palqee-prisma-client 0.4.1

Prisma AI platform client for validations, datasets, and jobs

Typed async Python client for the Palqee Prisma oversight intelligence platform.

---

What is this?

palqee-prisma-client is the low-level HTTP client for Palqee Prisma — the oversight intelligence platform that replaces manual review with automated, regulatory-grade validation across human and AI workflows. Use it to drive Prisma resources directly from Python: projects, runs, datasets, jobs, analytics, and validations.

Most applications should reach for palqee-prisma-ai — the high-level SDK that wires OpenTelemetry and validators into your stack with one call. Pick palqee-prisma-client when you need:

• Direct platform control — backfill historical data, orchestrate batch validation jobs, build admin tooling.
• Custom integrations — power BI dashboards, custom UIs, internal compliance tools, or CLI utilities.
• Async-first ergonomics — httpx-based, fully typed via Pydantic, designed for high-throughput pipelines.

Installation

pip install palqee-prisma-client

Imports as palqee_prisma_client:

from palqee_prisma_client import PrismaClient

Requires Python 3.10+.

Quickstart

import asyncio
import os
from palqee_prisma_client import PrismaClient

async def main():
    async with PrismaClient(
        api_key=os.environ["PALQEE_PRISMA_API_KEY"],
        workspace_id=os.environ["PALQEE_PRISMA_WORKSPACE_ID"],
        base_url=os.environ["PALQEE_PRISMA_BASE_URL"],
    ) as client:
        project = await client.projects.get_or_create("complaints-review")
        run = await client.runs.create(project.id)
        print(f"Created run {run.id} under project {project.id}")

asyncio.run(main())

With env vars set, the constructor collapses to PrismaClient().

Resources

The client exposes one resource per Prisma domain object. Every method is async and returns a typed Pydantic model.

Projects

projects = await client.projects.list()                       # Page[Project]
project  = await client.projects.get(project_id)
project  = await client.projects.update(project_id, name="new-name")
stats    = await client.projects.stats(project_id)            # total_runs, total_traces, ...
await client.projects.delete(project_id)

Runs

runs    = await client.runs.list(project_id, status="completed", limit=50)
run     = await client.runs.get(run_id)
run     = await client.runs.update(run_id, status="completed")
metrics = await client.runs.metrics(run_id)                   # validator scores per run
await client.runs.delete(run_id)

Datasets

datasets = await client.datasets.list(project_id)
dataset  = await client.datasets.create(project_id, name="golden-set", items=[...])
items    = await client.datasets.items(dataset.id)

Jobs

job   = await client.jobs.create(project_id, dataset_id=dataset.id, validators=["correctness"])
job   = await client.jobs.get(job.id)
jobs  = await client.jobs.list(project_id, status="running")

Validations

# Trigger validators on a run
result = await client.evaluations.run(run_id, evaluators=["correctness", "hallucination"])
print(result.status, result.total_evaluated)

# Project-wide validation summary
summary = await client.evaluations.summary(project_id, period="7d")
print(summary.avg_score, summary.scores_by_evaluator)

# List validators configured for a project
validators = await client.evaluations.evaluators(project_id)

Note: the API surface uses client.evaluations.* for backwards compatibility. The platform now describes this domain as validations — both terms refer to the same resource.

Analytics

result = await client.analytics.query(
    dimensions=["model_name"],
    measures=["trace_count", "avg_latency_ms"],
    filters=[{"field": "project_id", "op": "eq", "value": project_id}],
)
overview = await client.analytics.overview(project_id, period="7d")
costs    = await client.analytics.costs(project_id, period="30d")
export   = await client.analytics.export("csv", project_id=project_id)

Pagination

List endpoints return a Page[T]:

page = await client.projects.list(limit=20, offset=0)
print(page.total, page.has_more)
for project in page.items:
    ...

Configuration

Parameter	Environment variable	Description
api_key	PALQEE_PRISMA_API_KEY	Prisma API key (pq_prisma_sk_...)
workspace_id	PALQEE_PRISMA_WORKSPACE_ID	Workspace UUID
base_url	PALQEE_PRISMA_BASE_URL	Base URL of your Prisma deployment
timeout	—	Request timeout in seconds (default 30)
max_retries	—	Retries for transient errors (default 2)

Full reference: docs.palqee.com/docs/overview.

Error handling

All exceptions inherit from PrismaError. Catch the specific class you care about, fall through to the parent:

from palqee_prisma_client import PrismaNotFoundError, PrismaAPIError, PrismaError

try:
    project = await client.projects.get(project_id)
except PrismaNotFoundError:
    ...
except PrismaAPIError as e:
    print(f"API error {e.status_code}: {e.message}")
except PrismaError:
    ...

Exception	HTTP status	Meaning
PrismaAuthenticationError	—	Invalid or missing API key (client-side)
PrismaAuthorizationError	401, 403	Server rejected credentials
PrismaNotFoundError	404	Resource does not exist
PrismaConflictError	409	Conflict with existing state
PrismaUnprocessableEntityError	422	Server-side validation failure
PrismaAPIError	other	Any other HTTP error
PrismaTimeoutError	—	Request timed out
PrismaNetworkError	—	Connection or transport failure
PrismaValidationError	—	Client-side input validation failure

When to use this vs palqee-prisma-ai

Need	Reach for
Instrument an app once, ship traces and run validators	palqee-prisma-ai
Drive Prisma resources directly (projects, runs, jobs, datasets)	palqee-prisma-client (this package)
Both	Install both — the SDK uses this client under the hood

Links

• Documentation: docs.palqee.com/docs/overview
• High-level SDK: palqee-prisma-ai
• Tracing wrapper: palqee-prisma-otel
• Website: palqee.com
• Support: [e-mail]

License

MIT © Palqee