The core agent loop for AgentLayer. Wraps Vercel AI SDK's `streamText` in a resumable, approval-aware tool-execution loop, and defines the interfaces that platform packages (`agentlayer-filesystem`, `agentlayer-justbash`) implement. Ships isomorphic tools

Filesystem-backed implementations of `agentlayer-core`'s tool interfaces (read, write, edit, apply_patch, glob, grep, list, bash), plus the hooks and prompts needed to assemble a real coding agent that operates on a local directory. Where `agentlayer-core

OpenAI Codex provider for AgentLayer and the AI SDK. It talks to the ChatGPT Codex responses endpoint (`https://chatgpt.com/backend-api/codex/responses`) and supports ChatGPT OAuth/API-key auth through `@humanlayer/agentlayer-provider-auth`. Every factory

Universal AgentLayer wallet installer for OpenClaw, Codex, Claude Code, and Hermes.

`codelayer` is AgentLayer's reference coding agent — a multi-provider CLI (`@humanlayer/codelayer`) built on `agentlayer-core` and `agentlayer-filesystem`. It wires a `LanguageModel` (Anthropic, OpenAI, Codex, Copilot, or Fireworks/"firepass") to a filesy

Yjs-backed implementations of `agentlayer-core`'s filesystem tool interfaces (`read`, `write`, `edit`, `apply_patch`, `glob`, `grep`, `list`), executing against `@humanlayer/yjs-fs`'s `YjsFilesystem` — a CRDT virtual filesystem — instead of disk. Because

Conformance runner for AgentLayer Connector Protocol v1.

Bridges [`@humanlayer/yjs-fs`](../yjs-fs) (a Y.js CRDT filesystem) into [`secure-exec`](https://github.com/rivet-dev/secure-exec)'s sandboxed Node runtime, and exposes it as an `agentlayer-core` tool. Code the agent runs inside the sandbox reads/writes th

A `bash` tool for AgentLayer agents that executes commands against a [`YjsFilesystem`](../yjs-fs) instead of the real OS filesystem, using [`just-bash`](https://github.com/vercel-labs/just-bash) as the shell interpreter. Every file read/write/mkdir/rm/cp/

TypeScript SDK for building AgentLayer read-only connectors.

l33tAgents SDK (ex AgentLayer) — typed agent-side client (tasks, SSE, MCP discovery, idempotency) and studio-side Express middleware (HMAC-verified server-authoritative handlers, checkpoints, artifacts, well-known discovery).

Implements AgentLayer's core tool interfaces (`BashTool`, `ReadTool`, `WriteTool`, `EditTool`, `GrepTool`, `GlobTool`, `ListTool`, `ApplyPatchTool`, `WebFetchTool`, `WebSearchTool`, `CodeSearchTool`, skills) on top of [`just-bash`](https://github.com/verc

Storage and normalization for provider credentials (OAuth tokens and API keys) used by AgentLayer's model providers. It defines a small `AuthStore` interface with an in-memory implementation for tests and a file-backed implementation that persists to `~/.

The SDK for AgentLayer

An AI SDK v3 (`ProviderV3`/`LanguageModelV3`) provider that talks to the GitHub Copilot chat API. It handles Copilot's device-code OAuth flow, injects the Copilot-specific auth/User-Agent/intent headers on every request, and routes each model to either th

The official AgentLayer SDK — financial governance for AI agents

Logging utilities for AgentLayer

Core scanner, extractor, evaluator, and artifact generator for AgentLayer.

This package contains the global `agentlayer` command, its command registration layer, and the core modules it orchestrates.

This package exposes the MCP server so coding agents can query and log project memory through AgentLayer.