The full stack to ship agents in customer environments.

Gateway, control plane, and runtime — so every agent runs under each user's and each agent's own permissions, with hard budgets, an off switch, rollback, and a verifiable record of every action.

  • Any model, cloud or local
  • Any agent engine
  • One UI for every agent
control.agenthippo.ai
support-agent v2 · signed
running
Spend this month $3.20 of $50 cap
Audit record
aliceQ3 refunds by region · 14 rows$0.04
bobsame question · 0 rows$0.02
loopcall over cap · refused$0.00

The stack around your agent

Hosted by us — or on your own VM
Control plane Live usage, spend caps, a kill switch, version pinning, and rollback.
Customer environment
Runtime
Your agent
  • LangChain
  • CrewAI
  • Agent Pack
  • any SDK
Acts as each user, within each agent's permissions. Holds no credentials.
Gateway Routes every call to your providers, your internal model proxy, or local models — budget-checked first.
Data access Each user gets only their own rows.
  • Postgres RLS
  • Unity Catalog
  • DynamoDB
Evidence Who asked, which signed version ran, what it touched, what it cost.
  • OpenTelemetry
  • S3 Object Lock

Run your first governed agent

On your laptop or a server — either way, you govern it from control.agenthippo.ai.

  • Docker Desktop with WSL 2 + Ubuntu
  • AgentHippo IDE
  • An agent pack — any folder with agent.yaml, or a demo pack
  1. Turn on Docker for WSL once

    Docker Desktop → Settings → General: tick Use the WSL 2 based engine. Then Settings → Resources → WSL Integration: switch on Ubuntu and click Apply & Restart.

  2. Register your agent and download its bundle

    Sign in at control.agenthippo.ai, click + register agent, and download the bundle. Extract it anywhere — for example C:\Users\<you>\Downloads\<agent>-bundle\<agent>.

  3. Bootstrap the agent

    In the Ubuntu terminal, point the bundle at your pack:

    Ubuntu (WSL)
    cd /mnt/c/Users/<you>/Downloads/<agent>-bundle/<agent>
    bash ./bootstrap.sh /mnt/c/path/to/your-pack

    It worked if the output ends with:

    ✓ agent up on 127.0.0.1:8095
    ✓ IDE: open stack/ide-pack-local in the AgentHippo IDE — the loopback edge on 127.0.0.1:9095 verifies your sign-in
  4. Chat with it in the IDE

    Open the AgentHippo IDE and sign in. In the chat panel, click the agent picker at the top right (it shows Default) → Install Agent from Folder, and choose the bundle's stack/ide-pack-local folder. Pick your pack from the same menu and start a chat — it runs on your local agent.

  5. Control it from the console

    Watch usage at control.agenthippo.ai. Kill the agent and send a message: it's rejected. Resume it and send again.

Then put it to the test

Cap the spend

Set a budget and watch token costs meter live. The call over the cap is refused, not flagged.

Hit Stop

Kill the agent from the console. It stops within one poll — no SSH — until you resume it.

Ship v2, then roll back

Run ./bootstrap.sh with a new pack folder, then roll back to the last version.

Pull the audit trail

Who ran what, which version, and what it cost — for every turn.

No agent yet? Start with a demo pack

Running quickstack-analyst? Ask us for the seed scripts.

The IDE and CLI are free

AgentHippo IDE

Build packs and chat with your agents.

Download IDE for Mac
  • Any model, cloud or local Anthropic OpenAI Gemini Bedrock Vertex local models
  • Any agent engine Claude Code Codex OpenClaw your own
  • One UI for every agent plus Slack WhatsApp Telegram

AgentHippo CLI

Serve, automate, and run in CI.

Download CLI

After download: unzip agenthippo.zip && ./install.sh

Or install from a terminal
curl -fsSL https://agenthippo.ai/install.sh | bash

All builds are signed. Checksums · Release manifest

Going to production? Hosted Control Plane per agent, or the whole stack self-hosted in your VPC. Want us to ship the first one? The Production Sprint: one real workflow in production in three weeks — measured, and proven under control.