Govern AI agents
at runtime
A control plane that sits between your agents and the real world — deciding what runs, recording every step, and showing you what happened, why, and what it cost.
drag to set alice's limit — the control plane decides in real time
Every action passes through the control plane
Not a pile of features — a pipeline. A single agent request is intercepted, validated at each checkpoint, and either allowed through or stopped cold.
01Runtime Guardrails
ENABLEDThe kill switch and configured limits are evaluated first. Anything outside your boundaries is denied before it ever reaches a model or tool.
02Concurrency Control
ENABLEDCaps how many runs execute at once. Prevents resource exhaustion and keeps capacity predictable under load.
03Per-User Limits
ENABLEDEach SDK user carries its own daily budget. Users move NORMAL → WARNING → BLOCKED as they approach the line.
04Cost & Budget Check
ENABLEDToken cost is computed per call and checked against workspace and per-user daily budgets before the action proceeds.
05Complete Observability
ENABLEDEvery model call, tool use, and decision is recorded with full context — what ran, why it was allowed, and what it cost.
Dashboard & Analytics
Every checkpoint above surfaces here — runs in real time, step timelines, usage by model and user, and blocked-user state. The whole path, observable.
A gateway on the wire
The control plane sits inline between your agent and everything it touches. Every request is inspected before it's forwarded — or dropped.
{
"user": "alice",
"model": "gpt-4o",
"action": "model_call"
}click the kill switch ↑
Is this allowed?
Guardrails evaluate in real time. Actions outside your configured limits are denied automatically, before they run.
What happened?
Every action is logged with full context. Trace decisions, debug issues, and audit agent behavior end to end.
What did it cost?
Track cost across models, tools, and APIs. Set budgets, watch spend, and stop runaway charges before they land.
Wrap your agent in three lines
Drop the SDK around your existing code. Every model call and tool use is governed and recorded — no rewrite required.
from pikarc import AsyncPikarc, PikarcBlockedError
guard = AsyncPikarc(
api_key="lg_xxxx_yyyy",
base_url="https://api.pikarc.dev",
)
try:
async with guard.run(user_id="alice") as run:
response = await run.model_call(
fn=lambda: openai.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": query}],
),
model="gpt-4o",
token_extractor=lambda r: (
r.usage.prompt_tokens,
r.usage.completion_tokens,
),
)
result = await run.tool_call(
fn=lambda: search_database(query),
tool_name="search_database",
)
except PikarcBlockedError as e:
print(f"Blocked: {e.reason}")Automatic Enforcement
Budget, concurrency, and kill switch checks on every call
Full Cost Tracking
Token usage and cost calculated automatically per step
Complete Trace
Every step recorded with decisions, tokens, and duration
Built for production AI
Autonomous agents or customer assistants — keep control without slowing down.
Autonomous AI Agents
Let agents act independently while you keep control over spending and execution.
Customer-Facing AI
Deploy AI assistants that talk to customers while keeping costs under control.
Multi-Agent Systems
Run many agents in parallel with resource controls and full visibility.
Team AI Workflows
Give your team AI tools while maintaining budget governance across users.
Simple, transparent pricing
Every plan includes the full governance and observability core. Scale up when you need more runs, retention, and controls.
Basic
Solo devs / Prototypes
Perfect for getting started
- [✓]1,000 runs/month
- [✓]3 concurrent runs
- [✓]1-day retention
- [✓]Run tracking & cost analytics
- [✓]Email support
Pro
Growing AI Startups
Per-user daily budgets
- [✓]50,000 runs/month
- [✓]Configurable concurrent limits
- [✓]14-day retention
- [✓]Per-user daily budgets
- [✓]Run tracking & cost analytics
- [✓]Priority email support
Scale
Production-ready Agents
Kill switch
- [✓]Unlimited runs
- [✓]Configurable concurrent limits
- [✓]90-day retention
- [✓]Per-user daily budgets
- [✓]Kill switch
- [✓]Run tracking & cost analytics
- [✓]Priority email support
Enterprise
Large Org Governance
SSO, audit logs & dedicated support
- [✓]Everything in Scale
- [✓]Custom retention & run limits
- [✓]SSO / SAML
- [✓]Audit log export
- [✓]Dedicated support & SLA
- [✓]Custom deployment options
// all plans include core governance and observability
Start governing your AI agents today
Budget enforcement, cost tracking, and full observability from the first request.