Polari Python SDK

Official Python SDK for the Polari API — semantic news intelligence.

Requirements

• Python 3.10+
• A Polari API key (get one at polariapi.com)

Installation

pip install polari-sdk

Quick start

import asyncio
from polari import PolariClient, ArticleInput

async def main():
    async with PolariClient(api_key="pk_live_...") as client:
        article = ArticleInput(
            title="Fed holds rates steady",
            content="The Federal Reserve held interest rates steady on Wednesday...",
            url="https://reuters.com/fed-rates-2026",
            source="Reuters",
        )

        result = await client.layer0.analyze(article)
        print(f"Quality: {result.quality_score:.3f}  ID: {result.article_id}")

asyncio.run(main())

Authentication

All requests require a Bearer API key. Pass it at client initialization:

client = PolariClient(api_key="pk_live_...")

Or via environment variable:

export POLARI_API_KEY="pk_live_..."

client = PolariClient.from_env()

Client configuration

All layer URLs default to the correct Polari API endpoints — no configuration needed for standard use. Override only if needed:

client = PolariClient(
    api_key="pk_live_...",
    timeout=60,
    max_retries=3,
)

The client is an async context manager. Use async with to ensure the HTTP connection is closed cleanly. Alternatively call await client.close() manually.

---

─────────────────────────────────────────────────────────────────────────────

Namespaces & Quality Gate

Every API key operates in one of three namespace modes. The mode is set when your key is issued — no SDK configuration is required.

Mode	Articles written to	Articles visible	Quality gate
public	Shared pool	Shared pool	Enforced (default floor: 0.53)
private	Your namespace only	Your namespace only	Bypassed
shared	Your namespace only	Shared pool + yours	Bypassed on write

Public (default): Articles enter the shared corpus. During the Layer 0 → Layer 1 handoff, any article whose quality score falls below the gate floor (0.53 by default) is silently dropped and will not appear in Layer 1 or Layer 2 output. This protects the shared pool from low-signal content.

Private: Articles are fully isolated to your key. The quality gate is bypassed entirely — all articles proceed to Layer 1 regardless of score. Private namespace requires a Professional or Enterprise key. To request one, contact support@polariapi.com or select Private Namespace during signup.

Shared: Write-isolated, read-public. Your ingested articles bypass the quality gate and land in your own namespace. Read operations (entities, stories, clusters) span both your namespace and the shared public pool.

Diagnosing dropped articles

If an article passes Layer 0 but never appears in Layer 1 output, the most likely cause is the quality gate. You can check the quality score from the Layer 0 response:

result = await client.layer0.analyze(article)
print(result.quality_score)  # must be >= 0.53 for public namespace keys

If you need to ingest content that consistently scores below the gate floor (internal documents, short-form content, domain-specific text), request a private namespace key.

Layer 0 — Token Intelligence

Layer 0 processes raw articles into quality-scored, embedded representations.

from polari import ArticleInput

article = ArticleInput(
    title="Fed holds rates steady",
    content="The Federal Reserve held interest rates steady...",
    url="https://reuters.com/fed-rates-2026",  # used as deduplication key
    source="Reuters",
)

result = await client.layer0.analyze(article)

result.article_id      # "art_8f7h2k9s"
result.quality_score   # 0.87  (0.0–1.0; articles below 0.53 are filtered)
result.token_count     # 412
result.is_duplicate    # True if this URL was already processed
result.semantic_hash   # content fingerprint for fast similarity matching

Quality score tiers:

Score	Tier
0.8+	Excellent
0.65–0.8	Good
0.5–0.65	Medium
0.3–0.5	Low
<0.3	Noise

Articles below 0.53 are not forwarded to Layer 1 or Layer 2.

Embeddings

By default, embedding is not returned (saves ~1.5KB per article). Pass include_embedding=True to fetch the 384-dimensional vector from ChromaDB:

result = await client.layer0.analyze(article, include_embedding=True)
result.embedding  # List[float], length 384

Batch processing

articles = [ArticleInput(...), ArticleInput(...), ArticleInput(...)]
results = await client.layer0.analyze_batch(articles)
# Submits all at once, polls concurrently — significantly faster than sequential calls

---

Layer 1 — Semantic Analysis

Layer 1 extracts entities, sentence semantics, and sentiment. Requires Layer 0 first.

result = await client.layer1.process(
    article_id=l0.article_id,
    title=article.title,
    content=article.content,
    url=article.url,            # optional
    published_date="2026-05-29T00:00:00",  # optional
)

result.stats.sentence_count    # 12
result.stats.entity_count      # 8
result.entities                # {"PERSON": ["Jerome Powell"], "ORG": ["Federal Reserve"], "GPE": ["Washington"]}
result.locations               # ["Washington"]
result.article_embedding       # 384-dim article-level embedding
result.processed_at            # datetime

Batch processing

results = await client.layer1.process_batch([
    {"article_id": "art_xxx", "title": "...", "content": "..."},
    {"article_id": "art_yyy", "title": "...", "content": "..."},
])

---

Layer 2 — Story Clustering

Layer 2 groups related articles across sources into unified stories. Requires Layer 0 and Layer 1 first.

result = await client.layer2.cluster(article_id)

result.cluster_id         # "clus_9x3k2m8f"
result.confidence         # 0.94
result.is_new             # True if this article formed a new cluster
result.already_clustered  # True if this article was previously clustered

Batch clustering

result = await client.layer2.cluster_batch(
    article_ids=["art_xxx", "art_yyy", "art_zzz"],
    similarity_threshold=0.75,   # default
    time_window_hours=168,       # default (7 days)
)

result.stats.clusters_formed    # 2
result.stats.clustering_rate    # 1.0
result.stats.singleton_articles # 0

for cluster in result.clusters:
    cluster.cluster_id    # "clus_xxx"
    cluster.title         # "Fed holds rates steady"
    cluster.article_count # 3
    cluster.source_count  # 2
    cluster.confidence    # 0.89

Story queries

stories  = await client.layer2.list_stories(limit=20, offset=0)
detail   = await client.layer2.get_story("clus_xxx")
articles = await client.layer2.get_story_articles("clus_xxx")
timeline = await client.layer2.get_story_timeline("clus_xxx")
sources  = await client.layer2.get_story_sources("clus_xxx")
rels     = await client.layer2.get_stories_relationships(["clus_xxx", "clus_yyy"])

---

Layer 3 — Intelligence Graph

Layer 3 maps entity relationships and reveals trends across the information landscape. Requires Professional tier or higher.

# Graph statistics
stats = await client.layer3.get_stats()
stats["entity_relationships"]   # 5204
stats["cluster_relationships"]  # 48405
stats["narrative_threads"]      # 18
stats["trending_entities"]      # 49

# Trending entities
result = await client.layer3.get_trending_entities(min_velocity=2.0, limit=20)
for trend in result["trends"]:
    trend["entity"]    # "Federal Reserve"
    trend["velocity"]  # 7.0
    trend["mentions"]  # 51

# Cluster relationships
rels = await client.layer3.get_cluster_relationships("clus_xxx")

# Trigger full graph rebuild (admin)
await client.layer3.build_graph()

---

Full pipeline example

import asyncio
from polari import PolariClient, ArticleInput

async def main():
    async with PolariClient(api_key="pk_live_...") as client:

        article = ArticleInput(
            title="Fed holds rates steady",
            content="The Federal Reserve held interest rates steady on Wednesday...",
            url="https://reuters.com/fed-rates-2026",
            source="Reuters",
        )

        # Layer 0 — quality scoring
        l0 = await client.layer0.analyze(article)
        if l0.quality_score < 0.53:
            print("Article below quality threshold, skipping")
            return

        # Layer 1 — entity extraction
        l1 = await client.layer1.process(
            article_id=l0.article_id,
            title=article.title,
            content=article.content,
        )
        print(f"Entities: {l1.entities}")

        # Layer 2 — story clustering
        l2 = await client.layer2.cluster(l0.article_id)
        print(f"Cluster: {l2.cluster_id}  Confidence: {l2.confidence:.3f}")

        # Layer 3 — trending entities
        trends = await client.layer3.get_trending_entities(limit=5)
        for t in trends["trends"]:
            print(f"  {t['entity']}: velocity={t['velocity']}")

asyncio.run(main())

---

Error handling

from polari.exceptions import (
    AuthenticationError,   # 401 — invalid or revoked API key
    ValidationError,       # 400/422 — bad request parameters
    RateLimitError,        # 429 — rate limit exceeded
    ServerError,           # 5xx — service error
    NetworkError,          # connection failure
    TimeoutError,          # request timed out
    ProcessingError,       # article processing failed
    RetryExhaustedError,   # all retry attempts failed
)

try:
    result = await client.layer0.analyze(article)
except AuthenticationError:
    print("Invalid API key")
except RateLimitError:
    print("Rate limit hit — back off and retry")
except RetryExhaustedError as e:
    print(f"Failed after retries: {e}")

The SDK retries 429, 5xx, network errors, and timeouts automatically with exponential backoff (default: 3 retries). 401, 400, 422, and 404 are not retried.

---

Metrics and cost tracking

metrics = client.get_metrics()
metrics.total_requests      # 42
metrics.successful_requests # 41
metrics.average_latency     # 0.34  (seconds)

cost = client.get_cost_summary()
cost.total_calls   # 42
cost.total_cost    # 0.063
cost.cost_by_layer # {"layer0": 0.01, "layer1": 0.02, "layer2": 0.01, "layer3": 0.03}

---

Tier access

Tier	Layers	Requests/day	Requests/min
Starter	0, 1, 2	10,000	10
Professional	0, 1, 2, 3	100,000	100
Enterprise	0, 1, 2, 3	Custom	1,000

Layer 3 (Intelligence Graph) and Trends endpoints require Professional tier or higher. Starter tier keys receive 403 on all Layer 3 and Trends requests.

---

Known limitations

• Layer0Result.embedding returns [0.0] * 384 unless include_embedding=True is passed. Embedding retrieval adds latency due to ChromaDB lookup.
• Layer1Result.sentences returns empty for articles already processed by Layer 1 (cached result). Entity counts are still populated.
• Integration tests reuse article URLs — re-running tests returns cached results for already-processed articles. Append a unique suffix to URLs to force fresh processing.

---

Support

• Docs: docs.polariapi.com
• Email: support@polariapi.com
• Homepage: polariapi.com