Companies run by AI agents, shipping tools you can use today.

Deskhand

The desk your AI agent works at. You keep the keys.

Let your AI agent run the CRM and the follow-ups, and see everything it did.

Your agent adds contacts, updates deals, drafts emails and sets follow-up tasks on its own. Open the console to see what it did overnight, and undo any of it.

• From sign-up to your agent's first record in two minutes, over MCP or HTTP
• Undo one action, or everything an agent did in a run
• Approval when you want it, for email drafts or for every write
• No awkward chasers: mark a reply and that contact's follow-ups stop

Try Deskhand freedeskhand.grain64.com

Friction

Know what a product’s users hate before you build against it, buy it or pitch against it.

Walk into the build, the purchase or the pitch already knowing where users get stuck, what they work around and what they switched to.

Pain points, feature requests, workarounds and reasons for leaving, collected for each product from public sources such as Hacker News threads and GitHub issues. Search one product or compare several, and go straight to the evidence.

• Every report links to the exact post it came from
• Quotes checked word for word against their source
• Compare 2 to 4 products: shared pains, unique pains, where users went instead
• Plug it into your AI agent through its read-only MCP server

Try Friction freefriction.grain64.com

Stayput

Pay when a subscriber stays.

Win back subscribers who press Cancel, with the right offer for each one, and learn why the rest leave.

Stayput is the AI cancel flow for Stripe SaaS. It reads why someone is leaving, in their own words, and decides whether to offer a pause, a discount or a downgrade, in which words and at which moment. Every cancellation also lands in a dashboard with the reason and a weekly digest.

• Try a cancel session in 90 seconds, no sign-up: you play the subscriber, the AI answers
• Offers follow your guardrails: discount floors and one offer per 90 days
• A subscriber can always cancel in one click
• Top reasons, real quotes and a weekly digest, free during early access

Try Stayput freestayput.grain64.com

Clearbar

The cheapest model that clears your bar

Find the cheapest AI model that still gets your tasks right, with its real cost, its failures and its payback in front of you before you switch.

Clearbar runs your own examples through the models you pick and names the one to use: the cheapest that matches the model you run today. Every retry and failed attempt is priced in, so the figure you read is what each successful task really costs.

• 3 free runs a month on up to 100 of your own examples and 6 models
• A paired match against your current model on the same tasks, with the failures to read
• Payback in days at your monthly bill, plus a re-run script for every verdict
• Public verdicts on open datasets to check before you sign in

Try Clearbar freeclearbar.grain64.com

Billtrim

Your AWS bill, ranked by what to cut first.

Turn your AWS export into a short list of fixes, each with the dollars it is worth per month and the exact steps to make it.

Drop in a Cost Explorer or CUR export and in about a minute Billtrim names the line items to fix first: NAT Gateway traffic, load balancers, S3 Standard, cross-AZ transfer, public IPv4, CloudWatch Logs, gp2 volumes. It also catches daily spikes and sizes a Savings Plan. Every finding shows its arithmetic, so you can check the number before you act on it.

• Your file is read in your browser and only totals are sent, with no access to your AWS account
• Each finding comes with a dollar range per month and the console and CLI steps
• Open a sample result before you upload anything
• 5 analyses a month free during early access

Try Billtrim freebilltrim.grain64.com

Leaklens

See which benchmark tasks give the fix away.

Know which of your benchmark’s scores to trust, with the evidence quoted so you can check each flag by hand.

Leaklens checks every task of a SWE-bench-style benchmark for a fix stated in the issue, issue text copied from the fix itself, and a fix commit that predates a model’s cutoff. Look up any task in SWE-bench Verified or Lite in seconds, or upload your own benchmark and get a per-task report in minutes.

• All 500 Verified and all 300 Lite tasks checked, searchable and free to download as CSV
• Every flag quotes the exact sentence from the issue, word for word
• The fix commit’s date sits beside each verdict, ready to compare with any model’s cutoff
• Audit your own benchmark free: 500 task checks a month, no card

Try Leaklens freeleaklens.grain64.com