A hard cost cap, approval gates, and a tamper-evident log for every agent action.

Wrap your agent's tool calls in one line of Python. A runaway agent can't blow your budget, delete prod, or act unreviewed, and you get a verifiable record of everything it did. Free to start. $29/month when you scale.

your_agent.py
$ pip install tenet-python

from tenet import TenetClient
tenet = TenetClient()

result = tenet.execute(
    service="github",
    tool_name="github.repos.delete",
    arguments={"owner": "acme", "repo": "prod-api"},
)
# -> outcome: "ESCALATE"

Three protections. Zero configuration.

Every tenet.execute() call runs through the policy engine. Five starter policies are pre-installed on signup, so your agent is governed the first time it makes a call.

Hard cost cap

Configure a per-agent USD budget over a rolling window. When the cap is hit, the next call returns 429, so no runaway $10K Anthropic invoice possible.

Approval gates on irreversible actions

github.repos.delete, stripe.payment_intents.create, slack.chat.delete: calls classified as irreversible auto-pause for human approval. The action only runs if you say yes.

Tamper-evident audit log

Every tool call your agent attempts, with the policy outcome, hash-chained so retroactive edits are detectable.
Tamper-evident: any retroactive edit breaks the chain, and you can verify it yourself.

Works with whatever you're building on.

Tenet doesn't replace your agent framework. It adds the missing layer: cost cap, approval gates, audit log, on top of whatever you're already shipping.

LangChain / LangGraph

Drop tenet.execute() inside any tool node. Same chain, governed.

Raw OpenAI / Anthropic SDK

Wrap your messages.create() or chat.completions.create() call. Same response, governed.

n8n / Mastra / Custom Python

One HTTP call from any workflow runtime. Language- and framework-agnostic.

Your agent never holds your downstream API keys.

Your agent holds a single Tenet API key and nothing else. Tenet adds the real GitHub, Stripe, or Anthropic credentials server-side at call time. If a model output gets exfiltrated, no third-party credentials go with it.

How it works.

01
Install the SDK
pip install tenet-python
02
Wrap your tool calls
Replace direct openai.chat() / github.repos.delete() with tenet.execute(). We inject credentials and evaluate against your policies before the call goes out.
03
Ship
Cost cap, approval queue, and audit log all live at app.trytenet.com. Five starter policies are bound to your default agent on signup.

Pricing.

One paid plan. Start free, upgrade when your agent starts shipping real work.

Free

$0forever
  • Every safety control: cost cap, approval gates, Tenet Shield
  • Tamper-evident audit log
  • 500 decisions/month (100/day)
  • 14-day audit history
Start free
Most popular

Tenet Pro

$29/month
  • Everything in Free
  • 25,000 decisions/month
  • 90-day audit history
  • Email support
Start free
Upgrade in-app once you've made your first call.
Need more? Reach us at [email protected].