Self-hosted orchestration

Orchestrate anything.
Approved by people.

Describe a process across your apps in plain language. AI drafts the Flow, a person approves it, and the engine runs exactly that version, with each member's own access.

Apache 2.0 · Runs on your own infrastructure · Your data stays with you

Why Dittoly

Do it with AI once. Repeat it without.

Many tasks we hand to an AI are the same task, asked again and again: each time it costs tokens and energy, and the same request can come back worded or done differently. Dittoly turns that task into small, tested modules joined in a Flow, so it comes back fast, the same way every time, without asking a model again.

Asking an AI every time
  • CostSpends tokens on every request.
  • ResultThe same request can come back differently each time.
  • SpeedWaits for a model to think through the task again.
  • EnergyA model runs for every repeat of the task.
  • ProofWhat happened is in a chat, not in an approved version with an audit trail.
A Dittoly Flow
  • CostTokens once, to draft it. Steps that call your apps run as code, with no tokens.
  • ResultSame input, same approved steps, same result. Modules are tested before anyone uses them.
  • SpeedRuns straight through your apps, with no model to wait for.
  • EnergyModel compute only where you keep an AI step on purpose.
  • ProofAn approved version, an audit log, and a count of the tokens you did not spend.

Need judgment in a step? Keep an AI step there, with a pinned model and a cost cap. The Insights page counts the tokens and money each Flow saved by not asking an AI every time.

How it works

From an idea to a governed run, in four steps

  1. 01

    Describe

    Say what should happen across your tools, or upload a plan as a PDF. Dittoly asks about anything unclear before it drafts.

  2. 02

    Draft

    AI drafts the Flow from approved modules. Every step, input and permission is visible, and you can edit it on a canvas.

  3. 03

    Approve

    A person reviews what the Flow can read, write and reach, then approves that exact version. Policies can require more than one approver.

  4. 04

    Run

    The engine runs only the approved version, with each member's own credentials. If you lack access, a teammate runs that one step with theirs.

What you get

Built for teams that automate carefully

Modules

Connectors to your apps, each tested in a sandbox and approved before anyone can use it. Version diffs and drift checks show what changed.

Flows and Super Flows

Automate a process, then connect Flows into bigger ones that run in parallel, with pinned versions and recorded tests shown to the approver.

Error to verified fix

From an alert to a draft pull request with a regression test that fails before the fix and passes after. Java, Kotlin, Python, Go, Rust, C#, Ruby, PHP and TypeScript.

Apps

Turn approved Flows into internal apps for your team, laid out by AI, versioned and approved like everything else, and exportable.

AI with limits

Pinned models, per-run cost caps, team budgets and automatic stops when a control fails. Text a model wrote waits for a sign-off before it can change anything.

Works with your AI assistants

Expose approved Flows to Claude, ChatGPT or Copilot as tools over MCP, signed in as each person, with scoped and expiring agent identities.

Security and governance

Your security team stays in control

Dittoly runs on your servers. Code it builds and tests runs in locked-down containers that cannot see Dittoly, its database or your network, only the hosts each step declares.

  • Single sign-on with OpenID Connect, and SCIM provisioning from your identity provider
  • Hash-chained audit log, streamed to your SIEM, with SOC 2 and ISO 27001 evidence mapping
  • Per-phase network allowlists, inside limits your admins set for each team
  • Credentials encrypted at rest; every member keeps their own
  • Undo for reversible writes, break-glass access and regular access reviews
  • Sandboxes capped and cleaned up automatically, even after a crash

Get started

Run it on your own server

A Linux host with Docker 28 or later. 2 vCPU and 4 GB of memory is the minimum; 4 vCPU and 8 GB is recommended for sandboxed build and fix steps. The server guide covers requirements, single sign-on, HTTPS, upgrades and backups.

# get the code and configure
git clone https://github.com/SudoDevStudio/dittoly.git dittoly
cd dittoly && cp .env.example .env

# start Dittoly, PostgreSQL and RabbitMQ
docker compose up -d --build

# create the first admin
docker compose exec dittoly \
  node dist/src/cli/bootstrap.js admin

Documentation

Everything you need, step by step

Set up your first module and Flow in five steps, then find a short guide for each feature: what it is, when to use it and exactly what to click.

  1. 1Install and sign inOn your computer or on a server, with Docker.
  2. 2Approve a moduleDescribe the app connection; AI drafts it, the sandbox tests it, you approve.
  3. 3Store your credentialsPaste each app token once; it is encrypted and only used by your runs.
  4. 4Create a FlowDescribe the process or build it on the canvas, then approve the version.
  5. 5Run itRun it now, on a schedule, from a URL or from your AI assistant.

Setup

  • Install on your computerRun Dittoly on your own machine. You are the only user, and admin and approver at once.
  • Run it on a server for a teamOne shared Dittoly for your team, with Docker: Dittoly, PostgreSQL and RabbitMQ. People sign in with your single sign-on.
  • Connect your AIThe AI drafts modules and Flows from plain words. It uses your organization's own key, and data goes only to the provider you choose.
  • Store your credentialsEach person stores their own app tokens once. A run you start uses your access and nobody else's.

Modules and Flows

  • Create a moduleA module is how Dittoly talks to one app, for example posting to Slack. It may only reach the hosts it declares.
  • Create a FlowA Flow is a process across your apps: steps that call module actions, ask people, wait or check pages.
  • Build a Flow visuallyBuild or change a Flow by hand on a canvas, without an AI.
  • Connect Flows into a Super FlowA Super Flow runs other approved Flows, in stages. The Flows in one stage run at the same time.
  • Run a FlowRun it by hand, on a schedule, from another system, or for each new item.

AI and apps

  • Use Flows from your AI assistantClaude, ChatGPT, Copilot and other assistants that speak MCP can run your approved Flows as tools, as you, with your access.
  • Build an app for your teamAn app is a set of pages over your team's approved Flows: forms, tables, cards, an inbox and a runs list.
  • Turn a PDF into FlowsUpload a document describing what you want automated. Dittoly breaks it down and drafts the modules and Flows.
  • Keep everything in gitDittoly writes what people approve to a git repository, and changes merged in git come back as versions that wait for approval.
  • Export an app that runs on its ownPackage approved Flows as a TypeScript project with no database of its own and no connection back to Dittoly. Flows that wait for a person keep their runs in a Redis or Postgres you point it at; the rest run in memory.

Admin and safety

  • Set approval rules and policiesRules decide who must approve what, when runs may start and where modules may connect.
  • Audit log and SIEMEvery action is in a hash-chained audit log. Dittoly can stream it, as it happens, to your SIEM.
  • Single sign-on and SCIMPeople sign in with your identity provider, which can also create, update and offboard them, and turn groups into teams.
  • Undo what a run changedReverse what a Flow run or an AI agent changed, after an approver signs off.

Open the guides Full reference

Questions

Frequently asked questions

What is Dittoly?

Dittoly is self-hosted orchestration software. You describe a process across your apps in plain language, AI drafts it as a Flow from approved modules, a person approves that exact version, and the engine runs it with each member's own access.

How is Dittoly different from Claude, Codex or other AI models?

It is a different kind of tool. Claude, Codex and other AI models and assistants answer questions and do tasks when you ask them. Dittoly is orchestration software on your own servers that turns a repeatable process into a Flow: AI drafts it, a person approves that exact version, and the engine runs it the same way every time, with each member's own access and a full audit trail. It works with these models rather than replacing them: Dittoly uses Anthropic, OpenAI or Ollama models to draft and for AI steps, and it can give approved Flows to Claude, ChatGPT or Copilot as tools over MCP.

Does Dittoly use AI tokens every time a Flow runs?

No. AI is used once, to draft the modules and the Flow. After that, steps that call your apps run as code, without a model, so they spend no tokens and give the same result for the same input. Tokens are spent only by AI steps you choose to keep, which have pinned models and cost caps, and the Insights page shows the tokens and money each Flow saved.

Is Dittoly free and open source?

Yes. Dittoly is released under the Apache License 2.0, and you run it on your own infrastructure.

Where does Dittoly run, and where does my data go?

Dittoly runs on your own servers with Docker, alongside PostgreSQL and RabbitMQ. Your data stays on your infrastructure. Code that Dittoly builds and tests runs in locked-down containers that cannot reach Dittoly, its database or your network, only the hosts each step declares.

What do I need to run Dittoly?

A Linux host with Docker Engine 28 or later and the Compose v2 plugin. The minimum is 2 vCPU and 4 GB of memory; 4 vCPU and 8 GB is recommended for sandboxed build and fix steps.

How does Dittoly keep AI-drafted automation safe?

Nothing AI drafts runs until a person approves it. The approver sees what the Flow can read, write and reach, policies can require more than one approver, and the engine runs only the approved version. Text a model wrote waits for a sign-off before any step that changes something; text from an app or a webhook waits wherever it would choose a destination or something to run. An approver can accept a named path in advance, with a written reason, and that acceptance is recorded in the audit log.

Which languages does error-to-fix support?

Java, Kotlin, Python, Go, Rust, C#, Ruby, PHP and TypeScript. A fix arrives as a draft pull request with a regression test that fails before the fix and passes after.

Does Dittoly work with AI assistants like Claude, ChatGPT or Copilot?

Yes. Dittoly can expose approved Flows as tools over the Model Context Protocol (MCP), with each person signed in as themselves and with scoped, expiring agent identities.