Data Loom

Introducing Data Loom 🦉 Private AI meets proof of data. Turn human knowledge into AI answers through Phala Confidential Compute. Earn without selling your source files. 70% of each paid query goes to contributors, settled in USDC on Solana. CA : 8SymmzsWdQBweqCK1p35MGNCwgw8FgPRdL4KPckdata

Does private data make your AI better? 🦉 Data Loom’s Data Impact Trials compare the same model on the same cases, with and without a private tool. Compare accuracy gains, category results and tool costs. See an illustrative walkthrough below.

Data Impact Trials are live on Data Loom 🦉 Would a private warranty tool help your AI model make better eligibility decisions? And would the improvement justify the cost of calling the tool on every case? Data Impact Trials let you measure that. Pick one model and one private benchmark. Data Loom evaluates the same model on the same cases twice. In the baseline run, the model answers on its own. In the assisted run, the private tool receives the structured inputs for each case, and the model receives the tool’s decision before answering. The scorecard shows baseline accuracy, assisted accuracy, the change by category, and the cost of the tool calls. For example, a trial might show that accuracy rose from 60% to 80% on refurbished products and stayed flat on standard returns. That gives your team a concrete way to decide where the tool earns its place. Every model response and tool call requires a verified Phala inference receipt. The scorecard includes those receipts and aggregate results. Individual test cases, reference answers, tool source files, and model outputs stay private. Contributors opt in by pairing a private benchmark with a tool they own. Buyers approve one USDC payment on Solana to unlock the scorecard, and the contributor receives 70% of the payment. Open Marketplace → Blind Model Trials → Data Impact Trial. Try it on Data Loom

Data Loom shipped four features in the last 24 hours 🦉 AI teams can now compare models across private benchmarks, give agents access to contributor-approved tools, run up to 200 cases in a single batch, and pay for eligible confidential answers with $DATALOOM. 1. Private Benchmark Pools Compare two models across 2–3 benchmarks from different contributors, with up to 30 cases in a single run. You get one aggregate scorecard while contributors keep their test cases and answer keys private. Each model response requires a verified Phala inference receipt. Contributors split 70% of the payment based on how many cases each supplied. Benchmark Pools 2. Private Inference API for Agents A contributor can turn a private dataset version into a tool with defined inputs and allowed decisions. Think of a warranty tool that returns eligible, ineligible, or needs review. The contributor grants a buyer’s Solana wallet access with a USDC spending cap. The buyer can then create an API key for that grant, and each paid call returns a structured decision, a short reason, and a verified Phala inference receipt. API guide 3. Private Batch Inference Submit 2–200 cases to one private tool and approve a single USDC payment on Solana. Download one JSON file with each case’s row number, decision, reason, and verified Phala receipt. The export leaves out the submitted inputs. Contributors enable batch access per grant and receive 70% of the payment. Buyers can prepare one batch per dataset per day. Batch Inference guide 4. $DATALOOM payments for confidential answers Contributors can add a $DATALOOM price to eligible confidential query offers. Buyers can choose $DATALOOM when the offer supports it, and 70% of the payment goes to the contributor in $DATALOOM. Each purchase unlocks an answer from the approved dataset version with a verified Phala inference receipt. Query marketplace

Private Batch Inference on Data Loom supports 2–200 cases, with a verified Phala receipt per case. Original inputs stay out of the export. Contributors earn 70%. Watch the example workflow 👇

Private Batch Inference is live on Data Loom 🦉 Have 200 warranty claims to review? Submit them together to a private inference tool and approve one USDC payment on Solana. Download a single JSON file with a decision and a verified Phala inference receipt for each case. The export leaves out the original case inputs. Each batch supports 2–200 cases. Contributors choose which grants allow batch access and receive 70% of each payment. Buyers can prepare one batch per dataset per day. Find it under Account → Manage tools and grants → Private Batch Inference. Open Data Loom

Pay for confidential AI answers with $DATALOOM 🦉 $DATALOOM is now an optional payment currency for confidential queries on Data Loom. Contributors can set a $DATALOOM price alongside their USDC price. When a buyer pays in $DATALOOM, 70% goes directly to the contributor and 30% goes to Data Loom, all in the same Solana transaction. Data Loom verifies the Phala inference receipt before preparing checkout. Once the payment finalizes, the buyer gets the answer and its receipt. The source dataset stays with its owner. Contributors can opt in now under Account → Query Access Policies. Buyers will see $DATALOOM at checkout on offers that accept it. Solana mint 8SymmzsWdQBweqCK1p35MGNCwgw8FgPRdL4KPckdata

Data Loom’s Private Inference API lets agents request structured decisions without downloading the underlying dataset. Watch an illustrative refund check workflow below. Contributors control access, pricing and earn 70% per paid call.

Introducing the Private Inference API for Agents on Data Loom 🦉 An AI agent can now call a tool backed by a contributor’s private dataset. The contributor defines what the tool can answer, which wallet can use it, and what each call costs. Imagine a support agent that needs to check whether an order qualifies for a refund. It calls check_refund_eligibility with the country, order age, and product type. Data Loom runs confidential inference and returns a structured decision such as eligible, ineligible, or unknown, along with a short reason and a verified Phala inference receipt. The agent receives the decision without downloading the underlying policy. How it works Create a tool. Choose one private dataset version, define the input fields and allowed decisions, and set a price per call. Grant access. Authorize a buyer’s Solana wallet and set a total USDC spend cap. The buyer creates an API key scoped to that grant. Access can be revoked at any time. Call the tool. The agent submits typed inputs. Data Loom verifies the Phala inference receipt before preparing checkout. Pay and unlock the result. The authorized buyer wallet signs the USDC transaction on Solana. Once payment finalizes, the agent can retrieve the decision and receipt. The contributor receives 70% of each paid call. A refund policy is one example. The same workflow can serve product compatibility checks, warranty decisions, or internal support rules. Contributors can earn each time an agent uses their knowledge while keeping control over who gets access. Create your first tool under Account → Private Inference API for Agents. Open Data Loom · Read the API guide dataloom.live/tools/private-…

Here's a demo for our new feature "Private benchmarks" Compare two AI models across 2–3 private benchmarks in one run. Contributors share 70% of each purchase by case count. Contributor opt in is open. Pools need at least two contributors. dataloom.live/#marketplace

Introducing Private Benchmark Pools on Data Loom 🦉 An AI team can compare two models across private benchmarks from multiple contributors in one run. A single evaluation could cover billing questions, multilingual support, and troubleshooting, with one scorecard showing how each model performed across the combined test set. How it works Contributors opt in. Each contributor chooses which dataset version can be used in pooled evaluations and can pause future use. Buyers build a pool. Pick 2–3 benchmarks from different contributors, with up to 30 cases total, then choose two models to compare. Data Loom runs the evaluation. Each inference requires a verified Phala receipt. The scorecard shows aggregate results, while prompts, reference answers, and individual model responses stay private. One payment unlocks the scorecard. Buyers pay in USDC on Solana. Contributors share 70% of the purchase, weighted by how many cases each contributed to the run. Contributors can opt in now. A pool becomes available once at least two contributors have opted in. Find Private Benchmark Pool under Marketplace → Blind Model Trials. Open Data Loom

This Data Loom example confirms a reviewed fictional dataset, saves a local draft and downloads its reviewed JSON. The archive shows the saved result.

Choose what leaves your source. 🦉 Using fictional research text, this Data Loom example detects contact details and manually removes project information while preserving the useful evaluation method.

Your source. Your edits. Your decision. See a fictional Data Loom contribution edited manually and saved as a reviewed draft in the local archive. Saving the draft does not publish it. dataloom.live/#import

Introducing Private Release Gates on Data Loom 🦉 Would you ship a model that scores 92% overall but drops 12 percentage points on Spanish support? With Private Release Gates, you can set both limits before the evaluation starts. Open Blind Model Trials and choose a private benchmark, a baseline model and a candidate model. Attach a release gate to your commit SHA. Set a minimum accuracy for the candidate and, optionally, the maximum drop you’ll accept in a specific category. Data Loom locks those rules before the run. Every model response requires a verified Phala inference receipt, and the benchmark’s test cases and answer keys stay private. Once your wallet payment finalizes, download the signed pass or fail receipt. The CI verifier checks the signature, trial ID, commit SHA and expiry. A failed or invalid receipt can stop the deployment. Benchmark contributors receive 70% of each paid evaluation. This first release works with Data Loom’s supported hosted models. The commit SHA ties the receipt to your release workflow, and you supply the receipt to CI.

What should stay out of your contribution? This Data Loom example detects and redacts an email address and phone number in fictional text, directly in the browser. Automated detection is a first pass; review the full copy before sharing. dataloom.live/#privacy

From a product question to a traceable source. 🦉 Filter Data Loom’s Planning collection, open the sample workflow and inspect its source details.

Follow the source behind a creative workflow. 🦉 Explore Data Loom’s Creative sample, switch between its conversation path and metadata, and inspect storage, license and publication status.

A weak benchmark result can become a request for better data. 🦉 Data Loom Dataset Gap Bounties collect private evaluation cases. Buyers unlock selected packs with USDC; selected contributors split 70% of the award.

Share the proof, keep the answer private. 🦉 Data Loom Proof Cards link dataset versions, authorized use and inference receipt status. Share a link, then revoke it when needed.

Data Loom Model Regression Watch compares private benchmark runs against your baseline and flags accuracy drops. Set daily or weekly checks. Approve each USDC unlock. dataloom.live/#marketplace