Keld
Keld provides an observability and marketplace platform that helps businesses track, control, and optimize AI inference spending across people, apps, and agents. Its Atlas feature automatically attributes every prompt and dollar to defined workstreams without code changes, while the Marketplace routes prompts to the best-priced model that meets quality and compliance standards. Prompt content never leaves users' devices, with zero data retention and GDPR compliance.
- Artificial Intelligence
- Enterprise Software
- Software Only
- Data & Analytics
Funding
Founders
Founder details are not available yet.
Product
Problem
Businesses struggle to forecast and control their AI spending because inference is purchased off fixed rate cards, with no visibility into which teams, projects, or tasks are driving costs. AI is priced in tokens, and the number of tokens a model needs to finish a task is hard to predict, making budgets unpredictable while invoices from multiple providers arrive without context about what the spend was for.
Solution
Keld builds a marketplace and observability platform for AI work, enabling businesses to track AI usage across people, apps, and agents. The Keld Atlas feature reads each prompt, classifies it into activity types, and attributes every prompt and dollar to defined workstreams without requiring code changes, covering both personal inference tools and agentic API usage. Atlas connects spend to projects, teams, and activities, produces finance-ready reports with CapEx/OpEx splitting, and generates ranked optimization recommendations. The Keld Marketplace then routes prompts to the best-priced model from various providers that meets the company's quality and compliance standards, with prompt content never leaving the user's jurisdictional and regulatory boundaries.
Target Audience
Primary users are finance and engineering leaders at companies running AI across multiple tools, apps, and agents who need to forecast, control, and optimize their AI inference spending.
Features
- Atlas attributes every prompt and dollar to defined workstreams across people, apps, and agents, with no code changes required
- Activity-level classification reads each prompt and assigns an activity type such as coding, text creation, extraction, or Q&A to make token usage manageable
- Marketplace routes prompts to the best-priced provider model based on quality scores and price per million tokens, with price comparisons shown against list prices
- Workstream definitions can be imported from Notion, Monday.com, or Jira, or defined manually to match existing reporting structures
- Keld Signal runs on each person's machine, ensuring prompts never leave users' devices
- Zero data retention, SOC 2 certification in progress, GDPR compliance, and EU AI Act-ready controls that bound activity by jurisdiction and model origin
- Finance reporting includes total cost by workstream, team, and model, CapEx/OpEx splitting, daily CSV exports, and scheduled email reports