The self-improving AI harness for teams

Turn team feedback into better AI agents.

For supported classification skills, Longhand uses your team’s corrections to test updates. Your admins decide which changes to publish for everyone who uses the skill.

Bring one workflow. We’ll show you where to start.

Set up shared tools, model defaults, and API spending limits once. Switch models, use eligible subscriptions, or run compatible open-weight models on your own hardware.

See Longhand in action

Product overview · 2:19

A closer look at the workspace.

Explore real app screens, workflow tools, and local model selection.

A product overview, including private, on-device AI. Watch on YouTube

Want updates first? Join the waitlist. Already have access? Download for Mac or Windows.

Fix an AI result today. Improve the skill for future runs.

When your team corrects AI work, Longhand can use that feedback to test changes to the instructions the team shares.

Enable skill improvement for configured classification skills on supported model routes.

Illustrative example · Support request classification

  1. Capture the correction

    A reviewer spots a duplicate charge and changes a support ticket’s category to Billing. When the corrected result is approved, Longhand saves it as an example.

  2. Measure the proposed update

    Longhand proposes checking for duplicate charges before assigning a category. It tests both versions of the skill on examples set aside for evaluation and compares their error rates.

  3. Approve and share the better version

    An admin reviews the results and decides whether to publish the tested update. Once published, the new version is available to everyone who uses that shared skill.

Set up your team’s AI in one place.

Share tools, choose model defaults, and manage API spending from your workspace.

Shared tools and model defaults

Connect tools once and give the right teams access. Choose a default AI provider and model for members who haven’t picked their own.

Budgets for metered API runs

Set shared daily and monthly budgets, plus per-member limits by workflow, day, and month.

Longhand checks recorded cost before each metered step. In-progress usage can add cost, and unavailable prices can leave gaps in estimates. These limits apply to metered API usage.

Tool permissions and human review

Preapprove selected tool actions for skills, queue runs, and workflows. Add review checkpoints so staff can correct results and decide whether the workflow should continue.

Switch models.
Keep your workflows.

Switch among supported models as your quality, cost, and privacy needs change. Choose different models for individual workflow steps where supported.

You can export the skills, agent configurations, and workflows you build.

Bring your own API keys
Connect your own API keys to use supported cloud models from OpenAI, Anthropic, Google, and xAI.
Use an eligible subscription
Where enabled, use Claude Max through Claude Code or an eligible ChatGPT subscription through Codex. Personal plans require someone to be present during the run.
Run open-weight models privately
Run compatible open-weight models on hardware you already own if it can support the workload. Local inference has no per-token model API charges. Connected services may still contact their providers.
Explore Owned AI on your hardware

Longhand Academy

Your LMS, with an AI coach and skills built in.

Bring learning and practical work into one platform. Academy is Longhand’s learning management system (LMS), combining structured courses, an AI coach, and guided skills practice to help your team apply what they learn.

Courses & progress

Give your team a clear learning path.

Publish courses with video, written lessons, and reference materials. Manage course access for your organization while learners follow their progress through the lessons.

AI coach

Ask questions beside the lesson.

Clarify a concept and explore how it applies to your work. The AI coach uses the course and lesson material for context, so the conversation stays connected to what you’re learning.

Skills & guided practice

Put the material to work as you learn.

Try guided activities built around the course’s skills and frameworks. Capture notes, turn takeaways into tasks, and practice applying the lesson to real work.

Learning belongs beside the work.

Learn in the same app where your team builds skills and runs workflows. Pair Academy with live workshops and coaching when your team needs guided support.

Book an AI strategy session

Owned AI · Private deployment

Your business knowledge.
Your private AI.
Your hardware.

For workflows that need private models, train on your organization’s knowledge and run on your own hardware. Owned AI is one deployment option within Longhand’s model-agnostic platform.

Train for your work
Train or fine-tune private models on your organization’s knowledge and processes.
Run on your hardware
Keep model training and inference inside an environment your organization controls.
Choose your deployment
Match the model and hardware to your workflows, data needs, and team.
Read the Owned AI deployment guide Book an AI strategy session

Bring a workflow. We’ll discuss a private deployment that fits.

Inside your environment

  1. Your business knowledge

    Authorized material and examples you choose for model training.

  2. Private model training

    Train or fine-tune on hardware you control.

  3. Local inference

    Run the trained model as part of your team’s work.

Model training and inference stay on your hardware.

Privacy here describes the Owned AI model path. Optional cloud models and connected services use their own data paths. See how data is handled.

Enterprise workflow example

A classification workflow, demonstrated.

An enterprise demo explored how agents could help a team facing a 2,000-item classification backlog: gather evidence, make an initial assessment, and bring uncertain work to an analyst.

Research time was reported at under one minute per analysis in the demo. That timing covers research; assess full turnaround, accuracy, and review effort against your own baseline.

These figures describe the demo scenario and its benchmark. Production impact has yet to be established here.

  1. 01

    Gather the evidence

    The agent researches the item and brings the source material into the classification process.

  2. 02

    Make an initial classification

    Apply the workflow’s criteria and identify results that need a closer look.

  3. 03

    Put judgment with the analyst

    Route uncertain results to a person for review and capture their corrections.

Built from the AI playbook used by leaders from

  • Vercel
  • LaunchDarkly
  • Fivetran
  • GitLab
  • Calendly
  • Amazon
  • Nike
  • Netflix
  • Dell
  • Salesforce
  • Red Hat
  • Peloton
  • Stryker
  • Worldline
  • LegitScript
  • Cash App

Features & benefits

Built for the work you do again.

Build your team’s capability in Academy and run repeat work with connected tools, model choices, and review points you control.

Reusable skills & workflows

Stop rewriting the same instructions.

Save a task as a skill, connect skills into a multi-step workflow, and carry context from one step to the next. Reuse the process when the next report, research request, or review arrives.

Academy LMS, AI coach & skills

Turn learning into daily practice.

Bring courses, an AI coach, and guided activities built around lesson-linked skills into one learning workspace. Help your team understand the material, apply it to their work, and follow their lesson progress.

Explore Longhand Academy

Connected apps & documents

Spend less time copying between tools.

Give workflows the email, documents, spreadsheets, and calendar context they need, then save results back to connected tools. Start with the Google Workspace accounts you authorize.

See how Google access works

Recurring schedules

Put repeat work on a schedule.

Schedule a saved skill or workflow instead of starting it manually each time. Run it on demand, pause it, or change the schedule as priorities shift.

Desktop schedules run while Longhand is running, subject to your company and provider settings.

Human review & approval controls

Keep judgment with your team.

Choose where a workflow pauses for a person to review the result. Approve or reject work at those checkpoints, and configure which tool actions you trust to run without another prompt.

Cloud & local model choice

Match the model to the work.

Choose supported cloud models or compatible models on your own computer. Use different models for different workflow steps as your quality, cost, and data needs change.

Explore your model options

Run history & API spending limits

See what ran. Track API spend.

Inspect step status and model usage, review available API cost estimates, and set workflow, daily, and monthly spending limits for metered API runs.

Limits are checked between steps; usage already in progress can add cost. Estimates depend on available model pricing.

Bring one process your team repeats. We’ll explore how these features fit the work.

Book an AI strategy session

Bring your Google Workspace into Longhand.

Sign in with Google to identify your Longhand profile. Separately connect Google Workspace to let your assistant and workflows work with the Google data you authorize.

Gmail
Find and read messages, draft and send email, manage labels, and move messages to trash.
Google Drive
Find and read files for your work, download or export them, and create, copy, or update files.
Google Sheets and Docs
Read source material, analyze spreadsheet rows, and create or update spreadsheets and documents.
Google Calendar
Show your schedule, prepare for meetings, and create, update, or delete events.

The desktop connection requests broad read and write permissions for these services so workflows can use existing content and save results. Scheduled workflows and calendar sync can run in the background. Write actions follow your approval and saved tool-trust settings.

When you use cloud AI, relevant Google content can be sent to the selected model provider to complete your task. Trial or company AI can route through Longhand servers. Local models process AI requests on your device; Google connections still contact Google.

Connection settings let you disconnect an account. Read our Google data disclosures for permissions, storage, sharing, and deletion.

Before you get started

A few practical questions.

How skill updates work, which models you can use, and what your team owns.

Already have desktop access? Download Longhand. Waiting for access? Join the waitlist.

How does Longhand turn corrections into shared skill updates?

With skill improvement enabled for a supported classification skill, Longhand captures approved corrections and measures proposed updates against the current version on past examples. An admin can approve and publish the tested update to the people who use the skill.

Which AI models can we use?

Choose supported cloud models from OpenAI, Anthropic, Google, and xAI using your own API keys. Where enabled, you can also use an eligible provider subscription. Run compatible open-weight models on your own hardware when you need private inference.

What does Owned AI include?

Owned AI supports private model training or fine-tuning and inference on your own hardware. You choose authorized business knowledge for training and control where the model runs. Connected data remains subject to its provider’s usage rules. It is an available deployment option within Longhand.

What hardware do we need?

Requirements depend on the model, whether you are training or running it, the size of your data, and how many people will use it. We’ll discuss those needs in your strategy session and scope the hardware before a deployment.

Do we need to train a model before we can start?

No. You can start with an existing model and build reusable skills and workflows around your processes. Adding instructions, source material, or a standard operating procedure to a skill is different from training a model. Private training or fine-tuning is an option when your use case calls for it.

Who handles setup and support?

We’ll use the strategy session to discuss your team’s needs, model choices, and deployment responsibilities. We’ll agree on what your team will manage and where Longhand will help before implementation.

Do we own the workflows we build?

Yes. The skills, agent configurations, and workflows you build are yours, and you can export them. Your workflow ownership is independent of the model provider you choose. Longhand lets you select models for the work and keep your team’s processes under your control.

How does our team stay in control of AI work?

Workspace admins can share approved tools, set a default model, and set shared and per-member limits for metered API runs. You can add approval checkpoints wherever people need to review, correct, or approve work.

Start with one workflow.

Bring a repeatable process to a 15-minute call. We’ll look at where AI can help, where your team should review the work, and what to measure as you improve it.

Want updates on desktop access?

Join the waitlist for desktop access updates. Already have access? You can download Longhand now.

Loading the signup form…