[go: up one dir, main page]

  • DeepSeek Harness
  • Claude Code
  • Codex
  • Hermes Agent
  • claude-opus-5-5
  • deepseek-v4-pro
  • gpt-6.1-sol
  • gpt-5
  • claude-sonnet-5-5
  • simple-summary-agent
  • weather-brief
  • test-fixer
  • pr-reviewer
  • weekly-report
  • docs-writer
  • oncall
  • support-desk
  • dep-upgrader
  • reviewer
  • slack-digest-mailer
  • support-triage
  • text-summary
  • support-bot
  • standup
THE AGENT CLOUD

Deploy your agent. Make every run count.

Bring your agent to the cloud with its instructions and tools. Follow each run and test changes on the work it actually does.

Codex~/example-store/support-bot

Deploy this agent to LagrangeCloud

  1. Read AGENTS.md, 2 skills and a script
  2. Added MCP servers linear and stripe
  3. Stored 2 keys as secrets, without printing them
  4. Deployed support-bot v1 to LagrangeCloud using Codex

Live https://lgr.tensorblock.co/v1/invoke/ag_5d02c7a3b81e

Example store supportlive

Where is order 42?

It shipped yesterday and arrives Friday.

Lagrange found an optimizationLook up each order once
Cost per run
$0.031 → $0.013
Median time
4.8s → 2.0s
Promoted. support-bot v2 is live.

Take your agent from local to live.

Deploy your existing agent with its instructions, tools, and dependencies. Lagrange handles the infrastructure, stores credentials as secrets, and gives you one place to manage and inspect executions.

  • Codex
  • Claude Code
  • Hermes Agent
  • DeepSeek Harness
  1. Your agent

    Codex App, Codex CLI, Claude Code, Hermes Agent or DeepSeek Harness

    $ npx -y @openai/codex plugin marketplace add https://github.com/TensorBlock/lagrange-plugins
    $ npx -y @openai/codex plugin add lagrange@tensorblock
    Then, in your agent's folder: "Deploy this agent to LagrangeCloud"
  2. Lagrange Cloud

    Hosted execution

    support-botv1 live
    Endpoint
    /v1/invoke/ag_5d02c7a3b81e
    Harness
    Codex
    Secrets
    2 keys stored
    Executions
    Every run, inspectable
  3. Where it runs

    One deploy, three ways

    • From your appCall your agent through an APIPOST /v1/invoke
    • On your websiteMake it available to usersstore.example
    • On a scheduleRun recurring tasks automatically0 8 * * *

Deploys exactly as you built it. API keys stay in secrets, never in chat.

LagrangeCloud hosts your agent. A harness runs its instructions and tools: Codex, Claude Code, Hermes or DeepSeek Harness today, with our self-improving Lagrange harness coming soon.

~/example-store/support-bot
AGENTS.md
skills/orders/SKILL.md
skills/refunds/SKILL.md
scripts/lookup_order.py
.codex/config.toml  linear, stripe
.env                LINEAR_API_KEY, STRIPE_SECRET_KEY
support-botv1 live
Instructions
AGENTS.md
Skills
orders, refunds
Files
lookup_order.py
MCP servers
linear, stripe
Secrets
LINEAR_API_KEY, STRIPE_SECRET_KEY
Harness
Codex in this example

Infrastructure that improves with your agent.

  • Learns from every run

    Most agent infrastructure stays the same after deployment until someone changes it by hand. Every execution generates data that can reveal better ways to run your agent.

  • Proposes and tests changes

    Lagrange analyzes execution traces to find ways to reduce latency, lower costs, and improve reliability, then tests each change against the current version.

  • You approve what ships

    Review what changed, why it was proposed, and how it performed. Each accepted improvement becomes the baseline for the next round of optimization.

Optimization proposalPending review

Example

Use a lighter model for classification

Route a straightforward classification step to a smaller model, while keeping the remaining workflow unchanged.

steps.classify:
- model: large
+ model: small
steps.*: unchanged

Current and proposed

CurrentProposed

Execution time−28% 8.2s 5.9s
Cost per run−31% $0.042 $0.029
Output qualityUnchanged 98% 98%

Changes require your approval

Illustrative example, not measured Lagrange results

Each accepted improvement becomes the baseline for the next cycle.

  1. Deploy once

    01

    Deploy

    Get your agent running

    support-bot v1 live
  2. Then on every run

    v1v2v3…

    02

    Run

    Execute real tasks

    APIWebsiteSchedule
  3. 03

    Trace

    Capture execution data

  4. 04

    Improve

    Test and approve changes

    v2 proposed

Try it on your agent.

Choose how you want to deploy your agent to Lagrange. Then, once it's running, Lagrange will begin collecting execution data and identifying opportunities for improvement.

$ npx -y @openai/codex plugin marketplace add https://github.com/TensorBlock/lagrange-plugins
$ npx -y @openai/codex plugin add lagrange@tensorblock
Then, in your agent's folder: "Deploy this agent to LagrangeCloud"
PRICING

Pay for what you use.

Runs, instances, and fleets from

$0.05/ sandbox hour

  • Free credits to get started
  • Model tokens billed separately at list price