AdaL is a multi-agent development tool built by SylphAI, Inc. It autonomously orchestrates a set of worker agents that handle coding, deep research, browser automation, and code review, switching between frontier models such as Claude, GPT, and Gemini as it works through a task. You can use it through several interfaces — a CLI, a desktop app, and a browser-integrated agentic IDE — and the SDK and headless mode let you embed it in CI/CD pipelines and server environments. It is named after Ada Lovelace, often called the world’s first programmer, and launched in December 2025. In July 2026 it gained “Engineer mode,” where you hand over a goal and it autonomously carries the work from planning through implementation, verification, and handoff. August brought the “@Skills Hub,” a place to discover and share reusable agent skills — the feature set keeps expanding.
Key features
- AdaL Engineer (autonomous execution mode): A goal-oriented mode launched with
--mode engineer. Give it a clear goal and it will handle planning, implementation, verification, and handoff on its own. It can call Claude Code, Codex, or AdaL itself as workers, so it bundles your existing coding agents rather than replacing them - AdaL Worker (interactive exploration mode): A mode for development where the goal is still taking shape. It orchestrates the Coding, Deep Research, and Browser worker agents, switching models as needed, and moves through an ask → confirm → decide → continue loop
- Persistent memory and self-evolving skills: It remembers decisions, failures, and preferences across sessions (persistent memory), and it observes repeated work to extract and automatically codify the steps as skills (self-evolving skills)
- Multi-model support: Claude, GPT, Gemini, GLM, DeepSeek, Kimi, MiniMax, and Qwen, plus local models through Ollama and similar — all switchable with the
/modelcommand. BYOAK (bring your own API key) is supported - @Skills Hub: A skills marketplace launched in August 2026 (atskills.one). Search for skills from the CLI or the web, then install by copying and pasting a path. AdaL itself serves as the reference implementation of the “@skills” protocol
Pricing
| Plan | Monthly price | What you get |
|---|---|---|
| Free | $0 (first week only) | 7-day full-access trial |
| Pro | $20/month | Short sprints on small codebases |
| Max | $100/month | 5× the usage of Pro. Built for daily use (most popular) |
| Max+ | $200/month | 20× the usage of Pro. For power users |
| Teams & Enterprise | Contact sales | Up to 150 seats, SSO, SAML/SCIM integration, Zero Data Retention, and more |
Pricing reflects information as of August 2026. Annual billing takes 17% off, and students get a 50% discount. Check the official site for current pricing.
Pros and cons
✅ Pros
- It can call existing coding agents such as Claude Code and Codex as workers, so a team can orchestrate work while keeping the tools it already knows
- You can switch across a wide range of models — Claude, GPT, Gemini, GLM, DeepSeek, and others — and match the model to the task and the budget
- Coding, deep research, and browser work are split across specialist workers, which makes it easier to carry complex tasks through to completion autonomously
- Persistent memory and self-evolving skills are designed so that repeated work improves its own procedure each time around
- Three interfaces (CLI, desktop, browser) plus an SDK and headless mode cover everything from solo use to CI/CD integration
⚠️ Cons
- The free plan is a 7-day trial rather than a standing free tier, so continued use requires the $20/month Pro plan at minimum
- It launched in December 2025, so third-party reviews and adoption case studies are still limited
- The difference between Pro, Max, and Max+ is expressed as a usage multiplier, which makes it hard to know how many hours or tasks that actually buys until you try it
- Because the design leans on running multiple agents and models together, the sheer number of choices can be disorienting at first compared with a single AI coding tool
How it compares
| Dimension | AdaL | Claude Code | Cursor |
|---|---|---|---|
| Positioning | An orchestration layer that directs Claude Code, Codex, and others | A coding agent that lives in your terminal | An AI code editor (IDE) |
| Form factor | CLI, desktop, browser | CLI | Desktop IDE |
| Model switching | Works across Claude, GPT, Gemini, GLM, DeepSeek, and more | Centered on Claude models | Choice of several models |
| Calls existing agents | Can launch Claude Code and Codex as workers | ─ | ─ |
| Entry price for individuals | From $20/month (after a 7-day free trial) | Usage-based, or included with a Claude Pro subscription | Check the official site |
What distinguishes AdaL is that it isn’t really something you pick alongside a standalone coding agent or IDE like Claude Code or Cursor — it sits above them as an orchestration layer that directs them as “workers.” In practice, you can name Claude Code or Codex directly as the executor for AdaL Engineer mode.
Who it’s for
- Developers who want to use several coding agents — Claude Code, Codex, and others — within a single workflow
- Engineers who want to hand over a clear goal and delegate planning, implementation, verification, and handoff
- Anyone doing exploratory development where the spec isn’t settled yet and the work benefits from conversation
- Teams who want to divide specialist tasks — coding, deep research, browser automation — across agents
- Anyone optimizing for cost and accuracy by switching between multiple frontier models
Wrapping up
AdaL’s distinctive position is that it’s designed less as a single AI coding agent and more as an “orchestration layer” that directs multiple worker agents, Claude Code and Codex among them. You can move between Engineer mode, which aims to finish the job from a goal alone, and Worker mode, which progresses through conversation, and persistent memory plus self-evolving skills mean repeated work sharpens its own procedure over time. The official site says engineers from Google, GitHub, Stripe, Meta, NVIDIA, Anthropic, and OpenAI use it. It’s still a young service — it launched in December 2025 — but if you’re interested in a tool that bundles your existing coding agents rather than replacing them, it’s worth a look.