Understand the system. Explore what happens next.
Obelisk gives engineering teams one shared model of their systems and how they connect. Draw diagrams for any audience, trace dependencies, and run rule-driven simulations, all on infrastructure you control.
Diagrams drift apart. A shared model doesn't have to.
Teams document the same systems in different tools, so the same database gets drawn, named and described several ways. Obelisk starts one level lower. You define the objects and their connections once, and every view draws from them.
- Define it once
- Objects and connections live in the model. A diagram decides how they appear, not whether they exist.
- Make it investigable
- Trace what depends on a service, narrate a flow, run a scenario against the same model.
- Keep the context for change
- Drafts, snapshots, comments and decision records stay attached to the landscape.
- Keep control
- Run it in your own infrastructure. Reach it through the REST API and import and export.
Follow one system through the whole product
Mealwheel runs a food-ordering platform. Here is its ordering area in Obelisk, in five steps. Each step uses the same objects. See every feature
- 1. Define objects
- Add the Order service, the Orders DB and the Checkout queue. Give each an owning team and connect them. They belong to the model, not to any one drawing.
- 2. Build views
- The app team draws the Order service internals. The platform team draws the deployment. Both show the same Orders DB. Taking a box off a diagram does not delete the object. Deleting it is a separate, deliberate act on the model.
- 3. Trace dependencies
- A reviewer asks what relies on the Checkout queue. Obelisk walks the connections upstream and downstream, hop by hop. It answers from what your team has modelled. It does not measure production.
- 4. Simulate
- Put a source on Customer apps, add rules to the Order service and send a steady stream, a spike or a ramp through the model. Every dropped message keeps a trace of the rules it met. The result is only as good as your rules and inputs.
- 5. Propose a change
- Someone suggests a queue between the Order service and the Notification service. It goes in as a draft, with a list of every change and a decision record beside it. Merging changes the model. It does not touch anything that is running.
Your architecture. Your environment.
An architecture model lists every system, every database and every weak spot you have. Obelisk runs on your servers, so that list stays with you. Nothing here needs a vendor's cloud. How self-hosting works
Obelisk sends no telemetry. It has no analytics, no error reporting, no update check and no licence check over the network.
- Docker Compose
- Postgres and the app start together. Database migrations apply when the app starts.
- Sign-in
- SAML and OIDC with your identity provider. Roles are admin, editor and viewer.
- REST API, import and export
- Read and change the model from your own scripts. Export the model as JSON or CSV and diagrams as PNG or SVG.
- Air-gapped
- Runs with no outside network once installed. The offline install bundle is not published yet.
What works today, and what does not yet.
Plain labels, so nobody buys a roadmap by mistake.
| Capability | Status | What it means |
|---|---|---|
| Shared model, diagrams and Landscape Map | Available | Objects and connections defined once, drawn in as many views as your teams need. |
| Dependency analysis and flows | Available | Trace upstream and downstream. Narrate an interaction step by step. |
| Simulation with replay and traces | Available | Sources, rates, payloads and rules. Results follow your rules and inputs. |
| Drafts, snapshots and decision records | Available | Propose changes, keep named states, record why. |
| Docker Compose deployment, REST API, import and export | Available | Self-hosted with Postgres. Documented interfaces for your scripts. |
| Air-gapped deployment | Planned — check status | Runs with no outside network once installed. The offline install bundle is not published yet. |
| Command-line interface | Available | Start the server, run first-time setup and manage the license from a terminal. |
| MCP server and AI agent connection | Available | AI agents can search the model, trace dependencies, add objects and run simulations through the MCP server. |
| Comparison runs and assertions for simulations | Planned — check status | On the roadmap. Not part of the current simulation engine. |
| Helm chart | Planned — check status | Planned. Compose is the supported way to deploy today. |
| Automatic discovery or live sync with production | Not offered | Your team maintains the model. Obelisk does not read or control your running systems. |