GETTING STARTED WITH TALANTON

Quickstart in 30 Seconds

Talanton runs 100% locally on your machine or server. It has zero external dependencies on Redis, Postgres, or SaaS telemetry servers.

1. Installation via PyPI

pip install talanton-py

2. Verify Terminal CLI

$ talanton count "Hello production world" --model gpt-4o
{"model": "gpt-4o", "tokens": 4, "chat_overhead": 7, "total_input_tokens": 11}

Core Architecture: Blazing Fast Local Storage

Talanton embeds an ultra-high performance SQLite database directly on your computer. Every token count and cost record is saved privately in milliseconds without ever slowing down your app.

Tracking Cost Events with Python SDK

from talanton.tracker import CostTracker

tracker = CostTracker()

# Record an LLM inference call
event = tracker.record_call(
    model="gpt-4o",
    prompt_tokens=450,
    completion_tokens=180,
    team="search-team",
    project="agent-orchestrator",
    latency_ms=312.4
)

print(f"Recorded event: {event.id} | Cost: ${event.cost:.6f}")

BudgetGuard: Hard Caps & Soft Alerts

Unbounded autonomous agents can run into recursive loops, generating hundreds of dollars in API spend within minutes. BudgetGuard intercepts calls before sending them to the provider, halting requests in sub-2ms if thresholds are breached.

from talanton.guardrails import BudgetGuard, HardBudgetExceededError

guard = BudgetGuard(
    hard_limit_usd=50.00,      # Absolute stop: raises error
    soft_limit_usd=35.00,      # Warning alert trigger
    team_quotas={"agents": 20.00, "evals": 15.00}
)

# Intercept before making LLM invocation
try:
    guard.check_and_charge(
        estimated_cost=0.045,
        team="agents"
    )
    # Proceed to provider call...
except HardBudgetExceededError as e:
    print(f"REQUEST HALTED: {e}")

OpenAI, LiteLLM & LangChain Wrappers

Talanton wraps existing provider SDKs with a single line of code, automatically extracting tokens, computing exact dollar spend, and enforcing guardrails.

OpenAI Official Python SDK Wrapper

from openai import OpenAI
from talanton.integrations.openai_wrapper import wrap_openai

# Initialize standard OpenAI client
client = OpenAI()

# Wrap with Talanton observability
client = wrap_openai(client, team="production-backend")

# Normal OpenAI calls are now automatically measured & recorded!
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Analyze quarterly retention metrics."}]
)

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

Terminal CLI Commands

Talanton provides an ergonomic CLI tool for CI/CD pipelines, pre-commit validation, and terminal introspection.

Production Benchmarks & Latency SLAs

Every commit is validated against industry-grade performance SLAs. Talanton guarantees sub-2ms guardrail overhead and high-concurrency multi-threaded writes without database locking errors.

Local-First Architectural Guarantees

Unlike SaaS observability platforms that stream your prompts and completions to third-party cloud infrastructure, Talanton stores everything on your sovereign machine.