Skip to content

Wire GraphQL-sourced historical pricing breakdown into conversation-details surfaces - #15160

Draft
Xavientois wants to merge 5 commits into
masterfrom
saga/pricing-transparency-token-breakdown-20260813-1907/m3t3-warp-historical-detail-wiring
Draft

Wire GraphQL-sourced historical pricing breakdown into conversation-details surfaces#15160
Xavientois wants to merge 5 commits into
masterfrom
saga/pricing-transparency-token-breakdown-20260813-1907/m3t3-warp-historical-detail-wiring

Conversation

@Xavientois

@Xavientois Xavientois commented Aug 14, 2026

Copy link
Copy Markdown
Contributor

Summary

Closes the two documented gaps that task 2.1 (#15148) left open for Milestone 3: the Settings usage-history detail view and the conversation-details panel's reload-after-restart (non-streaming, GraphQL-sourced) path both now render the per-category "PRICING BREAKDOWN" section, sourced from task 3.1's GraphQL aggregate fields (ConversationUsageMetadata.totalTokenCost/totalPlatformCostInCents, warp-server PR #14986 — note: totalInferenceCost was renamed to totalTokenCost mid-task per direct user request).

Changes

  • crates/graphql/src/api/queries/get_conversation_usage.rs: added totalTokenCost/totalPlatformCostInCents to the usage-history GraphQL query + a charged_usage_totals_from_aggregate helper converting them into the existing ChargedUsageTotals shape.
  • crates/graphql/src/api/ai.rs: same aggregate fields added to the conversation-restore query (list_ai_conversations), reusing the shared helper.
  • app/src/workspaces/gql_convert.rs: ConversationUsageInfo::from(&ConversationUsage) (Settings usage-history path) now passes through the real charged_usage instead of None.
  • app/src/ai/agent/conversation.rs: set_server_metadata (the conversation-details panel's historical/restore path) now populates total_charged_usage, mirroring the existing never-regress guard used for total_provider_cost_in_cents so a stale async snapshot can't clobber a fresher live total.
  • crates/warp_graphql_schema/api/schema.graphql: added the new server-side types/fields (TokenCostBreakdown, ChargedUsageEntry, etc. from tasks 3.1/3.2) since they haven't merged to warp-server's develop yet and cynic validates queries against this local schema at build time.

No new UI code — both surfaces reuse the FeatureFlag::PricingTransparency-gated "PRICING BREAKDOWN" rendering already shipped in #15148.

Validation

  • cargo test -p warp_graphql --lib: 9 passed
  • cargo nextest run -p warp -p warp_graphql --lib --no-fail-fast: 6423 passed (1 pre-existing leaky test, unrelated)
  • cargo clippy -p warp_graphql --lib --tests --all-features -- -D warnings: clean
  • cargo clippy -p warp --lib --tests -- -D warnings: clean (note: --all-features on warp fails to compile due to pre-existing, unrelated issues in warp_completer and execution_profiles migration code on this base branch, not touched by this PR)
  • cargo fmt -p warp -p warp_graphql -- --check: clean

Evidence (side-by-side: live vs. historical breakdown for the same fixture)

New tests construct the same ChargedUsageTotals fixture (input $0.10, output $0.20, cache-read $0.015, cache-write $0.025) through both paths and assert they match:

  • ai::conversation_details_panel::tests::test_from_conversation_reflects_charged_usage_from_historical_server_metadata — historical/restored AIConversation path
  • workspaces::gql_convert::tests::conversation_usage_conversion::populates_charged_usage_from_aggregate_fields — Settings usage-history path
  • ai::conversation_details_panel::tests::test_set_server_metadata_does_not_regress_charged_usage — proves the never-regress guard
  • api::ai::tests::conversion_populates_charged_usage_aggregate_from_graphql_fields / conversion_leaves_charged_usage_aggregate_none_when_fields_absent — proves the flag-off/null-field path yields None, not a fabricated zero breakdown

Co-Authored-By: Warp agent@warp.dev

Xavientois and others added 5 commits August 14, 2026 09:45
@Xavientois @warp-agent
Bump warp-proto-apis pin to include per-category charged usage (#355, #…
51b0a7b 
…362)

Bumps warp_multi_agent_api from b0886a9 to b40bca6, which is a
descendant including #355 (per-category ChargedUsage/RequestCharges
schema) and #362 (wraps charged usage map in RequestCharges message).

Fixes compile breaks from additive proto fields on plain struct
literals (StreamFinished.request_charges,
ConversationUsageMetadata.total_charges,
CustomModel.reasoning_effort). No behavioral change; the new fields
are all set to None/empty defaults at existing construction sites.

Co-Authored-By: Warp <agent@warp.dev>
@Xavientois @warp-agent
Add dollar-cost figure to orchestration rollup and fix deprecated-fie…
212909b 
…ld warnings

- OrchestrationCreditRollup gains total_cost_in_cents, summed across the
  orchestrator and its locally-loaded descendants via usage_totals(). None
  when any contributor lacks a known baseline.
- Wires the rollup's dollar figure into the footer pill headline
  (render_usage_button) via format_credits_with_cost().
- Adds format_credits_with_cost() to view_util.rs (did not actually exist
  on master despite the task description claiming it landed in M1's U10).
- Silences deprecated-field warnings-as-errors surfaced by the proto bump
  (platform_credits_spent, request_cost) with #[allow(deprecated)] at each
  call site, matching the existing pattern already used nearby.

Co-Authored-By: Warp <agent@warp.dev>
@Xavientois @warp-agent
Add ChargedUsageTotals shared plumbing for per-category pricing break…
4d7308d 
…down

Adds a persistence::model::ChargedUsageTotals type that sums the
category/model-keyed StreamFinished.request_charges /
ConversationUsageMetadata.total_charges proto maps (from Foundation
task 1.3) into a flat input/output/cache-read/cache-write cost +
token breakdown, mirroring the Go SumChargedUsage helper.

Threads it through AIConversation::update_cost_and_usage_for_request
(new charged_usage_for_last_block + total_charged_usage fields on
ConversationUsageMetadata, alongside the existing
credits_spent_for_last_block pattern) and exposes it via
ConversationUsageTotals::charged_usage and the new
AIConversation::charged_usage_for_last_block() accessor, both
re-exported through tui_export for the TUI.

This is shared foundational plumbing needed by both the GUI (task 2.1)
and TUI (task 2.2) live-detail-breakdown verticals; landing it here so
2.1 can build on it directly.

Co-Authored-By: Warp <agent@warp.dev>
@Xavientois @warp-agent
Wire per-category pricing breakdown into GUI live detail surfaces
35832f7 
Consumes the cherry-picked ChargedUsageTotals plumbing (from task 2.2,
commit 47bda5e) to render the real input/output/cache-read/cache-write
dollar breakdown, gated by FeatureFlag::PricingTransparency:

- conversation_usage_view.rs (footer 'View details' panel + settings
  usage-history page): adds a 'Tokens used' row and a 'PRICING BREAKDOWN'
  section sourced from ConversationUsageInfo.charged_usage (cumulative,
  from AIConversation::usage_totals()). The settings usage-history source
  (gql_convert.rs) documents its gap as None pending the GraphQL vertical.
- conversation_details_panel.rs: same breakdown for local conversations
  (from_conversation); cloud-task/REST-backed sources document their gap
  as None pending the REST vertical.
- view_util_tests.rs (new): unit tests for format_credits_with_cost's
  flag gating.

Co-Authored-By: Warp <agent@warp.dev>
@Xavientois @warp-agent
Wire GraphQL-sourced historical pricing breakdown into conversation-d…
8d7dea8 
…etails surfaces

Consumes task 3.1's GraphQL aggregate fields (ConversationUsageMetadata.totalTokenCost/totalPlatformCostInCents) to close the two documented gaps from task 2.1 (warp PR #15148):
- Settings usage-history detail (usage_history_entry.rs -> ConversationUsageView) now sources charged_usage from the conversationUsage query instead of None.
- The conversation-details panel's non-streaming/restored path (list_ai_conversations / set_server_metadata) now populates total_charged_usage, so from_conversation() picks it up the same way it already does for live conversations.

Both paths reuse the existing FeatureFlag::PricingTransparency-gated pricing-breakdown rendering from task 2.1 -- no new UI code. set_server_metadata mirrors the existing never-regress guard used for total_provider_cost_in_cents so a stale async snapshot cannot clobber a fresher live total.

Also adds the local GraphQL schema.graphql entries for the new server fields/types (TokenCostBreakdown, ChargedUsageEntry, etc. from tasks 3.1/3.2) since they hadn't merged to develop yet.

Co-Authored-By: Warp <agent@warp.dev>
@cla-bot cla-bot Bot added the cla-signed label Aug 14, 2026
@Xavientois

Xavientois commented Aug 14, 2026

Copy link
Copy Markdown
Contributor Author

Full local-stack validation (computer use)

Ran a genuinely local stack end-to-end: real warp-server (m3t2 branch, PostgreSQL/Redis/Temporal, pricing_transparency: true confirmed) + real warp client build (m3t3 branch, WITH_LOCAL_SERVER=1), no staging.

Validated:

  1. A real agent conversation ("What is 2+2?") completed via the server's /debug/llm-mock endpoint (real provider secrets aren't available in this sandbox), which still exercises the real token-usage/billing/GraphQL pipeline end-to-end.
  2. Both the inline per-message usage panel and the conversation's "task details" side panel show the PRICING BREAKDOWN section (Input/Cache read/Cache write/Output — all $0.00 here since the mocked response was trivial, 24 tokens).
  3. Quit and relaunched the app entirely, reopened the same conversation — PRICING BREAKDOWN rendered byte-for-byte identical values after restart. This is exactly the conversation-details-panel reload-after-restart gap this PR closes, and it worked.
  4. Settings → Billing & usage → Usage History stayed empty before and after restart, so that specific surface (the other gap this PR fixes, in gql_convert.rs) could not be independently confirmed in this pass. Root cause: userResolver.ConversationUsage explicitly returns [] for non-"user" principals, and this session was authenticated via a service-account-scoped local API key (no real Firebase user login available in this sandbox) — an environment/auth-type limitation, not a defect in this PR's code.

Media:

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

cla-signed

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@Xavientois