Skip to content
Browse docsSources

Documentation Sources

Know where every number stops.

Workspace connections and local CLI readers have different setup and availability. For both, source support, reader validation and the evidence behind a number remain separate.

Updated October 7, 2026

Agree what the report can cover

With your onboarding contact, record the workspace, connected accounts, currency, inclusive UTC dates, source freshness and known missing days. Confirm source-specific administrative permissions before connecting; do not send provider keys in email or chat.

  • OpenAI and Anthropic API costs: invited setup, subject to accessible provider records and retained history. Project detail depends on what those records contain.
  • Historical limits: available history varies by connected source. Check coverage before treating a subtotal as a complete-period answer, and do not add overlapping records.
  • Coding activity: optional Claude Code and Codex sharing needs explicit setup and consent. Sessions and repository labels do not allocate the provider bill.
  • Cursor and Copilot: separate pilot checks are required. Broad real-account coverage is not established; do not depend on these facets without explicit acceptance for your setup.

If the requested period is incomplete, keep the gap visible. Agree a narrower useful review explicitly or pause that question. An unavailable answer is not zero spend. Prepare the review →

Three axes, one source row

Availability

Published, experimental, beta, planned, or unavailable in the version you are running.

Reader validation

live_verified, fixture_verified, untested, or failed—the path’s test coverage.

Financial evidence

verified, estimated, detected_unverified, or missing—the basis of a particular number.

A live-verified local reader still produces API-equivalent estimated values, not billed spend. Approving a folder is a separate read boundary and verifies neither the reader nor the money.

In CLI output, verified means provider-reported and source-authoritative; it does not guarantee a final invoice. estimated states a calculation or qualified billing basis, detected_unverified is an unreconciled signal, and missing means a supported cost basis is absent. Workspace reports use their displayed source labels and coverage.

Hosted sources and optional shared activity

Workspace access is invitation-only. Source setup and availability are confirmed during assisted onboarding; a published CLI connector does not automatically enable that provider in Workspace.

OpenAI and Anthropic

Available through invited onboarding

Provider-reported API costs and source-specific usage

Connected admin sources support cost views with their reported dates, account scope, freshness and coverage. A known subtotal can exclude unavailable days; it is not a final invoice.

Claude Code and Codex activity

Optional machine sharing

Machine-reported sessions and tokens

Explicitly shared local activity can show sessions, tokens, days and repository labels where available. It stays separate from provider cost and does not allocate the bill to a person or agent.

Cursor and GitHub Copilot

Ask about pilot availability

Separate estimates; live partner validation pending

These connections are not generally enabled and require separate partner setup. Where enabled, Cursor cycle-to-date on-demand spend and Copilot current-month net charges keep their own periods and are not added to the OpenAI/Anthropic cost total or to each other.

Jev, Kimi and other planned integrations are not included as supported Workspace sources here. Historical availability depends on each connected source; there is no promise of complete account history.

On-device transcript metadata

The following entries describe the local aibill CLI and local MCP. The public CLI version is v0.9.12, checked October 3, 2026.

Claude Code

Anthropic

Published CLI reader

Reads supported local transcript metadata. API-equivalent values are estimates, never subscription charges or billed spend.

Reader validation
live_verified
Financial evidence
estimated or missing
Product surfaces
CLI, statusline cache, local MCP, Context Health, Glance

Read the generated format boundary ↗

Codex

OpenAI

Published CLI reader

Reads root-session-aware rollout metadata with fork accounting and snapshot deduplication. Unsupported rows remain missing.

Reader validation
live_verified
Financial evidence
estimated or missing
Product surfaces
CLI, statusline cache, local MCP, Context Health, Glance

Read the generated format boundary ↗

Gemini CLI

Google

Experimental CLI reader

Experimental, fixture-verified reader for supported chats JSON/JSONL token records. logs.json is detection-only and creates no financial row.

Reader validation
fixture_verified
Financial evidence
estimated or missing
Product surfaces
Financial CLI, report, and local MCP only; excluded from statusline, Glance, Context Health, Apply, plan/runway, and invocation evidence

Read the generated format boundary ↗

Official APIs, explicit local sync

These are optional local CLI connectors. The Workspace setup above is separate. CLI connection registration and an explicit provider sync are also separate steps.

OpenAI Costs and Usage APIs

Published CLI connector

Validation
live_verified
Evidence
verified, estimated, or missing by endpoint
Requires
Organization-owner Admin credential reference

Anthropic Cost Report and Claude Code Analytics

Published CLI connector

Validation
live_verified
Evidence
verified, estimated, or missing by row
Requires
Admin credential reference

Cursor Admin API

Beta CLI connector

Validation
fixture_verified
Evidence
estimated, detected_unverified, or missing
Requires
Business plan and team-admin credential reference

GitHub Copilot organization APIs

Beta CLI connector

Validation
fixture_verified
Evidence
estimated, detected_unverified, or missing
Requires
Organization or enterprise billing-admin credential reference

Connector validation never makes every returned row verified. The endpoint, record, coverage window, and final invoice boundary still determine the label. Open provider setup → · Review the versioned provider contracts ↗

Missing is a product answer

Cursor local session storage, Cline, Aider, and other long-tail local formats do not currently produce financial rows. The investigated Cursor local store did not provide sufficiently stable evidence for routed model, billing mode, token semantics, adjustments, or reconciled spend, so a speculative local financial parser is not currently planned; official admin APIs remain the financial path unless a stable, versioned local format emerges.

New local formats enter through the public parser registry with a descriptor, synthetic recorded fixtures, conservative evidence defaults, generated source documentation, and privacy checks. Unknown models or token shapes stay missing, never estimated as zero.