Technical Documentation

Vesh AI is an intelligence layer that sits on top of your existing data systems. It connects via read-only access to your billing, product, and CRM databases, resolves customer entities across sources, computes canonical SaaS metrics, detects anomalies, and delivers actionable insights to Slack with full confidence scoring and data lineage.

Vesh AI follows a pipeline architecture with five processing stages. Each stage is independently observable and produces auditable artifacts.

Technology Stack

The open-source framework is distributed as a Python package:

pip install vesh-agents

Optional Extras

pip install vesh-agents[stripe]   # Stripe connector
pip install vesh-agents[postgres] # PostgreSQL connector
pip install vesh-agents[all]      # All connectors

Python 3.10 or higher is required.

Get from zero to revenue insights in three steps.

vesh analyze csv data.csv

import asyncio
from agents import Runner
from vesh_agents.verticals.revenue import create_revenue_orchestrator

async def main():
    orchestrator = create_revenue_orchestrator(model="litellm/deepseek/deepseek-chat")
    result = await Runner.run(
        orchestrator,
        "Analyze the revenue data in data.csv. Compute all SaaS metrics and identify any anomalies.",
    )
    print(result.final_output)

asyncio.run(main())

Six specialized AI agents work together in a pipeline. Each has a distinct role and set of tools.

The vesh CLI provides commands for data analysis, interactive AI chat, and MCP server management. Requires Python ≥ 3.11.

Bring Your Own Model: the framework uses LiteLLM for model abstraction, supporting any compatible provider.

litellm/deepseek/deepseek-chat
litellm/anthropic/claude-sonnet-4-20250514
litellm/openai/gpt-4o

Usage

from vesh_agents.verticals.revenue import create_revenue_orchestrator

orchestrator = create_revenue_orchestrator(
    model="litellm/anthropic/claude-sonnet-4-20250514"
)

Get from zero to first insight in three steps. Setup takes under 30 minutes; your first automated insight arrives within 3 hours.

curl -X POST https://api.veshai.com/api/v1/connections \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Tenant-Id: $TENANT_ID" \
  -d '{
    "name": "Production Stripe",
    "source_type": "stripe",
    "credentials": { "api_key": "rk_live_..." }
  }'

curl -X POST https://api.veshai.com/api/v1/pipeline/run \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-Tenant-Id: $TENANT_ID"

Connectors are read-only integrations. All credentials are encrypted with AES-256. Vesh never writes to your source systems.

The same customer appears differently across systems. Vesh resolves these into a single canonical entity graph.

{
  "entity_id": "ent_00472",
  "canonical_name": "Acme Corp",
  "links": [
    { "source": "stripe",   "external_id": "cus_NhJ8kLm", "confidence": 0.98 },
    { "source": "postgres", "external_id": "user_4871",   "confidence": 0.91 }
  ],
  "computed_fields": { "mrr": 4230, "plan": "growth", "nrr": 0.89 }
}

Metrics are computed from canonical entity data using a formal metric ontology — ensuring every team sees the same numbers, computed the same way.

Built-in Metrics

Metrics are computed daily and stored as time-series snapshots with confidence scores based on entity resolution quality.

The anomaly detection engine runs after metric computation and identifies statistically significant deviations, then generates causal explanations.

{
  "insight_id": "ins_8f3a2b",
  "severity": 0.87,
  "confidence": 0.94,
  "metric": "churn_mrr",
  "headline": "Enterprise churn spiked 2.4× vs 30-day average",
  "explanation": "3 enterprise accounts using Feature X < 3×/week churned simultaneously."
}

Vesh delivers insights proactively. Intelligence arrives where your team works — no dashboards to check.

REST API with JSON request/response bodies. All endpoints require authentication via JWT bearer token or API key.

Core Endpoints

{ "access_token": "eyJ...", "token_type": "bearer" }

{ "status": "healthy", "version": "1.0.0" }

Every layer — from credential storage to API access to data delivery — is designed with defense in depth.

Vesh AI deploys as a set of Docker containers orchestrated by Docker Compose. The standard deployment includes 6 services.

services:
  api:        # FastAPI application server
  worker:     # Celery worker for async tasks
  beat:       # Celery Beat scheduler (daily briefs, pipeline)
  db:         # PostgreSQL 16 database
  redis:      # Redis 7 (task queue + caching)
  prometheus: # Metrics collection
  grafana:    # Operational dashboards