How it works

Stay in control at every step.

Start by observing. Turn on controls one policy at a time, without changing how your agents run.

Start free trial Book a demo

Step 1

Connect your agents

Wrap your model client with two lines of code, or record calls from any framework or provider.

About 15 minutes

Step 2

Observe in monitor mode

Agents, owners, spend and sensitive data near-misses fill the dashboard. Nothing is blocked.

First report in week one

Step 3

Turn on controls

Move policies from monitor to enforce, add approvals and rulebooks, and invite approvers.

At your own pace

import agentdash
from openai import OpenAI

ad = agentdash.Client(api_key="ad_ingest_...")

# Every call is tracked: agent, model, tokens, cost
llm = ad.wrap(OpenAI(), agent="support-copilot")
llm.chat.completions.create(model="gpt-4o-mini", ...)

import agentdash
from anthropic import Anthropic

ad = agentdash.Client(api_key="ad_ingest_...")

# Every call is tracked: agent, model, tokens, cost
claude = ad.wrap(Anthropic(), agent="contract-reviewer")
claude.messages.create(model="claude-sonnet-4-5", ...)

import agentdash
from google import genai

ad = agentdash.Client(api_key="ad_ingest_...")

# Every call is tracked: agent, model, tokens, cost
gem = ad.wrap(genai.Client(), agent="sales-writer", provider="google")
gem.models.generate_content(model="gemini-2.5-flash", ...)

import agentdash

ad = agentdash.Client(api_key="ad_ingest_...")

# Bedrock, Mistral, Azure, a self-hosted model or raw HTTP
ad.record_llm_call(agent="my-agent", provider="bedrock",
                   model="llama-3-70b", input_tokens=812, output_tokens=140)

Two lines for engineers. One dashboard for everyone else.

OpenAI, Anthropic, Google Gemini and xAI clients are wrapped directly. Calls to any other provider are recorded with one function call. Executives and approvers see friendly agent names, never raw IDs.

No rip and replace

Keep your stack. Add a dashboard.

AgentDash sits beside the agents, models and tools you already run. Nothing about how your agents work has to change.

Any model providerOpenAI, Anthropic, Google Gemini and xAI out of the box, and any other provider with one function call.
Your company sign-inPeople sign in with Okta or Microsoft Entra, with Admin, Approver and Viewer access.
Monitor mode firstObserve for as long as you like. Nothing is blocked until you turn a policy on.
Reports in the formats you useGovernance reports export to PowerPoint, Excel and CSV.

Every agent. One dashboard.

Start with one team. Within a week you will see how many agents you have, who owns them and what they cost.

30-day free trial with a preloaded demo organization
Monitor mode first, no production risk
A 30-minute demo tailored to your teams