Are you an LLM? Read llms.txt for a summary of the docs, or llms-full.txt for the full context.
Skip to content

Amodal Cloud

Build and operate agents from your repo

What is Amodal?

Amodal is a cloud platform for building domain-specific AI agents from a git repo of Markdown, JSON, and TypeScript. Teams connect their APIs, add domain knowledge, define skills, and deploy the agent through Amodal or the Platform API. Amodal provides the reasoning loop, hosted runtime, session replay, evals, cost tracking, deployment controls, and operational logs around it.

The core loop is explore → plan → execute: the agent gathers data, reasons about it, and takes action, compounding knowledge with every interaction. You define the agent in source. Amodal turns that source into a production deployment that can be embedded in an app, operated by a team, and monitored over time.

How it works

An Amodal project is a folder of plain files that describe what your agent knows, what it can access, and how it should think:

  • Connections define API access. Each connection is a directory with a spec, auth config, and natural-language documentation that teaches the agent what endpoints exist and when to use them.
  • Skills are markdown files that encode expert reasoning — investigation methodologies, triage procedures, escalation logic. They activate automatically based on the user's question.
  • Knowledge is persistent domain context: environment details, team structures, historical patterns, baseline metrics. Knowledge docs are loaded into the agent prompt at session start.
  • Tools are custom actions beyond the built-in set — HTTP calls, multi-system chains, or arbitrary function handlers with confirmation rules and rate limits.

Amodal and the hosted runtime compile these files into an agent context and drive a state-machine loop against supported LLMs. The agent reasons about user questions, calls configured tools and connections, and returns structured answers. Sessions, deploys, evals, activity, and cost data are tracked so operators can see what changed and why.

# Agent source lives in git.
amodal.json
amodal/connections/slack/
amodal/skills/triage/SKILL.md
amodal/knowledge/runbook.md
evals/escalation.md

Build

  • Build With A Coding Agent — Give Codex, Claude, or CI the API contract and workflow for creating, deploying, and testing agents.
  • Project Structure — Learn the repo layout before editing agent files.
  • Build A Support Agent — Create a working tutorial agent from source files and a public demo API.
  • Source & Deploy In Amodal — Connect source, deploy, promote, and inspect the runtime.
  • Connections — Define or install API access for the systems your agent should call.
  • Skills — Author expert reasoning frameworks in Markdown.
  • Tools — Define custom HTTP, chain, and function tools.
  • Knowledge — Persistent domain context that grows with use.

Operate & Embed

  • Providers — Configure model provider access and selected models.
  • React SDK — Embed agents in your product with @amodalai/react.
  • Evals — Test behavior across models and deployments.
  • Runtime Apps — Ship a custom SPA through the deployed agent URL.
  • Architecture — The explore-plan-execute loop, context isolation, and multi-agent dispatch.