segmentstream (Python)

Pipelines your agent can write and you can read.

Connections, ingestion, models, checks, AI evaluation and semantic models in plain Python, in your workspace repo. The platform schedules, runs and monitors it on managed compute.

Start a pilot Join the waitlist

Get started

pipeline/pyproject.toml

# pipeline/pyproject.toml
dependencies = ["segmentstream"]

At a glance

Python, on top of a production-grade orchestrator
Bronze, silver and gold tiers; daily partitions; backfills
Assets, checks, jobs and semantic models in one definitions file
Every connector on the platform uses the same SDK

What it does

Connections

Declare connections once; pipelines, apps and agents share them. OAuth tokens and keys stay in managed secrets.

connections/connections.py

from segmentstream.connections import GoogleAdsConnection, HubSpotConnection

connections = (
    GoogleAdsConnection(key="google_ads"),
    HubSpotConnection(key="hubspot"),
)

Ingestion

Load any API by day, with retries, rate-limit pools and backfills. Re-running a day replaces it exactly.

pipeline/ingestion/hubspot/contacts.py

@s.ingestion(
    tier=s.Tier.Bronze,
    mode=s.IngestionMode.Historical(
        partitioning=s.Partitioning.Daily(column="created_date", start_date="2020-08-13"),
    ),
    connections=("hubspot",),
    schema=CONTACT_SCHEMA,
    primary_key=("contact_id",),
)
def hubspot_contacts(context: s.Context) -> Iterator[s.IngestionRow]:
    yield from contact_rows(context)

Models

Models are dataframe expressions compiled to warehouse SQL, with declared schemas and inputs.

pipeline/models/revenue_summary.py

@s.model(
    tier=s.Tier.Gold,
    inputs={"orders": "silver/orders"},
    schema={"order_date": "!date", "total_revenue": "!float64"},
)
def revenue_summary(orders: ir.Table) -> ir.Table:
    return orders.group_by("order_date").aggregate(
        total_revenue=orders.amount.sum(),
    )

Checks

Checks evaluate a table and record the result next to it, where lineage and agents can see it.

pipeline/checks/orders.py

@s.asset_check(asset=raw_orders)
def positive_order_amounts(raw_orders: ir.Table, warehouse: s.WarehouseResource) -> s.CheckResult:
    invalid = int(warehouse.execute(raw_orders.filter(raw_orders.amount <= 0).count()))
    return s.CheckResult(passed=invalid == 0, metadata={"invalid_order_count": invalid})

AI evaluation

Ask a decision model typed questions about any record and store the probabilities next to it.

pipeline/ingestion/evaluation/lead_scores.py

client = s.evaluation_client("evaluation")
result = client.evaluate(
    state={"job_title": lead.job_title, "company": lead.company},
    questions={"relevant": {
        "type": "boolean",
        "instructions": "Would sales accept this demo request as an opportunity?",
    }},
)

Semantic models

Define dimensions and measures over a table; the platform compiles every query deterministically.

pipeline/semantic_models/advertising.py

advertising = SemanticModel(
    name="advertising",
    source="silver/cost_data",
    default_date_dimension="date",
    dimensions=(Dimension("platform", "data_source_type"),),
    measures=(
        Measure("spend", sum_(column("cost")), unit="currency"),
        Measure("cpc", divide(measure("spend"), measure("clicks"))),
    ),
)

The rest of the surface

App SDK

@segmentstream-cloud/app

Sign-in, sessions, warehouse and metrics for Next.js apps

CLI

segmentstream

Workspaces, deploys, jobs and data from the terminal

Agent Gateway (MCP)

Model Context Protocol endpoint

One controlled endpoint for every agent

Developer overview →

Your company’s context, ready for every agent.

$ segmentstream workspace create acme --region eu

▸ workspace ready · data, objects, metrics and access rules in one place

▸ connect Claude, ChatGPT or Cursor with one connector

▸ tell your agent: “add our billing API as a source”

Start a pilot

Our engineers build the first version on your data in two weeks, at a fixed price.

Or join the waitlist for self-serve:

Pilots are open now. Self-serve opens to the waitlist this fall.