Kits / code and agents

ApexSDK

Public SDKs and agent plugins for ApexAi products.

ApexSDK is the catalog of ApexAi's public developer kits: SDKs that put a product's contract into your code, and agent plugins that bring a product into Claude Code and OpenAI Codex. Each kit belongs to one product, talks only to that product, and is open source on GitHub.

At a glance

Who it is for: Developers who call Scrawl from Python, TypeScript, or a terminal; teams connecting their own systems to the DevOps engine; and anyone giving an AI coding agent Scrawl or DevOps.
What is here: Four kits: Scrawl's SDKs and command-line tool, Scrawl's agent plugin, the DevOps SDK, and DevOps' agent plugin.
Where it lives: Open source on GitHub under the Apache License 2.0. None is on PyPI or npm or in an official plugin directory yet: each installs from its repository.
Today: Scrawl SDKs and CLIBuiltScrawl agent pluginBuiltDevOps SDKIn developmentDevOps agent pluginIn developmentEach works only with its product, which ApexAi operates and does not offer publicly yet.
Next step: Compare the kits Discuss access

The catalog

Four kits. Each belongs to one product.

Every kit talks only to its own product, through that product's contract and policy. ApexSDK is a catalog, not a cross-product SDK: there is no shared package, base URL, or sign-on.

ApexSDK developer kits
Kit	Product	Kind	Runs on	Source and package	State
Scrawl SDKs and CLI	Scrawl	SDK	Python 3.10+ · Node 20+	Open source on GitHub. Not on PyPI or npm yet.	Built
Scrawl agent plugin	Scrawl	Agent plugin	Claude Code · OpenAI Codex	Open source on GitHub. Not in an official plugin directory: installs from its own marketplace.	Built
DevOps SDK	DevOps	SDK	Python 3.11+ on Linux	Open source on GitHub. Not on PyPI yet.	In development
DevOps agent plugin	DevOps	Agent plugin	Claude Code	Open source on GitHub. Not in an official plugin directory: installs from its own marketplace.	In development

Built, for a kit, means it is complete and tested against its product, and its source is public. It is not usable yet, because the product deployment it talks to is not offered to customers.
In development, for a kit, means its source is public but the product side it needs is not finished: the DevOps kits work with an engine that is still in development and not yet released.
A public repository is not a release of the product, a package on a registry, or a hosted service.

SDK, MCP, or plugin?

Different callers, different kinds of access.

An SDK and an agent plugin are kits around a product's own surfaces. They add no authority: every call still passes the product's identity, policy, and limits.

How SDKs, ApexMCP, agent plugins, and product plugins differ
Kind	What it is	Who calls it	Where it runs
ApexAPI	A product's typed HTTP contract, with scoped credentials, idempotent side effects, and stable errors.	Your software, over HTTPS	In the product
SDK	A library that puts one product's contract into your code: for Scrawl, clients and a command-line tool for its ApexAPI, with typed results, retries, and pagination; for DevOps, the host contract and adapters the engine uses to observe your systems.	Your code, or a terminal	In your application or integration
ApexMCP	A product's own MCP server: named tools and resources an AI agent discovers and calls, under the product's identity and policy.	An AI agent	Beside the product
Agent plugin	A package for an AI coding agent such as Claude Code or OpenAI Codex: skills that teach it the product, plus the connection to the product's MCP server or command-line tool. It holds no product logic.	An AI coding agent	In the agent's host
Product plugin	Operator control for one product's deployment: health, smoke tests, approvals, bounded maintenance, and evidence.	The operators of a deployment	Beside the deployment

Scrawl SDKs and CLI · SDK

Call Scrawl from Python, TypeScript, or a terminal.

Official clients for Scrawl's ApexAPI: scrape a page to Markdown or JSON shaped by your schema, map a site, run batches and crawls into versioned datasets, query and search datasets, extract and ask with citations on your own AI provider, extract a page's files, and read your credit balance. Both SDKs share one contract exported by Scrawl, and every ApexAPI operation is mapped to its Python method, TypeScript method, and command, or to the reason it is left out.

This kit: Built
Works with Scrawl ApexAPI: Built
Scrawl hosted service: Not publicly hosted

Source: ApexAiOfficial/scrawl-sdk Open source on GitHub · Apache License 2.0
Package: Not on PyPI or npm yet
Versions: Tagged releases on GitHub, before version 1.0
Runs on: Python 3.10 or later: synchronous and asynchronous clients, and the scrawl command
TypeScript and JavaScript on Node 20 or later, with no runtime dependencies
Needs: Access to Scrawl, which ApexAi operates, and an API key for your account with the scopes your calls need. Scrawl has no public hosted access or sign-up yet.
Your own AI provider, set on the server, for JSON extraction and cited answers.
Docs: Quickstart Python SDK TypeScript SDK Command line Authentication Errors and retries Compatibility Versioning

Install from a clone

# Not on PyPI or npm yet: install from the repository
git clone https://github.com/ApexAiOfficial/scrawl-sdk.git
python -m pip install ./scrawl-sdk/python    # Python SDK and the scrawl command
npm install ./scrawl-sdk/typescript          # TypeScript SDK

# Point it at the Scrawl endpoint for your account
export SCRAWL_BASE_URL=https://scrawl.example
export SCRAWL_API_KEY=...                    # an API key from its operator
scrawl doctor

Registry install commands in the repository work only once the packages are published; until then, install from a clone as above. The names scrawl on PyPI and on npm belong to unrelated packages: never install either in place of the SDK.

Python SDK

Typed results and errors, automatic idempotency keys, bounded retries, wait helpers for jobs and batches, and pagination.

TypeScript SDK

The same operations, error model, and retry rules for fetch-based runtimes, with bundled type declarations.

scrawl command

Scrape, map, batch, crawl, extract, ask, datasets, files, and a doctor check of the URL, TLS, and key, for a terminal or a CI job.

Contract and coverage

The contract Scrawl publishes for ApexAPI, and a coverage table that maps every operation to each SDK and the command.

Scrawl agent plugin · Agent plugin

Scrawl for Claude Code and OpenAI Codex.

Skills that teach an AI coding agent when and how to use Scrawl, and the connection to Scrawl: its MCP endpoint, authenticated on every request with your API key, or the scrawl command. The plugin holds no scraping logic; Scrawl, operated by ApexAi, does every fetch under its own policy, permissions, and limits, and the skills treat page text as untrusted data.

This kit: Built
Works with Scrawl ApexMCP: Built
Scrawl hosted service: Not publicly hosted

Source: ApexAiOfficial/scrawl-plugin Open source on GitHub · Apache License 2.0
Package: Not in an official plugin directory: installs from its own marketplace
Versions: Tagged releases on GitHub, before version 1.0
Runs on: Claude Code, from the repository's own plugin marketplace
OpenAI Codex, from the same repository
Not ChatGPT yet: it needs a sign-in flow and a public endpoint that Scrawl does not offer
Needs: Access to Scrawl, which ApexAi operates, and an API key for your account. Scrawl has no public hosted access or sign-up yet.
For the MCP tools, an MCP endpoint you can reach: Scrawl's MCP server accepts only local connections today, and hosted MCP access is planned.
For the command-line route, the scrawl command from the Scrawl SDKs.
Docs: Claude Code OpenAI Codex ChatGPT status Testing

Install in your agent host

# Claude Code
claude plugin marketplace add ApexAiOfficial/scrawl-plugin
claude plugin install scrawl@apexai-scrawl

# OpenAI Codex
codex plugin marketplace add ApexAiOfficial/scrawl-plugin
codex plugin add scrawl@apexai-scrawl

Then set your API key and Scrawl's MCP address in the host, as the repository's guides describe. The default address is a local one; there is no hosted Scrawl MCP endpoint yet.

Nine skills

Choose the operation, set up and check the connection, scrape, map, crawl, extract, ask with citations, work with datasets, and download a page's files.

MCP connection

The agent host connects to Scrawl's MCP endpoint, sending your API key on every request. Nothing connects until a key is set.

Command-line route

For downloads, the setup check, and Scrawl's HTTP API, the skills drive the scrawl command from the Scrawl SDKs.

No code of its own

No hooks, executables, telemetry, or network code: skills and configuration only.

DevOps SDK · SDK

Connect your systems to the DevOps engine.

The integration layer between your systems and the DevOps engine: the host contract the engine works through, adapters that observe systems, sinks that deliver attention notices, and contract checks you run in your own tests. You can build and test an adapter with the SDK alone; running it under the engine needs a DevOps engine installation, which ApexAi does not offer to customers.

This kit: In development
Works with DevOps engine: In development
DevOps hosted service: Not publicly hosted

Source: ApexAiOfficial/devops-sdk Open source on GitHub · Apache License 2.0
Package: Not on PyPI yet
Versions: Version tags on GitHub, before version 1.0
Runs on: Python 3.11 or later, on Linux, with no runtime dependencies beyond the standard library
Needs: Nothing to build and test an adapter: the SDK and its test stand-ins are self-contained.
A DevOps engine installation to run it: the engine is in development, and ApexAi does not offer it to customers.
Docs: Host contract Adapters and sinks Architecture Examples

Install from a release tag

# Not on PyPI: install a tagged release from the repository
pip install "apexai-devops-sdk @ git+https://github.com/ApexAiOfficial/devops-sdk@<tag>"

Pin a release tag or a commit, as the repository describes. An adapter never guesses: a system it cannot read is reported as unobservable, and the engine never acts on that.

Host contract

Observations with four outcomes, the questions an adapter answers about a system, remedies, and attention notices.

Adapters and sinks

Watch Docker Compose services and their health checks, read a status file another process writes, and deliver attention notices to Telegram chats.

Contract tests

Checks that prove your own adapter meets the contract, and offline stand-ins for the Docker daemon and the Telegram Bot API.

A read-only monitor

Wiring for running an adapter under the engine as a monitor that observes and alerts but never acts.

DevOps agent plugin · Agent plugin

Ask an AI agent what the engine did, and why.

Read-only inspection of a DevOps engine installation from Claude Code: its status and health, the systems it watches and who owns them, incidents, decisions, the changes it made, budgets, recovery holds, and redacted evidence, plus a triage command. Every tool it offers is a read: it never grants authority, requests a remediation, or runs a command.

This kit: In development
Works with DevOps engine: In development
DevOps hosted service: Not publicly hosted

Source: ApexAiOfficial/devops-plugin Open source on GitHub · Apache License 2.0
Package: Not in an official plugin directory: installs from its own marketplace
Versions: A development pre-release with no tag yet
Runs on: Claude Code, from the repository's own plugin marketplace
Needs: A DevOps engine installation at the exact version the plugin names; the repository lists the system requirements. ApexAi does not offer the engine to customers.
Read access to that installation's records. The engine is in development and is not offered to customers yet.
Docs: README OpenAI hosts

Install in Claude Code

# Add the repository's plugin marketplace
claude plugin marketplace add ApexAiOfficial/devops-plugin

Then install the plugin from that marketplace and point it at your engine installation, as the repository's README describes step by step, including the engine version it requires.

Inspection tools

Status, incidents and their decisions, changes, budgets, recovery holds, control events, and redacted evidence, read from the engine's own records.

Triage command

A guided look at what is unresolved, for the whole installation or one incident.

A version check first

The plugin starts the engine's inspection server only after checking that the installed engine is the exact version it names and that its files match the installation's own record.

No authority

Nothing it does changes the engine or your systems: standing policy still decides every action.

Before you start

A kit needs its product.

The kits are public so you can read them, build against them, and plan an integration. Using one needs a deployment of its product, which ApexAi does not offer publicly yet.

Access to the product

Every kit talks to one product, and ApexAi operates both: Scrawl, and the DevOps engine. The engines are proprietary and never distributed, and neither product is publicly hosted or offered for sign-up yet.

Credentials from that product

Scrawl API keys are issued for your account by ApexAi and carry named scopes. DevOps kits connect to a DevOps engine installation, which ApexAi does not offer to customers.

Your own AI provider, where it applies

Scrawl's JSON extraction and cited answers run on the AI provider you bring. The kits never hold an AI key of their own.

Versions and support

Pin a version. Report problems privately.

The kits change in the open on GitHub. Communications has no kit: software and agents reach it through ApexAPI and ApexMCP directly. Shop Bot is used inside Telegram.

Versions

Every kit is before version 1.0, so a minor version may change its interface. Pin a kit's tag where it has one. The DevOps agent plugin is a development pre-release with no tag yet: it follows its main branch and requires the exact engine version it names.

Compatibility

The Scrawl SDKs record the contract they were built from and check it in their own tests. The DevOps agent plugin requires the exact engine version it names.

Questions and issues

Open an issue in the kit's repository, or write to support@apexaiofficial.com. For access to a product, write to sales@apexaiofficial.com.

Security

Report a vulnerability privately to support@apexaiofficial.com, never in a public issue. The kits read keys from the environment, your agent host, or files only their owner can read, and never print them.

Every public ApexAi repository: ApexAi on GitHub