Stop your AI agents from burning money on runaway loops — and prove what you saved.
AI agents loop on failed calls and burn tokens, spend money without asking, delete data, and hit APIs with no checkpoint. Orka sits in front of every action: it cuts the runaway loop, caps the spend, blocks the dangerous call, holds risky actions for human approval, and logs everything to a tamper-evident ledger.
Real incidents, not hypotheticals:
- Teams report agents looping on a failed call and burning hundreds of dollars in tokens over a weekend — with no alert.
- An OpenAI Operator agent spent $31 on eggs without asking — its own safety check never fired.
- Replit's coding agent deleted a production database in 9 seconds, during a freeze meant to prevent exactly that.
- Air Canada was held legally liable for a refund policy its chatbot invented on the spot.
Most teams have no checkpoint between the agent's decision and the expensive — or irreversible — action.
| Capability | What it does |
|---|---|
| Loop guard | |
| Detects an agent repeating the same failed action and cuts it before it drains the budget | |
| Spend cap | |
| Hard token/cost limit per run — stops the execution when crossed | |
| Savings ledger | |
| Quantifies, in dollars, the waste it prevented — "Orka saved you $X" | |
| Human approval | |
| Holds high-risk actions for human sign-off before they execute | |
| Immutable audit trail | |
| SHA-256 chained ledger — tamper-evident record of every action and decision | |
| Policy engine | |
| Blocks actions by rule: task type, domain, quota, risk level | |
| Multi-protocol | |
| MCP, A2A, REST, and custom agent protocols |
pip install orkaia
python
import orka
orka.init(api_key="orka_...") # get a key at orka.ia.br
@orka.guard(agent_id="my-agent", task_type="web_search")
def search(query: str) -> str:
return your_llm.call(query)
result = search("latest quarterly report")
Every execution appears in real time at orka.ia.br/dashboard: input/output, duration, cost, status, risk score, and a searchable audit trail.
Agent → Orka → Loop guard → Spend cap → Policy check → Risk score → [Human approval?] → Execute → Ledger
The SDK intercepts the call and hands it to the Orka backend. The decision logic (loop detection, budget enforcement, policy evaluation, risk scoring) runs server-side. The ledger is immutable and cryptographically chained.
The reason most teams lose money on agents isn't a single catastrophic event — it's the slow bleed of retry loops, redundant calls, and uncapped sessions.
Orka's economy layer tracks cost per action, detects loops automatically, enforces budget caps, and generates a savings report showing exactly how much it prevented.
@orka.guard(agent_id="analyst", task_type="data_fetch", risk="LIMITED")
def fetch_data(source: str) -> dict:
return api.get(source)
| Layer | What it is | License |
|---|---|---|
SDKs (python/ , typescript/ ) |
||
@guard decorator, REST client, integrations |
||
| MIT, this repo | ||
| Backend | ||
| Policy engine, risk scoring, ledger, approval routing | Managed service at orka.ia.br |
The SDKs are open source — read them, fork them, audit them. The backend runs the decision logic. A fully local/offline mode is on the roadmap.
Works with any framework. Add one decorator or callback:
LangChain:OrkaCallbackHandler
— every LLM call and chain step logged automaticallyCrewAI / AutoGen: wrap any tool with@orka.guard
OpenAI:OrkaOpenAIAdapter
for function-calling loopsMCP: native MCP support — Orka as a tool server for Claude, Cursor, Windsurf
See python/examples/ and
python/orka/integrations/
Base URL: https://orka.ia.br/api/v1
| Endpoint | Method | What it does |
|---|---|---|
/agents/ |
||
| GET / POST | List or register agents | |
/handover |
||
| POST | Submit an action for Orka to process | |
/handover/{task_id} |
||
| GET | Check execution status | |
/xshield/policies |
||
| GET / POST | Manage governance policies | |
/approvals |
||
| GET | List pending human approvals | |
/approvals/{id}/approve |
||
| POST | Approve a pending action | |
/xledger/entries |
||
| GET | Query the immutable audit ledger | |
/xledger/verify |
||
| GET | Verify chain integrity | |
/assurance/risk-report |
||
| GET | Per-agent risk scores | |
/metrics/dashboard |
||
| GET | Real-time platform metrics |
Auth: X-API-Key
header.
Issues and PRs welcome on the SDKs. The governance backend is closed-source — if you want to discuss architecture, integrations, or the roadmap, open a discussion or reach out.
Built for teams that deploy AI agents and need to stay in control.
** orka.ia.br** · contato@orka.ia.br