gilb.ai Talk to us

Cases

Company Brain at work.

Weekly business reviews, growth discovery and vessel operations. What we connected, what we built, and what the team got.

WEEKLY BUSINESS REVIEW

A weekly view of what matters—and what holds the business back.

We connected CRM data and calls to a dashboard with scheduled analysis. Every Monday, the system summarized the week, highlighted important changes, and searched for the main business constraint.

≈300
customer contacts per day
≈6,600
contacts per month
≈19,800
contacts per quarter

The challenge

With around 300 customer contacts a day, the challenge was to turn activity into a clear view of the business: what happened, what needed attention, and where the main bottleneck might be.

What we built

We built an integration that regularly fetched CRM data from Kommo and stored it in PostgreSQL for agent workflows. An MCP connector gave Company Brain access to that regularly updated data. Customer calls added conversation context. Hermes powered the agent workflow, OpenViking provided persistent memory, and a dashboard brought the findings together. Scheduled jobs ran recurring tasks, including a weekly review every Monday.

  1. Kommo CRMSource customer and business records
  2. PostgreSQLCRM data regularly fetched and stored
  3. MCP connectorCompany Brain access to regularly updated data
  4. Company BrainAnalysis for the dashboard and scheduled reviews

A dashboard for business review

The dashboard brought the findings together: the week’s summary, important signals, and analysis of the business’s main constraint.

Recurring work, on schedule

Scheduled jobs ran tasks regularly. Every Monday, the system reviewed the previous week, highlighted important developments, and searched for the main business bottleneck.

A weekly focus for the team

The weekly review brought important changes and business constraints together, using CRM records, calls, and agent memory.

“What mattered this week, and what is the main business constraint?”

Discuss a similar deployment

C-LEVEL REPORTING & GROWTH DISCOVERY

A continuous search for growth.

We built an integration that regularly pulled HubSpot data into ClickHouse for the company’s agents. They prepared C-level reports, searched for growth opportunities, and generated hypotheses to test.

C-level reporting

Regular business briefings

Growth opportunities

Recurring analysis of CRM data

Business hypotheses

Automatically generated ideas to test

The challenge

Leadership needed a regular view of the business and a way to spot growth opportunities in its CRM data. The task went beyond reporting what happened: it also called for generating ideas about what to try next.

What we built

We developed a recurring data integration between HubSpot and ClickHouse. It regularly fetched CRM data and saved it in the database, providing an updated analytical foundation for the company’s agent stack. Recurring workflows used the data to prepare executive reports, look for growth opportunities, and generate business hypotheses automatically.

  1. HubSpotSource CRM data
  2. ClickHouseCRM data regularly fetched and stored for agents
  3. Agent stackRecurring analysis and hypothesis generation
  4. C-level briefingsBusiness updates and opportunities to explore

What the agentic stack makes possible

ClickHouse’s open AI stack combines analytical data access, conversational tools, specialist agents, and evaluation. These platform capabilities support workflows like this executive reporting system.

MCP access to analytical data

MCP lets agents connect to ClickHouse and query the data behind a business question. SQL results give analysis a concrete numerical basis.

LibreChat for follow-up questions

LibreChat provides a conversational interface to the data. Leaders can explore a finding and ask for a more detailed breakdown.

Specialist agents and code execution

Specialist agents can share analytical work. A sandboxed Python or JavaScript interpreter can transform query results and produce calculations or visualizations.

Langfuse for quality and operations

Langfuse can trace prompts, tool calls, and responses; evaluate outputs; and monitor model costs and latency.

Regular reports for leadership

The reporting workflow brought business findings into a recurring briefing for C-level leaders. Questions for this kind of review include: Which segments are converting? Where are deals slowing down? What changed since the previous reporting period?

Automatic growth hypotheses

The agents searched the data for opportunities and generated hypotheses for the team to assess. A useful hypothesis connects an observed pattern to a proposed change, a target segment, and a metric to test. For example, a drop in conversion can prompt an experiment with follow-up timing.

From business review to the next experiment

One workflow: review performance, find opportunities, and generate hypotheses to test.

Example hypothesis: “Could a different follow-up sequence improve conversion in a customer segment? Compare it with the current approach before expanding.”

Discuss a similar deployment

VESSEL OPERATIONS & LOADING PLANNING

From captains’ emails to a shared operational picture.

We built an email integration that parsed captains’ reports and updated vessel locations in PostgreSQL. A map showed the latest reported positions. Chat helped the team investigate delays and plan loading.

Updated vessel positions

Incoming reports turned into structured records

A shared fleet map

Latest reported vessel locations in one view

Chat with operational data

Questions about delays and loading plans

The challenge

Captains sent operational updates by email. The team needed to turn those messages into a current view of vessel locations and use that information to understand delays and plan loading.

What we built

We developed an ongoing email integration that processed incoming captains’ reports, extracted vessel updates, and saved them in PostgreSQL. The database supplied both the vessel map and the agents’ working context, so new reports became available for operational questions.

  1. EmailIncoming reports from captains
  2. HermesAgent workflow for processing reports
  3. PostgreSQLStructured vessel records updated from email
  4. Map & chatVessel locations and operational questions

A map grounded in reported positions

The map displayed vessel locations from the processed reports. It gave the team a shared view of where vessels were last reported, with new email updates feeding the underlying records.

LibreChat for operational questions

LibreChat provided the interface for asking questions about the data. The team could investigate where a delay was occurring and discuss how to plan loading around the reported vessel situation.

Hermes for agent workflows

Hermes powered the agent workflows connecting incoming reports and operational analysis. The integration kept supplying new data as captains sent further updates.

OpenViking for memory

OpenViking provided persistent memory for the agents, carrying context forward between conversations. PostgreSQL held the structured vessel data used by the map and chat.

One view for tracking and planning

The team could review vessel positions, investigate delays, and discuss loading plans using the same data.

Example questions: “Where is the delay?” and “How should we plan loading based on the latest vessel reports?”

Discuss a similar deployment

Want something similar?

Tell us which case is closest to your situation and what systems are involved.

All fields except WhatsApp are required.

We use your details only to reply to this request. Privacy notice