COVERAGE

Every place your AI spend comes from.

Ottto reconciles usage from coding agents, cloud model providers, and gateways - subscriptions, API keys, and cloud bills, all in one number.

Start free →

Free to start · no credit card · local runtime plus connected accounts

Coding agents

Where developers actually spend.

Every supported agent, account, and machine - read from local transcript metadata, live telemetry, and connected-account evidence, then labeled and reconciled. No proxy in the request path. Built by engineers for engineers.

Claude Code

subscription + API

Pro / Max capacityTeam and Enterprise seatsAPI token and cache pricing

Claude Code

Pro, Max, Team, Enterprise, API token usage, credits, cache behavior, skills, subagents, and MCP usage all show up as one Claude-shaped problem.

Claude is not one bill. A person can have Pro, Max 5x, Max 20x, Team or Enterprise access, plus direct API usage and credits.

The same allowance can be spent from the CLI, desktop app, Pi, or another agent path, so the surface where work happens is not always the surface where spend appears.

Skills, subagents, and MCP usage add context to the session and can consume the same plan allowance or API-priced model path.

The cost math changes by token type: input, output, 5m cache writes, 1h cache writes, and cache hits. Ottto ties those selectors back to the actual Claude sessions on the machine.

Codex

ChatGPT + usage

ChatGPT plan inclusionBusiness Codex usageAPI model rates

Codex

Codex spans ChatGPT plans, CLI sessions, app sessions, Pi or other agent routes, Business usage pricing, credits, and direct OpenAI API model rates.

Codex can arrive through ChatGPT Free, Go, Plus, Pro, Business, Edu, or Enterprise, while Business Codex adds a usage-priced path.

Developers can spend it in the CLI, app, IDE/cloud workflows, Pi, or another agent, while the evidence may land in plan seats, credits, or API usage.

OpenAI API usage adds model rates, cached input, batch discounts, data-residency modifiers, fast-mode behavior, reasoning-effort selectors, and credit balances.

Ottto keeps local Codex sessions, ChatGPT/Codex plan evidence, API spend, and credits in one reconciled picture.

Cursor

included + on-demand

Pro / Pro+ / UltraTeams and EnterpriseOn-demand usage

Cursor

Cursor mixes individual plans, Teams seats, Enterprise controls, included model usage, on-demand overage, Bugbot, and cloud agents.

Cursor is partly a seat product and partly a usage product. Plans include model usage, and on-demand usage can continue after included usage is consumed.

Teams adds centralized billing, usage analytics, privacy mode, SSO, shared context, and cloud automations. Enterprise adds pooled usage and deeper controls.

Ottto treats Cursor as a connected-account source so team usage and plan fit can sit beside local agent sessions.

route evidence

Local setupModel routeBilling channel

Pi is source and route evidence: local setup, session files, live path, model route, provider, billing provider, and auth mode.

Pi does not behave like a tidy public seat table. The important evidence is which route ran, which provider backed it, and which billing channel paid.

Ottto records route-level details: model provider, billing provider, source category, context/output size, auth mode, and local evidence when available.

That lets Pi sessions reconcile with OpenAI, Gemini, Vertex, Bedrock, or gateway-backed routes without flattening them into one vague app total.

Clouds, APIs & gateways

The bill side of the ledger.

Cloud model providers and pay-as-you-go APIs get pulled in too, so a Bedrock line item and a Claude Max seat land in the same total - not two dashboards.

aws AWS

CUR + Cost Explorer

Cost Explorer + CUR 2.0CloudWatch + invocation logsAWS Marketplace LLMs

AWS

AWS AI spend can land in Bedrock, SageMaker AI, EC2/GPU, ECS/EKS/Lambda gateways, logs, storage, data transfer, and Marketplace charges.

Cost Explorer is the primary bill view, but default Bedrock can be coarse. Per-model split needs usage type or Application Inference Profiles with cost-allocation tags.

CUR 2.0 is the finest cost path: per-model rows, token direction, cache read/write, service tier, region, tags, and IAM principal when enabled.

CloudWatch and invocation logs show tokens/calls without dollars; Budgets and Cost Anomaly Detection alert on spend. Some LLMs are first-party Bedrock, others are AWS Marketplace.

GCP Vertex AI

cloud billing

Billing Reports + BigQueryMonitoring + LoggingProject / labels / SKU

GCP Vertex AI

Vertex cost can be Gemini, partner models, endpoints, projects, BigQuery billing export, cache storage, grounding extras, and labels.

Cloud Billing Reports show service, SKU, project, and label totals; BigQuery export is the granular SQL path over SKU-level rows.

Cloud Monitoring shows token/request counts by model and publisher; Cloud Logging carries call metadata, not the bill.

Budgets, alerts, quotas, project, region, endpoint, and partner-model attribution stay in cloud billing. Ottto reconciles that with model usage.

Anthropic API

token selectors

Input and output5m / 1h cache writesCache hits and batch

Anthropic API

Claude API pricing is selector-heavy. A cache read is not an input token, and a 1h cache write is not a 5m cache write.

Batch, data residency, service tier, model family, and long-context behavior can change the right rate for the same apparent session.

Ottto preserves those selectors so a Claude Code session and an Anthropic API invoice can meet in the same cost story.

OpenAI API

models + credits

Model ratesFast / reasoning selectorsCredits and batch

OpenAI API

OpenAI API pricing changes by model and processing mode, including cached input, batch discounts, and data-residency modifiers.

A personal user or an organization can have ChatGPT seats, Codex usage, credit balances, and API keys, each with a different billing path.

Fast-mode behavior, reasoning-effort choices, cached input, and output tokens can make two similar-looking turns land at different effective rates.

Ottto separates those paths first, then reconciles them so Codex and OpenAI API spend are comparable without being confused.

Gemini

API + Vertex

Free and paid tiersCaching storageSearch and Maps grounding

Gemini

Gemini spans free and paid API tiers, Vertex paths, context caching, thinking/output tokens, audio/video/image inputs, and grounding fees.

Gemini pricing varies by model, modality, token direction, context cache storage, and whether grounding with Google Search or Maps is involved.

Gemini API and Vertex AI are different cost paths even when the model name looks familiar.

Ottto keeps the Google route, project, billing channel, and model pricing evidence visible so estimates do not hide the selector complexity.

Why this gets hard

Every source has its own billing maze. Each source looks like a single label until you manage it: plan windows, seats, credits, API token rates, marketplace invoices, cloud billing exports, monitoring logs. Keeping up is tiresome because the evidence is fragmented. We all increase our AI usage all the time, using it for more and more purposes, but because we add agents bottom up, it comes with operational complexity we need to manage.

One daemon. Every source.

Subscriptions, API keys, and cloud bills across Claude Code, Codex, Cursor, Pi and your clouds - reconciled into one number you can trust, with source labels still visible.

Start free → Get the Mac app

Free to start · no credit card · Apple Silicon · macOS Sonoma+