It reads, checks and drafts.You keep every decision.
AI does the reading, code verifies the numbers, a person approves. Nothing gets a status without your hand on it.
Read against what you agreed, before you open it.
One canonical run per upload. The artist sees a translation of that same result, never a second opinion.
Nothing runs until a file lands. Press run to see the check this invoice actually gets.
A verdict is a suggestion until you approve it. The check writes pass, warn or fail and stops there. A person flips every status, and the AI never approves or pays anyone.
One send. Every artist reads it in their own language.
You write the template once. Every token and conditional block survives untouched, because that part is verified by code.
Restyle a draft to match your venue's voice. The rules forbid adding a fact, a time or a price, and you approve before anything sends.
A long version for your site and a short one for social, drafted from the real lineup. It uses only the facts you gave it, and eventflow never submits it anywhere for you.
Paste a messy timetable. Fix it in plain language.
Rooms as headings, time ranges, artists, in whatever shape the promoter sent it. Nothing reaches the grid until you click create.
Nothing lands until you click create. Every row stays editable before it commits, and you fix what the parse got wrong by saying what is wrong.
Your workspace, over the Model Context Protocol.
You add the connector once and approve exactly what it may do. From then on every call is scoped, audited and revocable.
https://app.event-flow.ai/api/mcp14 read tools
7 write tools
3 send tools
4 more sends, not built yet
who is still not confirmed for Saturday?
1 of 8 set on Ostwerk SATURDAY is not confirmed.
Sends reach a real inbox. The three live send tools email a human through the same relay and ledger as the app, so they land in your communications log with delivery tracking. Every call, read or write, is written to an audit table.
It remembers the corrections your accountant made.
Every approval, override and thumbs-down becomes a short, evidenced lesson for your workspace alone.
Corrections with at least two pieces of evidence behind them. Contradictions are surfaced for you, never silently flipped, and there is a hard cap per workspace so it stays a short list you can read.
The constitution you write outranks anything the system learned. Memory never edits it.
Lessons live in your workspace and are visible to its owner. No model is trained on your data.
AI is a draft layer, not a decision layer.
Seven rules we hold ourselves to, and the reason each of them exists.
Nothing is approved, paid, sent or published by a model. The check proposes, you decide, and the decision is the thing that changes a record.
Language judgement is the model's job. Arithmetic, dates, tokens and totals are checked deterministically, so a confident wrong answer cannot become a number in your ledger.
We do not send it anywhere that keeps it, and no model is trained on it.
Every check shows what it looked at and what it found, row by row, so a verdict you disagree with is one you can argue with.
Each check, translation and connected-assistant call is recorded with what ran and what it cost. Sends land in your communications log with the same delivery tracking as any other message.
AI is metered per workspace against a monthly credit, with a warning at 80% and a hard stop at 100%. Every band includes credit and you can top up.
You can revoke them any time. No scoring or ranking of artists for booking, no profiling of artists or guests, and no AI writing to your financial records.
Less admin, more music.
We set every workspace up ourselves and come back within two working days.