Skip to content
Heighliner
Menu
Join early access

Early accessInfra for AI agent fleets

Run agent fleets you can trust.

Bring the agents you already use. Heighliner gives every fleet a dollar budget that hard-stops spend, a sandbox per agent, approvals for risky actions and a tamper-evident audit trail.

Join early accessRead the docs

Bring your own agent

  • Claude Code
  • Codex
  • Hermes
  • OpenCode
  • Custom container
  • Your own endpoint
fleet/growth-ops

Monthly budget

$18.40of $30.00 cap

warn 50 / 75 / 90% · hard stop at cap

  • seo-weeklyclaude-code
    running
    lse_041 $0.62/$1
  • pr-reviewcodex
    running
    lse_017 $0.34/$1
  • competitor-briefhermes
    running
    lse_009 $0.81/$1
  • nightly-reportopencode
    waiting on you

    IRREVERSIBLE email.send to 212 subscribers · held in escrow

  1. 06:14:00lease lse_041 granted $1.00 to seo-weekly
  2. 06:14:05pr.open allowed by policy r41 (reversible)
  3. 06:14:08nightly-report: email.send held in escrow, approval requested

nightly-report is waiting for approval to send an email.

A simulated fleet of four agents spending from a monthly budget. The governor's arms rise as spend approaches the cap. Each agent spends from a one dollar lease. One agent is waiting for approval to send an irreversible email.

Simulated fleet. Try the approval gate.

The problem

A VPS, a cron job and a .env file

That is how most agents run today, and it works until one loops overnight, emails the wrong list or spends a month's budget by morning. A machine is not the hard part. Everything around it is.

Today

On Heighliner

  • No spend cap

    Dollar budgets with a hard stop

  • No approvals

    Effect classes and an approval queue

  • Crash means start over

    Durable runs that resume

  • Raw keys with every scope

    Short-lived, narrow credentials

  • No audit

    Hash-chained trail and receipts

  • Agents share one box

    A sandbox per agent

  • Any egress

    Egress only through the gateway

What every fleet gets

Infrastructure for the agents, not the agents.

A fleet is one to many agents running one harness under one manifest, one budget and one policy. Whatever you run, the same five controls apply.

  1. Budgets01

    A dollar budget that hard-stops spend

    Set a monthly cap and a per-run cap in dollars. Every model and tool call spends from a small lease taken from the fleet's budget, so the worst overrun is one lease. At the cap, agents pause. They are never deleted, and they resume when you raise it.

    More on budgets

  2. Schedules and triggers02

    Runs start on a schedule or a signal

    Run hourly, nightly or on a cron expression, or start a run from Slack, a webhook, the API, or a chat in ChatGPT or Claude. Between runs an agent sleeps instead of holding a machine.

    More on schedules and triggers

  3. Sandboxes03

    A sandbox per agent

    Each agent runs in its own sandbox. Model keys and connector credentials are attached outside it at the gateway, and agents get short-lived, narrow credentials, never your keys. Network egress leaves only through the gateway.

    More on sandboxes

  4. Approvals04

    Approvals for risky actions, not every action

    Every tool call has an effect class: pure, reversible, compensatable or irreversible. Low-risk work runs within policy. Irreversible actions wait in escrow for a person, in Slack, ChatGPT, Claude or the console, and any approval can be saved as a rule.

    More on approvals

  5. Audit05

    A tamper-evident audit trail with receipts

    Every step is written to an append-only log where each event carries the hash of the one before it. Every run ends with a receipt: which agent, model and policy version, which checks passed, and the trace that led there.

    More on audit

  6. Your model key

    $0

    Bring a key from any provider or OpenRouter. Tokens are tracked for your budget and never marked up.

    See pricing

Approvals without rubber-stamping

You approve the rare risky action, not four hundred safe ones.

Every tool call carries an effect class. Policy decides what runs on its own; irreversible actions wait in escrow for a person. Any approval can become a rule.

  1. 01 Pure

    Agent, freely

    Read metrics, run a query

  2. 02 Reversible

    Agent, within policy

    Open a pull request, flip a feature flag

  3. 03 Compensatable

    Policy, often a person

    Charge a card (undo is a refund)

  4. 04 Irreversible

    Always a person

    Email customers, delete data

Agent-first, easy for humans

Plan, quote, apply.

The main caller is often another agent: fast, literal and sometimes wrong. So every change can be planned and priced before it runs, from the hl CLI, the SDKs, the API or a chat in ChatGPT or Claude.

fleet.yaml
fleet: seo-weekly
runtime: claude-code       # or codex, hermes, opencode
goal: "Find and fix SEO issues on example.com"
model: { provider: openrouter, key: secret://OPENROUTER_KEY }
triggers:
  - schedule: "0 6 * * 1"  # Mondays 06:00
  - slack: "#growth"
tools:
  github: { allow: [pulls.create], require_approval: [pulls.merge] }
budget: { monthly_usd: 30, per_run_usd: 0.50 }
terminal
$ hl plan fleet.yaml
  + fleet seo-weekly    runtime claude-code
  ~ budget              $30.00/mo, $0.50/run
  ! approval            pulls.merge
  quote                 ~$0.31/run + your tokens
  plan pln_7f2c expires in 15 min

$ hl apply pln_7f2c
  applied pln_7f2c  fleet seo-weekly is scheduled
How it worksCLI, SDKs and MCP

Three ways to run it

One codebase. Our cloud, your cloud, or your data center.

Start on the managed cloud in minutes. Move the data plane into your own AWS, GCP or Azure account when your data has to stay there, or run everything yourself, air-gapped if needed.

Compare deployment options

Why the name

In Dune, a heighliner is the Guild ship that carries whole fleets across space. Heighliner does the same for your agents: it carries the fleet, and the fleet does the work.

Questions

Is Heighliner an AI agent?

No. Heighliner is the infrastructure agents run on. You bring the agent, such as Claude Code, Codex, Hermes, OpenCode, a custom container or your own endpoint, and Heighliner adds budgets, schedules, sandboxes, approvals and an audit trail.

Who is it for?

Developers and indie hackers who run agents today on a VPS with a cron job and a .env file, and agencies running agents for clients. Enterprises can run it in their own cloud or fully self-hosted.

Which models can I use?

Any provider, or OpenRouter, with your own key. Heighliner tracks tokens so budgets work, and never marks them up.

Can I keep my data in my own cloud?

Yes. With BYOC the data plane runs in your AWS, GCP or Azure account and your data stays there. Fully self-hosted and air-gapped installs are also available.

Is it available now?

Heighliner is in early access. Join the waitlist and we will reach out as places open.

Early access

Give your agents a budget, a sandbox and a supervisor.

Heighliner is in early access. Join the waitlist and we will reach out as places open.

Join early accessRead the docs