Integrations — Agent Spend Firewall Connectors
Native integrations for LangChain, CrewAI, Vercel AI SDK, OpenAI Agents, AWS Bedrock, and custom frameworks.
sipi.bot integrates with LangChain agents via a single callback. Add spend evaluation to any LangChain agent in 3 lines …
CrewAI multi-agent systems need per-agent spend controls. sipi.bot provides isolated workspace policies for each connected API key.…
Vercel AI SDK agents can call the sipi.bot evaluation endpoint before any tool execution that incurs cost.…
OpenAI Agents SDK integration: wrap your payment tools with sipi.bot evaluation for no-model spend decisions.…
AWS Bedrock agents calling paid AWS services — add sipi.bot as a pre-execution guard for cost control.…
Any agent that can make an HTTP POST request can integrate. REST API, deterministic rules, JSON response.…
Detailed integration walkthrough for Integrations — Agent Spend Firewall Connectors
Integrating sipi.bot with Integrations — Agent Spend Firewall Connectors adds a policy enforcement layer between your agent's intent to spend and the actual transaction. The integration is deliberately minimal — a single API call with four fields (agent ID, merchant, amount, intent) — but the effect is comprehensive: every transaction must pass the policy check before money moves.
Step-by-step setup
- Install the sipi.bot SDK or use the HTTP API. If Integrations — Agent Spend Firewall Connectors runs on Python, run
pip install sipi-bot. For other runtimes, use the REST API atPOST https://sipi.bot/v1/transactions/evaluate. Authentication is via bearer token. - Identify the spend paths. Search your Integrations — Agent Spend Firewall Connectors agent code for every place it can initiate a payment, charge, API call, provisioning action, or purchase. Each one needs a policy check inserted before the actual transaction executes.
- Wrap each path with a policy check. The check takes four parameters: agent ID (string), merchant (string — the vendor or destination), amount (float in USD), intent (string — free-text description of what the transaction is for). Example:
sipi.check(agent_id="agent-1", merchant="openai", amount=0.50, intent="LLM call") - Handle the response. sipi.bot returns a JSON object:
{"decision": "APPROVED"|"BLOCKED"|"FLAGGED", "reason": "...", "transaction_id": "...", "policy_version": "..."}. Your agent must handle all three outcomes. Do not assume APPROVED is the only response. - Test each path. Run one test for each outcome: a small approved transaction, a blocked transaction (exceed a limit), and a flagged transaction (exceed the approval threshold). Verify the audit log captures all three with correct details.
Testing your integration
After setup, run the integration with sipi.bot's sandbox mode first (set header X-Sipi-Mode: sandbox). Sandbox mode evaluates policies and writes to the audit log but never blocks a real transaction. Use it to verify your integration and tune your limits for 24 hours before enabling enforcement. Most teams find this catches 2-3 configuration issues that would have caused false positives in production.
Pricing
sipi.bot starts at $99/month for unlimited transaction evaluations. No per-call fees, no tier upgrades for higher volume. The integration with Integrations — Agent Spend Firewall Connectors is supported out of the box, and setup typically takes under an hour for teams familiar with their agent's codebase.