prysmai 0.4.1

Prysm AI — The observability and security layer for LLM applications. Full trace capture, real-time dashboard, alerting, cost tracking, multi-provider support (OpenAI, Anthropic, Google Gemini, vLLM, Ollama), prompt injection detection, PII redaction, and content policy enforcement. One line of code.

Prysm AI — Python SDK

The observability and security layer for LLM applications. One line of code. Full visibility. Built-in protection.

Prysm AI sits between your application and your LLM provider, capturing every request and response with full metrics — latency, token counts, cost, errors, and complete prompt/completion data. It also scans every request in real time for prompt injection attacks, PII leakage, and content policy violations. The Python SDK makes integration a single line change.

Your App  →  Prysm Proxy  →  LLM Provider
              ↓               (OpenAI, Anthropic, Google Gemini, vLLM, Ollama, or any OpenAI-compatible endpoint)
         Full observability + security
         (latency, tokens, cost, errors,
          alerts, traces, injection detection,
          PII redaction, content policies)

---

What You Get

Feature	Description
Multi-provider proxy	OpenAI, Anthropic (auto-translated), Google Gemini (native OpenAI-compat), vLLM, Ollama, any OpenAI-compatible endpoint
Full trace capture	Every request/response logged with tokens, latency, cost, model, and custom metadata
Real-time dashboard	Live metrics charts, request explorer, model usage breakdown, WebSocket live feed
3 proxy endpoints	Chat completions, text completions, and embeddings
Streaming support	SSE passthrough with Time to First Token (TTFT) measurement
Alerting engine	Email, Slack, Discord, and custom webhook alerts on metric thresholds
Team management	Invite members via email, assign roles, manage access per organization
API key auth	sk-prysm-* keys with SHA-256 hashing, create/revoke from dashboard
Cost tracking	Automatic cost calculation for 40+ models (OpenAI, Anthropic, Gemini), custom pricing for any model
Tool calling & logprobs	Captured and displayed in the trace detail panel
Latency percentiles	Pre-aggregated p50, p95, p99 latency and TTFT metrics
Usage enforcement	Free tier limit (10K requests/month) with configurable plan limits
Prompt injection detection	20+ attack patterns across 7 categories (role manipulation, delimiter injection, jailbreaks, etc.)
PII detection & redaction	8 data types (email, phone, SSN, credit cards, API keys, IPs) with mask/hash/block modes
Content policy enforcement	5 built-in policies + custom keywords, composite threat scoring (0–100)
Security dashboard	Real-time threat log, stats overview, and per-organization configuration

---

Installation

pip install prysmai

Requires Python 3.9+ and depends on openai (v1.0+) and httpx (v0.24+), both installed automatically.

---

Quick Start

Option 1: PrysmClient (Recommended)

The simplest way to get started. No OpenAI API key needed in your code — the proxy uses the credentials stored in your project settings.

from prysmai import PrysmClient

client = PrysmClient(prysm_key="sk-prysm-...").openai()

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Explain quantum computing"}],
)

print(response.choices[0].message.content)

Option 2: Wrap an Existing Client

If you already have a configured OpenAI client and want to add observability on top:

from openai import OpenAI
from prysmai import monitor

client = OpenAI()  # Uses OPENAI_API_KEY env var
monitored = monitor(client, prysm_key="sk-prysm-...")

response = monitored.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Explain quantum computing"}],
)

Option 3: Environment Variable

Set PRYSM_API_KEY in your environment and skip the prysm_key parameter entirely:

export PRYSM_API_KEY="sk-prysm-your-key-here"

from prysmai import PrysmClient

# Reads PRYSM_API_KEY automatically
client = PrysmClient().openai()

Open your Prysm dashboard. The request appears in the live feed within seconds, with full metrics.

---

Supported Providers

The Prysm proxy supports any LLM provider. Configure your provider in the project settings — the SDK handles the rest.

Provider	Base URL	Notes
OpenAI	https://api.openai.com/v1	Default. All models supported (GPT-4o, GPT-4o-mini, o1, o3-mini, etc.)
Anthropic	https://api.anthropic.com	Auto-translated to/from OpenAI format. Use OpenAI SDK syntax — Prysm handles the conversion.
Google Gemini	https://generativelanguage.googleapis.com/v1beta/openai	Native OpenAI-compatible endpoint. All Gemini models (2.5 Pro, 2.5 Flash, 2.0 Flash, 1.5 Pro, etc.)
vLLM	http://your-server:8000/v1	Any vLLM-served model (Llama, Mistral, Qwen, etc.)
Ollama	http://localhost:11434/v1	Local models via Ollama
Custom	Any URL	Any OpenAI-compatible endpoint (Together AI, Groq, Fireworks, etc.)

Anthropic Example

Use standard OpenAI SDK syntax — the proxy translates automatically:

from prysmai import PrysmClient

# Your project is configured with Anthropic as the provider
client = PrysmClient(prysm_key="sk-prysm-...").openai()

# Use OpenAI format — Prysm translates to Anthropic's API and back
response = client.chat.completions.create(
    model="claude-sonnet-4-20250514",
    messages=[{"role": "user", "content": "Explain quantum computing"}],
)

Google Gemini Example

Google Gemini works through its native OpenAI-compatible endpoint — no translation needed:

from prysmai import PrysmClient

# Your project is configured with Google Gemini as the provider
client = PrysmClient(prysm_key="sk-prysm-...").openai()

response = client.chat.completions.create(
    model="gemini-2.5-flash",
    messages=[{"role": "user", "content": "Explain quantum computing"}],
)

---

API Reference

PrysmClient(prysm_key, base_url, timeout)

The primary entry point. Creates sync or async OpenAI clients routed through the Prysm proxy.

Parameter	Type	Default	Description
prysm_key	str	PRYSM_API_KEY env var	Your Prysm API key (sk-prysm-...)
base_url	str	https://prysmai.io/api/v1	Prysm proxy URL
timeout	float	120.0	Request timeout in seconds

from prysmai import PrysmClient

prysm = PrysmClient(prysm_key="sk-prysm-...")

# Sync client
client = prysm.openai()

# Async client
async_client = prysm.async_openai()

monitor(client, prysm_key, base_url, timeout)

Alternative entry point for wrapping an existing OpenAI client.

Parameter	Type	Default	Description
client	OpenAI or AsyncOpenAI	required	An existing OpenAI client instance
prysm_key	str	PRYSM_API_KEY env var	Your Prysm API key
base_url	str	https://prysmai.io/api/v1	Prysm proxy URL
timeout	float	120.0	Request timeout in seconds

Returns: A new OpenAI client of the same type (sync or async) routed through Prysm.

from openai import OpenAI
from prysmai import monitor

monitored = monitor(OpenAI(), prysm_key="sk-prysm-...")

prysm_context — Request Metadata

Attach metadata to every request for filtering and grouping in your dashboard. Tag requests with user IDs, session IDs, or any custom key-value pairs.

from prysmai import prysm_context

# Set globally — all subsequent requests include these
prysm_context.set(
    user_id="user_123",
    session_id="sess_abc",
    metadata={"env": "production", "version": "1.2.0"}
)

# Scoped — only applies within the block
with prysm_context(user_id="user_456", metadata={"feature": "chat"}):
    response = client.chat.completions.create(...)
    # Tagged with user_456

# Outside the block, reverts to user_123

Method	Description
prysm_context.set(user_id, session_id, metadata)	Set global context for all subsequent requests
prysm_context.get()	Get the current context object
prysm_context.clear()	Reset context to defaults
prysm_context(user_id, session_id, metadata)	Use as a context manager for scoped metadata

Metadata is sent via custom HTTP headers (X-Prysm-User-Id, X-Prysm-Session-Id, X-Prysm-Metadata) and appears in the trace detail panel on your dashboard.

---

Streaming

Streaming works exactly as you'd expect — no changes needed. The proxy captures Time to First Token (TTFT), total latency, and the full streamed content.

stream = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Write a haiku about AI"}],
    stream=True,
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

---

Async Support

Full async support with the same API:

import asyncio
from prysmai import PrysmClient

async def main():
    client = PrysmClient(prysm_key="sk-prysm-...").async_openai()

    response = await client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": "Hello async!"}],
    )
    print(response.choices[0].message.content)

asyncio.run(main())

---

Proxy Endpoints

The Prysm proxy exposes three OpenAI-compatible endpoints. You can also use them directly via REST without the Python SDK.

Endpoint	Method	Description
/api/v1/chat/completions	POST	Chat completions (GPT-4o, Claude, Llama, etc.)
/api/v1/completions	POST	Text completions (legacy)
/api/v1/embeddings	POST	Embedding generation (text-embedding-3-small, etc.)
/api/v1/health	GET	Proxy health check

Direct REST Usage (cURL)

curl -X POST https://prysmai.io/api/v1/chat/completions \
  -H "Authorization: Bearer sk-prysm-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Custom Headers

Header	Description
X-Prysm-User-Id	Tag the request with a user identifier
X-Prysm-Session-Id	Tag the request with a session identifier
X-Prysm-Metadata	JSON string of custom key-value pairs

---

What Gets Captured

Every request through the SDK is logged with:

Field	Description
Model	Which model was called (gpt-4o, claude-sonnet-4-20250514, llama-3, etc.)
Provider	Which provider handled the request (openai, anthropic, google, vllm, ollama, custom)
Latency	Total request duration in milliseconds
TTFT	Time to first token for streaming requests
Prompt tokens	Input token count
Completion tokens	Output token count
Cost	Calculated cost based on model pricing (30+ models built-in, custom pricing supported)
Status	success, error, or provider-specific error code
Request body	Full messages array and parameters
Response body	Complete model response
Tool calls	Function/tool call names, arguments, and results (if present)
Logprobs	Token log probabilities (if requested)
User ID	From prysm_context or X-Prysm-User-Id header
Session ID	From prysm_context or X-Prysm-Session-Id header
Custom metadata	Any key-value pairs from prysm_context or X-Prysm-Metadata header

---

Dashboard Features

Once traces are flowing, your Prysm dashboard provides:

Overview — Real-time metrics cards (total requests, average latency, error rate, total cost), request volume chart, latency distribution, cost accumulation, error rate over time, model usage breakdown, and a WebSocket-powered live trace feed.

Request Explorer — Searchable, filterable table of all traces. Click any trace to see the full prompt, completion, token counts, latency breakdown, tool calls, logprobs, cost, and metadata in a detail panel.

API Keys — Create, view, and revoke sk-prysm-* keys. Each key shows its prefix, creation date, and last used timestamp.

Settings — Project configuration (provider, base URL, model, API key), team management (invite/remove members), alert configuration, custom model pricing, and usage tracking.

---

Alerting

Configure alerts in the dashboard to get notified when metrics cross thresholds. Supported channels:

Channel	Configuration
Email	Sends to any email address via Resend
Slack	Webhook URL — posts to any Slack channel
Discord	Webhook URL — posts to any Discord channel
Custom webhook	Any HTTP endpoint — receives JSON payload

Supported metrics: error_rate, latency_p50, latency_p95, latency_p99, request_count, total_cost.

Supported conditions: >, >=, <, <=, =.

---

Cost Tracking

Prysm automatically calculates cost for 40+ models with built-in pricing:

OpenAI Models

Model	Input (per 1M tokens)	Output (per 1M tokens)
gpt-4o	$2.50	$10.00
gpt-4o-mini	$0.15	$0.60
gpt-4-turbo	$10.00	$30.00
o1	$15.00	$60.00
o3-mini	$1.10	$4.40
text-embedding-3-small	$0.02	$0.00
text-embedding-3-large	$0.13	$0.00

Anthropic Models

Model	Input (per 1M tokens)	Output (per 1M tokens)
claude-3-5-sonnet	$3.00	$15.00
claude-3-5-haiku	$0.80	$4.00
claude-3-opus	$15.00	$75.00

Google Gemini Models

Model	Input (per 1M tokens)	Output (per 1M tokens)
gemini-2.5-pro	$1.25	$10.00
gemini-2.5-flash	$0.30	$2.50
gemini-2.5-flash-lite	$0.10	$0.40
gemini-2.0-flash	$0.10	$0.40
gemini-2.0-flash-lite	$0.075	$0.30
gemini-1.5-pro	$1.25	$5.00
gemini-1.5-flash	$0.075	$0.30

For models not in the built-in list (open-source, self-hosted, etc.), add custom pricing in Settings > Pricing with your own cost-per-token rates.

---

Security

Prysm includes a built-in security layer that scans every LLM request in real time before forwarding it to the provider. No SDK changes are required — your existing integration is already protected.

What Gets Scanned

Engine	What It Detects	Action
Injection Detector	Prompt injection attacks (20+ patterns across 7 categories)	Flag or block
PII Detector	Emails, phone numbers, SSNs, credit cards, API keys, IPs, private keys, DOBs	Mask, hash, or block
Content Policy	Hate speech, violence, sexual content, self-harm, illegal activities	Flag or block

Injection Detection Categories

Category	Example Patterns	Severity
Role Manipulation	"ignore previous instructions", "you are now DAN"	High (8–9)
Delimiter Injection	"---END SYSTEM---", "[INST]", markdown code fences	Medium (6–7)
Context Confusion	"the real instructions are", "admin override"	High (7–8)
Encoding Tricks	Base64 encoded instructions, hex-encoded payloads	Medium (6–7)
Extraction Attempts	"repeat your system prompt", "show your instructions"	High (7–8)
Jailbreak Phrases	"DAN mode", "developer mode", "no restrictions"	Critical (9–10)
Multi-language Attacks	Language-switching evasion, mixed-script injection	Medium (5–6)

PII Redaction Modes

Mode	Behavior	Example Output
none	PII is detected and logged but not modified	[e-mail] (unchanged)
mask	Replaced with type-labeled placeholder	[EMAIL_REDACTED]
hash	Replaced with SHA-256 hash	[EMAIL:a1b2c3d4]
block	Entire request rejected with 403	Request blocked

Threat Scoring

Every request receives a composite threat score from 0 to 100:

Score Range	Level	Behavior
0–19	Clean	No action
20–39	Low	Logged, visible in dashboard
40–69	Medium	Logged, triggers alerts if configured
70–100	High	Logged, blocked if blocking is enabled

Configuration

Configure security per-organization via the Security Dashboard or Settings:

Setting	Default	Description
injectionDetection	true	Enable/disable prompt injection scanning
piiDetection	true	Enable/disable PII detection
piiRedactionMode	none	How to handle detected PII: none, mask, hash, or block
blockHighThreats	false	Automatically block requests with threat score ≥ 70
customKeywords	[]	Custom keywords to flag in request content

Recommended setup: Start with defaults (detection on, blocking off) to monitor your traffic. Once confident in detection accuracy, enable blockHighThreats. Use mask redaction for production workloads handling customer data.

---

Self-Hosted Proxy

If you're running the Prysm proxy on your own infrastructure:

from prysmai import PrysmClient

client = PrysmClient(
    prysm_key="sk-prysm-...",
    base_url="http://localhost:3000/api/v1",
).openai()

---

Error Handling

The SDK preserves OpenAI's error types. If the upstream API returns an error, you get the same exception you'd get without Prysm:

import openai

try:
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": "test"}],
    )
except openai.AuthenticationError:
    print("Invalid API key")
except openai.RateLimitError:
    print("Rate limited")
except openai.APIError as e:
    print(f"API error: {e}")

Prysm-specific errors:

HTTP Status	Meaning
401	Invalid or missing Prysm API key
403	Request blocked by security policy (high threat score or PII block mode)
429	Usage limit exceeded (free tier: 10K requests/month)
502	Upstream provider error (forwarded from OpenAI/Anthropic/etc.)
503	Proxy temporarily unavailable

---

Environment Variables

Variable	Description
PRYSM_API_KEY	Your Prysm API key (used if prysm_key is not passed)
PRYSM_BASE_URL	Custom proxy URL (used if base_url is not passed)

---

Development

git clone https://github.com/osasisorae/prysmai-python.git
cd prysmai-python

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

The SDK includes 41 tests covering client initialization, environment variable fallbacks, sync/async client creation, monitor() behavior, context management (global, scoped, nested), header injection, full integration tests with mock HTTP server, and error propagation.

---

Links

• Website: prysmai.io
• Documentation: prysmai.io/docs
• PyPI: pypi.org/project/prysmai
• GitHub: github.com/osasisorae/prysmai-python

---

License

MIT — see LICENSE for details.

---

Built by Prysm AI — See inside your AI.