WaypointOpen console

A research agent with a steering wheel

AI agents you can steer, step by step.

Describe a task. Waypoint drafts a plan you can edit before anything runs, then works through it one step at a time, searching the live web and citing its sources. Pause it, rewrite a step, or re-run from any point. You stay in the driver's seat the whole way.

No sign-up. Runs in your browser, powered by fast inference on Groq.

Running step 3 of 4

Compare Notion, Linear, Asana, ClickUp, and Monday.com for a 10-person startup

Pause
  1. Collect current pricing

    Find per-seat pricing for each tool at the 10-user tier.

    Linear Business is $14/user/mo1, Notion Plus $102, ClickUp Unlimited $73…

  2. Compare integrations

    Edited by you

    Focus on GitHub, Slack, and Figma. Skip the long tail.

  3. Evaluate AI featuresRunning

    What each AI add-on actually does, and what it costs extra.

    Notion AI is now bundled into Business plans, while Asana

  4. Write the recommendation

    Rank the five tools for a 10-person startup.

Why Waypoint

Autonomous agents are a black box. Waypoint is a glass one.

Handing a task to an AI agent usually means hoping for the best. Waypoint makes every decision visible and every step editable, so the work goes where you actually want it to.

×You can't see the plan

Most agents decide what to do on their own. You find out what they were thinking only after the answer arrives.

×You can't change course

Spot a wrong assumption halfway through? Your only option is to stop everything and start over from scratch.

×You can't trust the facts

Answers drawn from training data go stale fast, especially for prices, features, and anything that changed this year.

How it works

From a sentence to a finished brief, in four stops.

Waypoint splits work into a route of small, visible steps. You set the destination, approve the route, and can take the wheel whenever you like.

  1. Step 1: Describe the task

    Write what you need in plain language: a comparison, a research question, an audit. Up to 2,000 characters, as specific as you like.

    “Compare five project tools on pricing, integrations, and AI features.”

  2. Step 2: Review the plan

    A planning model breaks your task into 3–6 concrete steps. Rewrite any step, delete what you don't need, and only then press Run.

    Nothing executes until you approve it.

  3. Step 3: Watch it work

    Steps run in order. Each one searches the live web, builds on the results before it, and streams its answer into a trace you can read as it's written.

    Every claim is tagged with a numbered source.

  4. Step 4: Steer at any point

    Pause mid-step, edit the instructions, re-run a single step, or re-run everything from a given step onward. Waypoint shows you what's out of date.

    Fix the route without restarting the trip.

Features

Everything you need to keep an agent on course.

Editable plans

Every plan is a draft. Change a step's title or instructions, remove steps, and see your edits highlighted so you always know what you changed.

Live, streamed trace

Watch each step's answer appear token by token, rendered as clean Markdown with tables and lists, plus a progress bar for the whole run.

Grounded in the live web

Every step is required to search the web before answering, so prices, features, and dates come from current sources, not stale memory.

Inline citations

Claims carry numbered source chips right where they're made, so you can see which parts of an answer are backed by what.

Pause and resume

Stop the agent mid-sentence. The partial output is kept, and Resume picks up from the next unfinished step whenever you're ready.

Re-run one step, or from here

Not happy with a result? Re-run just that step, or re-run it and everything after it so later steps build on the new answer.

Stale-output tracking

Edit a step or redo an earlier one, and Waypoint flags every downstream result that may now be out of date. No silent inconsistencies.

Context that carries forward

Each step receives the finished outputs of the steps before it, so the final step can compare, rank, and recommend with the full picture.

Fast and recoverable

Runs on Groq's low-latency inference. If a step fails, you see the error in place and can retry and continue without losing earlier work.

Reading the route

Know the state of every step at a glance.

Each step sits on a route line that fills in as work completes. Its marker tells you exactly where it stands.

  • 1

    Waiting

    Queued. It runs once the steps before it finish.

  • 2

    Running

    Searching sources and writing its answer right now.

  • Done

    Finished, and its output matches its current instructions.

  • 4

    Out of date

    Its instructions or an earlier step changed after it ran. Re-run to refresh.

  • Paused

    You stopped it. Partial output is kept until you resume or re-run.

  • !

    Failed

    Something went wrong. The error is shown in place, with Retry and continue.

Use cases

Built for research that has to be right.

Waypoint shines on multi-part questions where facts change often and you want to check the reasoning before you act on it.

Founders & operators

Pick the right tool

“Compare five CRMs for a 20-person sales team on price, setup time, and HubSpot migration.”

Product & strategy

Map a market

“Survey the main players in AI meeting notes: positioning, pricing tiers, and recent launches.”

Students & researchers

Get oriented fast

“Summarize the current approaches to battery recycling and the companies leading each one.”

Procurement & finance

Audit what you pay

“Check whether our cloud storage, email, and design tools have cheaper plans at our seat count.”

Under the hood

Two models, one route, and nothing hidden.

Waypoint is a small, open-source Next.js app. You can run your own copy in a couple of minutes with a free Groq API key.

  1. 1

    Planner

    A larger reasoning model (GPT-OSS 120B on Groq) turns your task into 3–6 steps, validated against a strict schema and retried if malformed.

  2. 2

    Executor

    A fast model (GPT-OSS 20B on Groq) runs each step with a required browser-search tool, given the task and all prior step outputs.

  3. 3

    Stream

    Answers stream straight to your browser and can be cancelled instantly. Pausing aborts the request; nothing keeps running in the background.

Run it yourself

# 1. Install
npm install

# 2. Add your key to .env.local
GROQ_API_KEY=your_key_here

# 3. Start
npm run dev

Get a key at console.groq.com/keys. Built with Next.js, React, Tailwind CSS, Zustand, and Zod.

FAQ

Questions, answered.

Do I need an account?

No. Open the console and type a task. If you self-host, the only setup is a Groq API key in your environment.

Is my task or its results stored anywhere?

Waypoint has no database. Your plan and its outputs live in your browser tab's memory and disappear when you refresh or start a new task. Your task and step prompts are sent to Groq to generate answers.

What happens to later steps if I change an earlier one?

They're kept, but marked “Out of date” so you know they were built on different input. Use “Re-run from here” to refresh the step and everything after it in one go.

Can I change the plan while it's running?

Yes. Pause the run, then edit any step that isn't currently running. Edited steps are highlighted, and any output they already produced is flagged as stale.

How accurate are the answers?

Every step searches the live web and cites sources inline, which helps a lot with fast-changing facts. It's still an AI model, so check the cited sources before making important decisions.

What does it cost?

Waypoint itself is free and MIT-licensed. Running it uses your Groq API key, which is billed (or free-tier limited) by Groq.

Give the agent a destination. Keep the wheel.

Your first plan is one sentence away. Try the example task if you need a starting point.

Open the console