Valistry

Start read-only. Connect deeper as trust grows.

Read-only API access shows your spend and your AI apps. A browser extension or npx valistry shows who uses what. A short script shows how coding sessions run.

Start freeAsk about a source

One ledger

reconciled

$56,282

In this sample, every AI dollar in one total.

  • Has an owner $53,126
  • No owner yet $3,156
Claude APIOpenAIAWS BedrockGitHub CopilotCursorOpenRouterAzure OpenAIClaude EnterpriseGoogle Vertex AIFireworks AIxAIAWS cost and usageGoogle Cloud billingAzure cost managementLangSmithBraintrustW&B WeaveChatGPT EnterpriseCodexChatGPT EnterpriseMicrosoft 365 CopilotGemini Code AssistWindsurfGitLab DuoPerplexity EnterpriseQdrant Cloud
Prolific human dataZendesk resolved casesFreshdesk resolved ticketsIntercom closed conversationsSalesforce opportunity stagesGitHub merged pull requestsGitLab merged merge requestsJira delivered issuesn8n workflow runsTwilio completed callsArrowhead call outcomesYellow.ai resolved conversationsSend results by webhookDevRev closed workPortkey gatewayLangfuseLiteLLM proxyHeliconeCloudflare AI GatewayOpenTelemetryBrowser extensionLinear completed issuesVercel AI GatewayDatabricksSnowflake CortexGoogle Workspace directory

Go deeper, get sharper recommendations.

We count use and never keep prompts or replies.

  1. Light access

    Recommend

    How you give it

    Read-only API access

    Or an export. No install.

    What we read

    Spend, tokens, models, seats, timings, and the tags a source carries. AI apps by name, and how many people use each.

    What you get

    Cheaper models, caching and unused seats. Tools nobody approved. You make each change.

    Works with

    Model vendors, cloud, gateways, directories and exports.

  2. Medium access

    Copilot

    How you give it

    Browser extension, npx valistry or a short script

    IT installs them. An agent may write to GitHub or Jira.

    What we read

    Site visits and message counts, per person and day. Claude Code and Codex use. Counts from each coding session.

    What you get

    Shadow AI, person by person. A fix for each developer. A pull request or an issue, drafted for you to finish.

    Works with

    42 AI sites, desktop apps, Claude Code, Codex, GitHub and Jira.

  3. Deep access

    Autopilot

    How you give it

    A key or app for each agent

    For its one job. You set the limits.

    What we read

    Only what each agent needs for its job.

    What you get

    Agents set limits and models, and remove seats you choose. Each can be undone.

    Works with

    LiteLLM, GitHub Copilot, Microsoft Entra and Okta.

Light shows tools nobody approved. Medium shows who uses them.

What each source reads.

  • Claude APIapi keyworkspacemodelClaude Code person, terminal, sessions, edits accepted and rejected
  • OpenAIapi keyprojectmodelline item
  • Fireworks AIdaymodelAPI keyprompt, completion and cached tokens
  • xAIdaymodelendpoint kind (chat, image)what xAI charged
  • Coherethrough a gatewaymodelday

Every dollar gets an owner.

When a source does not name an owner, a rule or a trace adds one. Until then it shows as unallocated.

Two other ways in.

Send any usage or billing export, or point OpenTelemetry at us. Both land in the same ledger, reconciled the same way.

Connections start read-only.

The least access a vendor allows. Our code refuses any address a connector did not list.

How that is checked →

Start with one connection. See what your AI costs.