AI coding FinOps + agent observability

One meter for every supported AI coding agent you run.

Ottto reconciles local activity, subscriptions, provider usage, credits, and cloud bills—then keeps the source, freshness, coverage, and billing basis attached as you investigate what deserves attention.

Start free Download the macOS app

Illustrative values below. Out of the request path. No automatic changes.

Accounts & usage

Claude CodeObserved local sessions $840
CodexObserved local sessions $620
CursorTeam usage evidence $480
PiObserved local sessions $190
Anthropic APIOrganization analytics $510
$3,000 illustrative monthly total reconciled

Cloud & credits

AWS BedrockCost Explorer billing $280
Google VertexSelected billing export $120
Provider creditsWhere reported −$40

Supported agent evidence

Claude CodeCodexCursorPi + provider and cloud evidence

Illustrative example. No customer data. Availability and evidence depth vary by source.

Recurring agent work

See which workflows repeat—and what recorded runs cost.

Ottto groups repeated sessions only when evidence supports the relationship. Review observed cadence, runs, spend, tokens, models, machines, and freshness together.

nightly-code-review

checkout-service · illustrative recurring group

Observed group
Observed schedule
Daily · 02:00
Runs found
21
Recorded spend
$87.40
Tokens processed
18.2M
Models observed
2
Machines observed
3
Cost by observed run Most recent 4h ago
14 runs agoLatest

Evidence, not outcomes. Run history does not claim a merge, comment, or useful result. Ottto does not pause or cancel the schedule for you.

Illustrative group and values. Not customer data.

One cost layer

Reconcile subscriptions, APIs, cloud bills, and local activity.

Compare supported sources in one view while origin, freshness, and confidence remain visible. The source mix and values below are illustrative.

Claude Code

Session records · refreshed 2m ago

Exact

Codex

Session records · refreshed 4m ago

Exact

Cursor

Usage records · refreshed 12m ago

Observed

Pi

Session records · refreshed 18m ago

Observed

AWS Bedrock

Cost Explorer billing · refreshed 6h ago

Imported

Google Vertex

BigQuery billing export · refreshed 8h ago

Imported
6machines in view
14repositories in view
8connected sources

Illustrative source summary. Not customer data.

Observe, reconcile, improve

Turn fragmented agent activity into a defensible next step.

  1. 01

    Observe

    Read supported local telemetry and billing evidence.

  2. 02

    Reconcile

    Join sessions, families, repositories, machines, subscriptions, APIs, and invoices.

  3. 03

    Trace

    Keep source, freshness, confidence, and estimation method attached.

  4. 04

    Diagnose

    Compare context, configuration, caching, recurring work, and provider changes.

  5. 05

    You decide

    Review evidence-backed recommendations. Nothing changes without your action.

From cost to cause

See what drives the bill. Know what to inspect next.

Ottto connects spend to observed workflow behavior, then keeps recorded facts, derived estimates, and recommendations distinct.

Context + configurationObserved footprint

See which workspaces carry the heaviest context.

Compare loaded files, memory, MCP and tool configuration, context pressure, and compaction evidence across repositories.

checkout-service2.1× example median
catalog-apiexample median
docs-sitebelow example median
Cache activityObserved behavior

Spot shifts in cache reads and writes.

Compare recorded cache activity across periods, then inspect the source evidence before changing a workflow.

Cache reads↓ 24%illustrative change
Cache writes↑ 17%illustrative change
Agent lineageRecorded relationships

Trace cost from a root agent through its children.

Use map, timeline, list, and detail views to inspect family-level tokens, cost, models, and duration.

orchestrator
researchreviewtests
12 example children · $31.70 illustrative spend

All signals and values shown are illustrative—not customer results or promised savings.

Local workflow visibility

See how agent work unfolds across your machines.

The local app turns supported, content-free usage metadata into inspectable sessions, agent lineage, costs, tokens, models, machine activity, and repository attribution.

  • Observe local coding-agent activity without proxying model requests.
  • Attribute supported sessions to machines and repositories.
  • Trace recorded relationships between root agents and their children.
  • Compare local evidence with subscriptions, APIs, and cloud bills.

Coverage, available fields, and attribution depth vary by source.

Illustrative terminal
$ ottto status --json
{
  "protocol_version": 15,
  "daemon": "running"
}

Illustrative command output. Available fields vary by supported source.

Human control

Ottto explains and recommends. You choose what changes.

Use the evidence to decide what to keep, tune, or investigate. Ottto does not act silently.

✓

Ottto does

  • Reconcile supported local telemetry and billing evidence.
  • Preserve source, freshness, confidence, and estimates.
  • Group recurring work and map recorded agent lineage.
  • Surface traceable patterns and recommendations.
×

Ottto never

  • Sits in the model request path.
  • Equates repeated work with a useful outcome.
  • Cancels schedules or edits repositories automatically.
  • Turns missing or stale evidence into zero.

For one stack or a team

Start with your own agents. Add a shared view when the team needs one.

For individuals

Make sense of your AI coding stack.

  • Bring local tools, subscriptions, and provider usage together.
  • Compare machines, repositories, sessions, and agent families.
  • Keep recorded costs and estimates clearly labeled.
Start free

For teams

Put engineering and finance on the same cost evidence.

  • Review supported subscription, API, cloud-billing, and local evidence within connected-account and shared-report scope.
  • Connect provider changes to affected usage and workflows.
  • Prioritize evidence-backed opportunities without automatic changes.
Request team access

Before you connect

Know what Ottto sees—and what it does not do.

Clear answers on data, estimates, recurring groups, and control.

Will Ottto slow down my agents?

No. Ottto reads supported local telemetry and billing sources without proxying model requests or changing how your agents run.

What data does the local app read?

For supported sources, it reads content-free usage metadata such as timestamps, token counts, model identifiers, costs, session relationships, and machine or repository context. Available fields vary by source.

How does Ottto group recurring work?

Ottto groups repeated runs only when observed evidence supports the relationship. It shows the evidence and confidence without declaring the workflow stale, claiming an outcome, or cancelling a schedule.

Why are some costs exact and others estimated?

Invoices and provider bills can be exact. Local usage, subscription allocation, and incomplete source data may require estimates. Ottto keeps each confidence level visible instead of blending everything into an unexplained total.

Does Ottto work for individuals and teams?

Individuals can review their own connected tools, machines, and subscriptions. Teams can invite admins and members, then share selected saved reports with named teammates through read-only grants.

Start with what is measurable

Know what supported agent work costs. Review what deserves attention.

Connect supported sources, reconcile the bill, and inspect the workflow evidence behind your next decision.

Start free Get the Mac app