# AdaL

> AdaL is an agentic engineering harness that unifies deep research, coding, browser use, and code review into one continuous session with multi-model support.

AdaL, built by SylphAI, Inc., is an agentic engineering platform that brings fragmented AI-assisted development workflows into a single continuous session. It combines a CLI, an Agentic IDE, specialized worker agents, and browser automation under one subscription, with no setup required according to the product page.

## What It Is

AdaL positions itself as an "agent harness" — a layer that sits above frontier AI models and optimizes how those models plan, use tools, recover from errors, and verify results on real engineering tasks. Rather than replacing a single tool like a code editor or terminal, AdaL spans the full lifecycle: deep research from the codebase, autonomous coding, browser-based verification, and clustered code review. The platform supports routing tasks to specialized worker agents (Deep Research, Coding, Browser Use, Review) and switching between model families including Anthropic, OpenAI, Google, DeepSeek, Moonshot, Z.AI, and MiniMax.

## How the Agent Harness Works

AdaL's core workflow runs through four coordinated layers:

- **CLI (YOLO mode):** Starts an autonomous worker loop that chains research, coding, browser verification, and review steps without manual handoffs.
- **Worker Agents:** Specialized agents handle distinct tasks — Deep Research maps the problem space, Coding implements the solution, Browser Use validates the live experience, and Review clusters changes for human inspection.
- **Agentic IDE:** A browser and native desktop workspace where engineers inspect source code, review visual tasks, and collaborate with AdaL on clustered change sets.
- **Headless / SDK / Meta-Agent:** AdaL can run in CI pipelines, be composed into custom domain agents via SDK, or use a meta-agent to scaffold entire agent systems.

The platform maintains full chat history, visible tool inputs and outputs, and reviewable steps throughout, so the autonomous loop does not become a black box.

## Benchmark Claims

The AdaL website publishes benchmark comparisons against competing tools. On the vendor-published BU Bench V1 browser automation benchmark, AdaL claims 92% on BU100 versus 89.5% for Browser Use and 74% for Claude Code. On a vendor-published SWE-Pro-Curated-50 coding benchmark, AdaL claims 50% accuracy versus Claude Code's 46%, 1.5× faster completion (209s vs 315s), and 64% lower cost per success ($0.68 vs $1.87). These figures are self-reported by SylphAI and have not been independently verified.

## Platform and Interface Options

AdaL is accessible across multiple interfaces from a single subscription:

- **CLI** — terminal-first, fast autonomous sessions on Mac and Linux (Windows support noted)
- **Agentic IDE** — browser-based and native desktop (macOS Apple Silicon desktop app listed as Preview)
- **Web** — browser access to the full workspace
- **Headless** — triggered from CI, background jobs, or internal tools via SDK

Model switching and worker-agent routing are available across all interfaces via `/model` and `/agent` commands.

## Audience and Community

AdaL targets professional engineering teams and individual developers who want to automate the full engineering workflow rather than just code generation. The product page quotes engineers, CTOs, founders, and a university professor as users. SylphAI also runs a community program with hackathons and events, and an AI bootcamp at adal.techexpert.io for learning agentic engineering. A student discount program is offered. The company is SOC 2 Certified, and the Teams & Enterprise tier includes SSO, SAML/SCIM provisioning, and Zero Data Retention options for up to 150 seats.

## Features
- CLI with YOLO autonomous mode
- Deep Research worker agent
- Coding worker agent
- Browser Use agent for web verification
- Agentic IDE (browser and desktop)
- Clustered code review
- Multi-model support (Anthropic, OpenAI, Google, DeepSeek, Moonshot, Z.AI, MiniMax)
- Model switching mid-session
- Headless mode for CI/CD integration
- SDK for custom domain agents
- Meta-agent for scaffolding agent systems
- Long-term shared memory between agents and humans
- Full chat history and visible tool I/O
- SAML/SCIM provisioning (Enterprise)
- SSO integration (Enterprise)
- Zero Data Retention option (Enterprise)
- SOC 2 Certified

## Integrations
Claude (Anthropic), GPT (OpenAI), Gemini (Google), DeepSeek, Moonshot (Kimi), Z.AI (GLM), MiniMax, ElevenLabs, Veo, Remotion, ffmpeg, GitHub (CI/headless)

## Platforms
WINDOWS, MACOS, LINUX, IOS, WEB, API, DEVELOPER_SDK, CLI

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://adalagent.ai
- Documentation: https://docs.sylph.ai/
- Repository: https://github.com/SylphAI-Inc
- EveryDev.ai: https://www.everydev.ai/tools/adal