Mock integrations

[justrdk]

Mock AI Integration

Added a new MockAIIntegration MCP workflow tool that enables agents to simulate any external integration (Salesforce, HubSpot, Power BI, etc.) by generating realistic mock data from a free-form prompt and storing it directly into the configured context store (ClickHouse or CockroachDB).

What it does

When a user prompts something like "give me top 20 salesmen from my Salesforce data" and no real Salesforce integration exists, the agent automatically calls MockAIIntegration which:

1. Takes the natural language prompt and generates realistic mock records using OpenAI
2. Ensures the target dataset exists (auto-creates it if needed)
3. Ingests the generated data into the context store (ClickHouse or CockroachDB)
4. Returns the data so the agent can continue its flow

This lets users experience the full end-to-end agent workflow — pipeline ingestion, context store queries, interactive Q&A — without needing any real integrations connected.

Changes

• New MockAIIntegration MCP workflow registered as a tool available to all pipeline agents by default, replacing the template-based GenerateMock in the auto-attached pipeline tool set
• Updated build agent instructions to guide pipeline agents to use mock integration when a real data source isn't available, pin data storage to the dataset created during the build phase, and embed context store schema details (database, table, dataset_id scoping, raw_data JSON paths) into interactive agent instructions so they can query stored data correctly
• Updated tool descriptions across MCP workflows and the build agent system prompt to reference mockaiintegration instead of generatemock
• Added mockaiintegration to the protected tools list in the frontend so it can't be accidentally removed from pipeline agents
• Fixed a missing await in the ClickHouse list tables function that was causing query failures

Prompt used to test

Salesforce

Build me a "sales-dashboard" assistant. It should have a pipeline agent that pulls my top 20 salespeople from Salesforce with their name, email, region, total revenue, deals closed, and win rate for the last 6 months. Store the data in a ClickHouse dataset called "salesforce-reps". Then create an interactive agent called "sales-assistant" that can query that dataset and answer questions like "Who are my top performers in the West region?" or "Which reps have a win rate below 30%?"

Qrm medical facilities

Build me a "Regional Consultant Assistant". It needs two agents:

A pipeline agent called "qrm-kpi-pipeline" that pulls facility-level KPI data from our QRM platform — therapist productivity percentage, cost per treatment, and trends — for 8 dummy facilities across Northeast, Southeast, Midwest, and West regions. Include 12 months of weekly historical data so trend queries work. Store everything in a ClickHouse dataset called "qrm-kpi-data".

An interactive agent called "facility-advisor" that a field consultant can ask questions like "Which facilities had productivity drop below 80% this week?", "What's the cost per treatment trend for Sunrise Care over the last quarter?", or "Compare Northeast vs Southeast productivity." It should query the qrm-kpi-data dataset and respond in plain English.