We build Agents. You build the Future.
● B2B offer — Meetings & action trackingMeetings: the summary written, the actions tracked
What makes a meeting worthwhile is what comes out of it: clear decisions and actions that move. Your agent produces the summary, extracts the actions with their owner and their deadline, and presents the state of each at the next meeting. Data hosted in France; the AI model is called at its publisher, except with the Sovereign model option. With the 100% Sovereign package, your teams' exchanges stay with you. The chair approves the summary before it circulates and settles the priorities.
Updated on
Fictional simulation — not a client result
Seven actions extracted, each with its owner and its deadline as they were stated.
Two actions have no identified owner: they are flagged rather than assigned.
🔗 Sourced · taken from what was said in the meeting
Each refers back to the meeting where it was decided.
Reprioritising or reassigning an action commits your teams: that call is yours.
✎ Support · state presented, the chair's call
For an organisation established outside the European Union, the law of its own country applies; at its request, we put equivalent safeguards in place (human review before any action, a log of the criteria, information for the people concerned).
A Blue Lemon Agent meeting summary agent writes the minutes, extracts the actions with their owner and their deadline as stated, and presents at the next meeting the state of every open action. An action with no identified owner is flagged rather than assigned. Data hosted in France; the AI model is called at its publisher, except with the Sovereign model option. With the 100% Sovereign package, your teams' exchanges stay with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
These figures describe our offer, not results measured at a client. How large the gain is on your number of forums and meetings tracked is confirmed by a pilot.
Indicative estimate, to be confirmed by a pilot on your real files (general terms of sale, article 4).
One package, one agent
A meeting summary agent (minutes, actions, tracking), installed and operated for you.
Setup + controlled subscription
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Hosting in France, a dedicated and isolated resource
All inclusive, no setup fee
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Hosting in France, managed end to end
On site, you own it
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
What does the agent actually do?
One agent, from the meeting to the follow-up, with a link to a neighbouring agent. The agent carries out by itself what you authorise it to do; you decide what it does on its own and what it submits to you.
Meeting summary
Writes the subjects covered, the decisions taken and the points left open.
Extracting the actions
Picks up each action with its owner and its deadline, as stated.
Tracking through to the next meeting
Presents the state of the open actions and refers back to the original meeting.
Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
In 30 minutes we identify the most relevant agent — without oversizing the project.
See the agent at work
4 everyday situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data. The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds. The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The company in this demonstration
Fictional companyLanvère Ingénierie — engineering consultancy in building services and energy performance, fictional company
- Sector
- Engineering consultancy — building services and energy performance. Work is organised in committees, team check-ins and project reviews, and that is where decisions are taken.
- Headcount
- 150 employees spread across project teams; the agent covers the sessions of the three declared formats — committee, team check-in, project review — and the person chairing is its only counterpart.
- Who is served
- Public and private clients, followed project by project. Internally, what circulates is the approved minutes, two pages long: decisions with their authors, actions with owner and due date, untreated agenda items and how many times they have been carried over.
- Order of magnitude
- 60 meetings covered and 840 actions noted; a typical agenda holds 7 items, 4 of which are dealt with; the 23 sessions that produced neither a decision nor an action lasted 52 minutes on average, with 4 to 11 participants.
- Tools in place
- The session audio, the agenda, the meeting type as the company declared it, and the action tracker with its due dates. The recording is deleted when the minutes are approved and the deletion is logged; no full transcript is produced, because a transcript is a written recording.
- Who decides
- The person chairing the meeting reads, corrects and approves the minutes — it is at that approval that the recording is deleted — and it is they who sign the agenda. The agent attributes decisions and commitments, never an opinion, an objection or a hesitation: the disagreement is recorded, its author is not. An action with no owner is written down exactly as it was said and flagged as such, with nobody designated by default. Dropping a meeting or changing its format remains a company decision, and no participant is measured.
- Improvement points
- Of 840 actions noted, 344 — 41% — do not yet have an owner: 61 were done anyway, the other 283 come back 2.4 times word for word. The share has come down from 41% to 12% in six weeks, simply because the list is read out at the start of the session. 19 agenda items come back from one session to the next without being dealt with, 6 of them carried over more than three times. And 23 meetings out of 60 produced neither a decision nor an action — 17 team check-ins and 6 project reviews; it is the 6 that call for a decision on format, three of their open items having waited two sessions, that is 28 days.
Lanvère Ingénierie works in project teams, and its decisions are taken in committees, team check-ins and project reviews. Across the 60 meetings in scope, 840 actions were noted, and follow-up rested on minutes written by hand, uneven in precision from one session to the next. The agent listens, produces two pages of minutes, attributes decisions and commitments, visibly flags the actions with no owner, and carries at the end of the document the agenda items left untreated with how many times they have been carried over. The person chairing reads, corrects and approves; the recording is deleted at that approval, and it is this rule that lets a team speak in a meeting as it did before the agent. The exchanges that follow cover one morning of monitoring over sixty meetings, then three cases: what carries a name and what does not, what disappears between two sessions, and what the company keeps.
This company, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real client.
Fictional simulation — not a client result
· Forty-one per cent of the actions recorded have no owner. An action with no name is not an action.
· Twenty-three meetings out of sixty produced neither a decision nor an action. I record it, I do not judge it.
· Three of seven agenda items were not reached, and they vanished from the follow-up.
· I am asked for speaking time per participant. I will not produce it. morning-watch_4-flags.pdf41% of actions with no owner · 3 items of 7 vanished
⛓ Source · 60 meetings, 840 actions, agendas
What I record: of 840 actions recorded, 344 have no identified owner. They are phrased impersonally: "check with the supplier", "follow up on this", "this should be clarified".
What becomes of them: of those 344, I find 61 completed — almost always because somebody took them on themselves. The other 283 reappear on average 2.4 times in later minutes, word for word, with nothing having moved.
Why they get written that way: because naming somebody in a meeting means asking them, in front of the others, at a moment when the agenda is pressing. The impersonal avoids that moment. It is a chairing problem, not a drafting one.
What I do: I record the action as it was said, and I mark it "no owner" — visibly, in the minutes, not in an appendix. I assign it to nobody on my own initiative.
What that produced: at the following meeting, the list of ownerless actions is read first. The share fell from 41% to 12% in six weeks, without my having assigned anything. 344-actions_283-reappear.pdfThe impersonal avoids the moment of asking somebody
⛓ Source · 344 ownerless actions, 283 reappearing 2.4 times
What each meeting produces: the decisions taken, with who took them, the actions with their owner and deadline, the agenda items not reached, and disagreements mentioned without being attributed.
Routing follows who can act: an ownerless action is read at the top of the next meeting; an item not reached is explicitly carried to the next agenda, not forgotten; a late action goes to its owner, once; a meeting with no decision is counted, without comment.
With a monthly summary: ownerless actions, actions reappearing unchanged, agenda items never reached, and meetings with no decision.
Three operating rules, and what they give you. The recording goes when the minutes are approved — which is what lets a team speak in the room as it did before the agent, and a meeting where people can still hesitate aloud decides better. What commits carries a name, what searches does not: decisions and commitments always attributed, opinions and objections never — the disagreement is recorded, its author is not. No participant is measured, neither in speaking time nor in actions held: such a counter would make the meeting serve the counter. What I measure instead is already in figures: 344 ownerless actions, 283 reappearing 2.4 times, 19 agenda items never reached — and the share of ownerless actions fell from 41% to 12% in six weeks.
✎ Framework · no recording kept · no opinion attributed · no participant measured
Fictional simulation — not a client result
What I never attribute: an opinion, an objection, a hesitation, a naive question, a position stated and then abandoned during the discussion.
Why it is more than courtesy: somebody who knows their sentences will be written down, named and forwarded stops saying the same things. They stop testing a half-formed idea, they stop changing their position, and they say nothing when unsure. A meeting where you can no longer hesitate aloud produces worse decisions.
What I always attribute, by contrast: decisions and commitments. "Decided to postpone the launch" without who decided is not a decision, it is a rumour. And an action with no owner is not an action.
The line is there, and it is sharp: what commits carries a name; what searches does not.
What I do with disagreements: I mention them — "the decision was taken despite an objection on the timing" — without saying who raised it. The existence of the disagreement is useful information; its author is not. what-carries-a-name.pdfA meeting where you cannot hesitate produces worse decisions
✎ Framework · decisions and commitments named, opinions never
What was said, word for word: “we are going with the condensing option, subject to the inspection body's reply”. The decision is in the present tense, the proviso comes behind it, as an aside, in the ordinary phrasing of a site review — and that is the phrasing I read as a settled decision.
The measurement I publish about myself: across your 60 meetings, 37 sentences carried an explicit proviso. 34 came out with it, 3 without — Tuesday's is one of the three.
What the review produced, and that is exactly its job: the 3 were corrected by the person chairing before the minutes circulated, and no minutes went out with the gap. But a correction that has to be made three times is a correction I can spare you.
What I changed, and it has already been run again over the 60 meetings: a decision carrying a proviso now comes out as a “decision subject to a proviso”, in a section separate from firm decisions, with the proviso itself, who lifts it and the date by which it is lifted. Run again over your 60 meetings, this treatment brings out 37 provisos out of 37.
And what does not move: the person chairing rereads, corrects and approves — they settle the minutes, and the treatment does not replace that reading: it leaves it less to catch. decisions-subject-to-a-proviso_37-across-60-meetings.pdf34 provisos out of 37 carried through, the other 3 caught at review
⛓ Sourced · 60 meetings, 37 sentences carrying an explicit proviso, minutes reread by the chairs
What I do: I use the audio to produce the minutes, the person chairing rereads and corrects them, and the recording is deleted on approval. The deletion is logged; the minutes remain.
Why not keep it "just in case": because a kept recording can be replayed for something other than the minutes. A poorly phrased sentence, a tone, a silence — all of that exists in the audio and does not exist in the minutes, which are a text reread and corrected.
What it changes in the room: it is the one point of this whole arrangement that participants check. A team that knows nothing is kept talks as it did before the agent; a team that doubts it talks as if into a microphone.
What I do if somebody asks not to be recorded: the recording stops, and I take the minutes another way. No meeting requires being captured.
And the minutes replace the transcript, they do not come on top of it: a full transcript is a written recording, and it raises the same problem with the appearance of a working document. What I deliver instead fits on two pages — attributed decisions, actions with owner and deadline, items not reached and how often they have been carried —, reread and corrected by whoever chaired, then approved. That is the document of record, and it is the only one that remains. why-nothing-is-kept.pdfA transcript is a written recording
✎ Framework · audio deleted on approval · no full transcript
Fictional simulation — not a client result
What I record: the agenda had 7 items. 4 were dealt with, 3 were not — for lack of time. The minutes, accurately, contain the 4. The other 3 appear nowhere, and do not return to the next agenda.
What that produces over time: the items that get dropped are always the same — those placed at the end of the agenda, often the substantive subjects. Across 60 meetings, 19 items were never dealt with despite repeated listing, and 6 were carried more than three times.
What I do: the minutes carry a "not reached" section, with the items and how many times they have already been carried. Three lines, at the end.
What that changes: an item carried four times stops being an oversight and becomes a question — should it be handled elsewhere, differently, or dropped? That is a decision, and it was never taken because the question was never put.
What I put on the table, in figures: the 19 items never reached despite repeated listing, each with how many times it was carried, its position on the agenda and the time it would have needed. Six were carried more than three times — those are the ones that call for a decision. The agenda itself is signed by whoever chairs: it is their move, and they make it knowing the figures rather than by guess. 19-items_6-carried-three-times.pdfAn item carried four times stops being an oversight
⛓ Source · 60 meetings, 19 items never dealt with, 6 carried more than three times
What I record: of 60 meetings, 23 produced neither a decision nor an action. They lasted 52 minutes on average, with between 4 and 11 participants.
What that does not say: that they were pointless. An information meeting, a progress check, a catch-up after an absence do not have to produce a decision — and demanding that they do is the surest way to obtain premature ones.
What I do: I give the figure by meeting type as you have declared it — committee, team check-in, project review. A project review with no decision does not mean the same as a team check-in with no decision.
What that showed: of the 23, 17 were recurring team check-ins and 6 project reviews. It is the 6 that prompted discussion, and two of them changed format.
What I put on the table instead of a cost in cumulative hours: for each of the 6 project reviews with no decision, the agenda, the points left open and the sitting where they were finally decided — three of them waited two sittings, that is twenty-eight days. It is that delay that can be fixed, and it is fixed on the format, not by removal. The cost in cumulative hours exists, it is spectacular, and it says nothing about what a meeting produces invisibly — alignment, trust, what people learn from each other. Removing a meeting stays your decision; I give you what it takes to make it, sitting by sitting. 23-meetings_17-team-check-ins.pdfDemanding a decision per meeting produces premature ones
⛓ Source · 23 meetings with no decision, 17 team check-ins, 6 project reviews
Fictional simulation — not a client result
What I deliver to install it: the prior information notice (art. L1222-4), the works council consultation file, a proportionate scope.
What I am told about it, and it is sincere: that it helps balance meetings, spot those who go unheard. The intention is good and the effect is the opposite: the person who went unheard, knowing they are counted, speaks to appear, and what they say then is not what they were thinking.
The measure I propose that serves the same aim: the agenda items actually covered. In a seven-item meeting, 4 covered, 3 postponed without being announced as such. A person who goes unheard is often a person whose item was not reached — and that is fixed by reorganising the agenda, not by watching people.
What stays true in every case: I attribute no opinion. A meeting is a place where people think aloud; minutes are a document that circulates. I attribute decisions and commitments — what commits carries a name, what explores does not. speaking-time_and-the-agenda.pdfWhat opens · the effect opposite to the intention · the measure serving the same aim
⛓ Source · French DPA, art. L1222-4 · 4 items covered of 7, 3 postponed unannounced
What is kept: the decisions and who took them, the actions with owner and deadline, the actions with no owner, the agenda items covered and not covered, and the meetings with no decision or action.
The headline figure: of 840 actions recorded, 344 — 41 % — have no owner. An action with no name is not an action: it is written, reread, ticked at the next meeting, and nobody ever started it. I now raise them at the end of the session, while everybody is still there.
The point minutes never record: what was not said. A seven-item agenda, 4 covered — the other 3 appear nowhere, neither as covered nor as postponed. They will come back in three weeks as if they were new.
What I give without qualifying it: 23 meetings of 60 produced neither a decision nor an action. I do not know whether that is a problem — some meetings exist to inform, and that is legitimate. I give the figure, you know which ones.
The recording: it disappears when the minutes are approved. That is an operating condition — without it, people stop thinking aloud. what-you-keep_meeting-summaries.pdf5 items kept · the recording that disappears on approval
⛓ Source · 344 unowned actions of 840, 3 items of 7 vanished, 23 meetings of 60
Your case is not here? That is exactly what a 30-minute conversation is for. Discuss your project →
The stages of your AI agent project
Audit & scoping
30 minutes to target the use case with the best return.
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
Design
We design the agent and its guardrails.
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
Rollout
Going live and training your team.
Operation
Continuous supervision and improvement.
What does an AI agent bring to your meetings?
A decision taken and an action followed through are two different things. Linking the two from one meeting to the next gives the time spent together its full value.
! The issue
A meeting produces two things: a record and a list of actions. The first calls for drafting, the second for tracking through to the next session. The agent takes on both, staying as close as possible to what was said: owner and deadline are carried over as stated, and an action with no owner is flagged rather than guessed.
✓ Our answer
The chair arrives at the next meeting with the approved summary and the state of every open action. Reprioritising, reassigning or closing an action commits the teams: those calls remain theirs. Data hosted in France on an isolated resource; the AI model is called at its publisher, except with the Sovereign model option. With that option, internal exchanges, often the most sensitive a company has, are entrusted to no third party.
Your teams' exchanges and decisions: sovereignty & compliance
What is said in a meeting is among the most sensitive information a company has. Here is how the architecture of our agents protects it.
Local inference
With the 100% Sovereign package, the agent runs on a machine belonging to your organisation: no recording and no minutes leave the network.
Hosting in France
Otherwise, with the Agility and Serenity packages, a dedicated, isolated resource hosted in France, under French law: data hosted in France; the AI model is called at its publisher, except with the Sovereign model option.
Reduced extraterritorial exposure
For your teams' exchanges and decisions, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your organisation and its forums.
Owner and deadline carried over as stated
An action with no identified owner is flagged rather than assigned; encryption, role-based access and logging of every summary produced.
AI Act: governed deployment
The agent carries out what you have authorised; no action is assigned or closed automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
How much time can a team give back to the substance?
By taking on the drafting and the tracking, the effort shifts towards preparation and decision. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Four guarantees that matter to your meetings
Related resources
Your questions, our answers
Does the agent assign the actions?
Is the summary circulated automatically?
Do the meetings have to be recorded?
Are our teams' exchanges protected?
Can certain meetings be excluded?
How long does it take to deploy this agent?
Other agents for organisation
Let us estimate the potential across your meetings
30 minutes to frame your forums and your tools — hosted in France, supervised, with no commitment.