style, structure and requirements.

ops & revops

Operations across services

Connect CRMs, spreadsheets, drives and APIs. Extella ties the data into one process and removes the manual hand-offs between tools.

marketing & brand

Content in your voice

From plan to finished publication. Extella holds your rules, references and style so every piece reads like one brand.

dev productivity

Tools for development

Spin up internal tools, scripts and integrations with a single instruction. Finished solutions are saved to the library and run again.

personal os

A personal OS that forgets nothing

Notes, decisions, recurring routine. Extella remembers what usually gets lost — and sees things through.

analytics

Data, ready to use

Work with spreadsheets, exports and dashboards. Ask a question and get more than an answer — a finished artifact: a report, a file or a visualization.

Any model. Any API. Any tool

GitHub · Hugging Face · OpenAI · Anthropic · Google · xAI · Qwen · MiniMax · Local LLMs · Claude Code · OpenClaw

Telegram · WhatsApp · Slack · Gmail · Notion · Google Sheets · CRMs · Databases · Internal Systems · and more

Extella sits on top of your stack and orchestrates everything underneath.

One place for everything