loading experience
Coming soon

KODINN

The AI workspace that works for you.

A desktop agent that writes code, runs commands and works with Git, together with your mail, notes and documents. With your data encrypted on your computer, and more than twenty AI providers or fully offline.

Download for WindowsComing soon
Download for macOSComing soon

Want to be told at launch? Get in touch.

Scroll to explore
Why Kodinn

A real agent, with your data safe.

Data always encrypted

Database, conversations, credentials and mail are encrypted at rest with a key derived from your PIN.

Personal data stays out

PrivacyGuard masks names, emails, IBANs, tax codes and keys before a message leaves your computer.

Many models, one team

More than twenty providers, several instances, a different model per role and automatic failover. No lock-in.

Offline when you want

Local models run inside Kodinn on CPU or GPU through DesireeIA. No network, no account, no per-token cost.

It checks its own work

QualityGuard: static analysis across 27 languages with a Quality Gate, used by the agent on the code it writes.

Context that lasts

Caveman compresses the conversation instead of cutting it and aligns with the provider's cache.

A real application

Sessions, chat, Git, diff, browser and panels side by side, in Italian, English and Chinese.

Extensible without restarting

Hot-installable plugins, each with its own encrypted database and explicit consent.

QualityGuard

The only AI in the world with total code-quality control.

An AI that writes code should also know whether that code is any good. QualityGuard analyzes the project with a real static-analysis engine and tells the agent whether the Quality Gate passes, before it hands the work back to you.

  • 27languages, from C# to Python, Terraform to Kubernetes
  • 4000+rules, hundreds of them security rules, with data-flow analysis
  • A–Eratings for security, reliability and maintainability
  • GatePASSED / FAILED Quality Gate with the conditions not met

Bugs, vulnerabilities, hotspots, code smells, duplication and technical debt in a single panel, kept up to date by the agent while it works.

QualityGuard
DesireeIA

Our own inference engine. Proprietary, inside Kodinn.

Not a wrapper around someone else's tools: DesireeIA is an inference engine written for the .NET ecosystem, with a native C++ core and a clean public API. It runs models on your computer, inside Kodinn's process, with no network.

1

Detects the hardware

CPU (AVX, AVX2, AVX-512, NEON, NUMA), GPU, NPU and available RAM.

2

Builds the plan

RAM budget, threads, quantization and format picked automatically for your model and your PC; every parameter can be overridden.

3

Loads and runs

A native context with a multi-tier cache, in-process, offline.

Backends, in priority order
NPU AxeleraNVIDIA CUDAIntel ArcApple MetalCPU

Formats and architectures

Reads GGUF and safetensors; multi-architecture dense forward with GQA, RoPE, RMSNorm and gated FFN.

MoE models on low RAM

Experts streamed from disk with an LRU cache, a learned hot set that keeps the most used experts, next-layer prefetch and dual-SSD mirror.

Compressed KV cache

Long contexts with less memory, thanks to key-value cache compression.

Native core, .NET API

C++17 core with a stable C ABI and a .NET wrapper: native performance, clean integration into the app.

Inside Kodinn

See it at work.

Features

Much more than code.

The agent

  • Always has tools: over 50 (files, search, shell, build, Git, browser, memory, databases).
  • Asks before risky actions: every tool has an allow / ask permission.
  • Undo and redo every change, with a diff of each file.
  • Plan mode, sub-agents and teams, 45 bundled skills and slash commands.

DesireeIA

Our own inference engine: GGUF models on CPU or GPU, automatic tuning, local voice dictation.

Sinapsi

  • A full client for REST and SOAP APIs: build requests, inspect responses, save everything in collections.
  • GraphQL and WebSocket too, with environments and variables to move from test to production.
  • Import from OpenAPI, Postman and cURL; export your collections; a history of every call.
  • Bearer and OAuth 2 authentication with token exchange and refresh, and test scripts on the responses.
  • The agent can call APIs for you with the api-call skill, without ever seeing the credentials.

Passwords & vault

  • A sealed vault encrypted with the key derived from your PIN.
  • Passwords, tokens and API keys with notes, tags, expiry and a strength meter.
  • Connectors read credentials from the vault: the model never sees them.

Kanban Board & MemoFlow

  • A board with To do, In progress, Done and Reminders columns: drag cards between them.
  • Priorities, labels, due dates, attachments and a link back to the source mail.
  • “Plan my day”, “Import from mail” and quick entry in plain words: “call Marco tomorrow at 3pm”.
  • Notes, calendar, reminders and a daily planner, with voice dictation.

Telegram

Send messages, photos, documents, polls and locations, read updates and manage chats and members through a bot. The token stays in the vault.

WhatsApp

Click-to-chat links and international number formatting, offline; with WhatsApp Business you send and receive messages programmatically.

Full desktop control

  • The agent moves the mouse, clicks, types and uses shortcuts, just like you would.
  • Captures screenshots, detects monitors and brings an application window to the front.
  • Always behind a permission (allow / ask) and with the vault unlocked: nothing hidden.

Developer workspace

Built-in Git and GitHub, terminal, a browser the agent can drive, editor with language servers, MCP client and server.

EmailFlow

Full mail client (IMAP/SMTP, OAuth), rules, identities, scheduled and undoable send; the agent can read and draft.

Archive & GraphRAG

Encrypted knowledge base with a concept graph built from your documents, search and backup, plus database modules with enforced read-only.

79 connectors

External services with credentials prefilled from the sealed vault.

The guide

It walks you through, step by step.

From first setup to local models and security: Kodinn guides you with its mascot.

Guided setup
Providers & models
Local AI with DesireeIA
Everything under lock & key
Cowork

Work as a team, up to 5 people.

Create a team with people registered on digitalsolutions.it and share work in Kodinn. Prompts and local settings always stay on your computer.

Coming
  • Teams of up to 5 people
  • Sign in with your digitalsolutions.it account
  • Prompts and local settings never shared
Compared

An application, not just a terminal.

KodinnClaude CodeOpenCodeLangChain
InterfaceFull desktop applicationTerminalTerminalNone: it is a library
Data encrypted at restDatabase, conversations, credentials, mail: key derived from your PINConfiguration and history in the clearFiles in the clearNot provided
Personal-data masking before the modelBuilt in (PrivacyGuard), values restored in the answer––Build it yourself
Providers20+ preconfigured, several instances of eachAnthropic models onlySeveral providersMany, through integrations
A different model for each agent roleWith specialisations and priorityOne vendor for everythingModel selection, no per-role routingBuild it yourself
Automatic failover between providersShared health state, a reaction per error type–Left to the userBuild it yourself
Offline local modelsIn-process on CPU and GPU, with our own DesireeIA engineNeeds the remote APIThrough an external serverThrough external integrations
Built-in static analysisQualityGuard: 27 languages, thousands of rules, Quality GateReviews by the model, no engineExternal tooling–
MCPClient and server, dedicated UI, auto-detectionClientClientClient, via adapters
Built-in browser driven by the agentInside the app, the very session you seeSeparate browser extension–Only tools you program
Voice dictationOffline: your voice never leaves the machineRemote transcription––
Mail, notes, calendar, knowledge archive (RAG), databases, API clientAll built in––Build it yourself
REST and SOAP API clientSinapsi, with collections, environments and OAuth 2––Build it yourself
PIN-encrypted password vaultCredentials never exposed to the model––Build it yourself
Kanban board, tasks and plannerMemoFlow, with quick entry and linked mail––Build it yourself
Messaging (Telegram, WhatsApp)Built-in connectors with tokens in the vault––Integrations you code
Desktop control (mouse, keyboard, screen)Built-in tool, always behind a permissionDepends on the version and environment–Only tools you program
PluginsHot install, explicit consent, each with its own encrypted databasePlugins and extensionsCommunity extensionsIntegrations you code
Operational telemetryPanels for tokens, cost, per-minute limits, timeouts, permissionsText commandsPartialTracing to wire up
Context compressionCaveman: routing per content type, cache alignment, guard over critical contentSummarising compactionCompactionStrategies to assemble
Localised interfaceItalian, English, ChineseEnglishEnglishNot applicable

Where the others fit better. Claude Code and OpenCode run anywhere there is a terminal: Linux, containers, CI pipelines, SSH. LangChain is the right base for a pipeline written from scratch. Kodinn is a graphical application for Windows and macOS: when your data, your costs and the quality of the result are what matter, it is built to be the best place to work with AI.

The Kodinn column is verified against Kodinn's source. The other columns reflect the features known as of September 2026: those products are under active development and may change.

Kodinn

Kodinn is coming to Windows and macOS.

In the meantime you'll find examples, the plugin SDK and open code on GitHub.