running-example.csv in 0.4 s
ALL LINEWORK DRAWN FROM YOUR EVENT LOG. NOTHING DECORATIVE.
FlowMiner reads your event log and draws the process as it happened — every path, every rework loop, every queue where cases sit and wait. Filter to the variants that hurt, rank the bottlenecks, check reality against the model, and ask the built-in AI analyst why — every number it answers with is computed from your log, never improvised.
And it’s open: MIT-licensed and free to self-host on your own
hardware — one docker compose up, no per-seat licenses,
no telemetry, your data never leaves your machines. A managed cloud
exists for teams who’d rather skip the server.
The full analyst’s toolkit — discovery, variants, bottlenecks, conformance, object-centric — in one place, on one upload, with no re-ingestion between views.
Drop in a CSV, XES, or OCEL export. The column mapper auto-detects case, activity, and timestamp; four mining algorithms run in parallel; and you get an interactive map you can filter by frequency or performance — in seconds, not in a workshop.
Bottlenecks ranked by where cases actually wait — not just where activities are slow. Rework loops, ping-pong between teams, temporal profiles, and the social network of handoffs, each one click from the map.
Token-replay conformance against a discovered or imported model, log-skeleton checks, and full OCEL 2.0 — orders, items, and invoices analysed as the separate objects they are, not flattened into one artificial case ID.
Ask the chat panel “why are March cases slower?” and it runs the actual analyses against your log, shows you the charts it fetched, and answers with numbers you can trace. Works with a local model when the data is sensitive.
No Python required, no query language to learn. (When you do want to script it, there’s a documented REST API and an MCP server — but that’s a different sheet.)
Most “AI in process mining” is a chatbot guessing around your schema. FlowMiner’s assistant runs the real analyses — you see which one it ran, the chart it got back, and every number in its answer traces to your data.
Show me the top 3 bottlenecks and tell me which one is worst.
get_bottlenecks(event_log_id="ec7a…", limit=3)The slowest step is examine at 2.5 d average, running across 5 of 6 cases. check ticket is second at 1.2 d over 9 occurrences. decide trails at 1.2 d across 6 occurrences.
The worst offender is examine: it touches nearly every case and consumes roughly half of the end-to-end duration. Focus optimisation there first.
If you’re weighing this against a Celonis quote or another year of Disco licenses — every row below was checked against current product docs, community reviews, and public pricing. Last verified Q2 2026.
| Criterion | FlowMiner | Celonis | Apromore | Fluxicon Disco | MS Process Mining |
|---|---|---|---|---|---|
| Starting price | $0 | ~$100k / yr | ~$10k / yr | ~€1k once-off | M365 bundle |
| License | MIT · open source | Proprietary | LGPL + commercial | Proprietary | Microsoft EULA |
| Deployment | Self-hosted · SaaS | SaaS only | Self or SaaS | Desktop app | Azure only |
| Inductive / Alpha / Heuristic | All three | All three | All three | Inductive only | DFG only |
| Variant explorer | Graph-isomorphism | Flattened | Flattened | Flattened | Flattened |
| OCEL 2.0 native | Yes | Yes | Yes | No | Preview |
| AI chat (grounded) | Tool-use loop | Process Copilot | — | — | Preview |
| Works with local LLMs | Ollama-ready | No | — | — | Azure OpenAI only |
| REST API + MCP for agents | Both · documented | Paid tier | API only | — | API only |
| Telemetry / phone-home | None | Yes | Yes | Anonymous | Yes |
| Data never leaves your infra | Guaranteed | — | If self-hosted | Yes (desktop) | — |
The whole platform — every analysis, the AI assistant, the API — MIT-licensed with no feature gates and no trial clock. If you can run Docker, you can run FlowMiner: three commands, about forty seconds, even on the laptop you’re reading this on.
$ git clone https://github.com/flowminer-stack/flowminer $ cd flowminer && cp .env.example .env $ docker compose up -d ✓ Pulled 4 images (postgres · redis · python · node) ✓ Built 2 images (backend · frontend) ✓ 6 services started, 6 healthy ~40s $ open http://localhost:3000 $ █
Self-hosting is free forever and always will be. But if you’d rather analyse processes than babysit a server — or IT is six tickets away — we’ll run it for you, patched and backed up, with Ask AI working out of the box.
Shared FlowMiner instance, row-level isolation by workspace, provisioned in under a minute. For the team that wants FlowMiner today and can live on a shared box.
Everything in Basic, but on its own backend, Postgres, and Redis. The isolation tier — for teams whose compliance team has questions about multi-tenancy.
Every knob is on the table. Region, credits, SSO, contracts, SLA, and an optional private AI model we deploy just for you so prompts never leave your environment.
Managed tiers bundle AI credits so Ask AI works the moment you sign in; self-hosters bring their own LLM key and skip credits entirely. Otherwise: same source code, same features, same MIT license. The only thing that changes is who gets paged when Postgres runs out of disk at 02:00.
“Your process data is your process data. Nobody — not a vendor, not a consulting firm, not an AI provider — needs it more than you do. So why is the tooling for reading it closed?”
— REASON №1 FLOWMINER EXISTS
Click around the demo, or run it on your own machine before lunch. FlowMiner is MIT-licensed, built in the open, and will stay that way. If it’s useful, star it. If something is broken, file an issue.