[elyasmnvidian]

Before this PR, when a provider omitted a dollar charge and no price file was set, the server recorded the cost as unknown. Now the server and desktop app estimate costs from a price table bundled in the switchyard-costs crate. For example, calls to claude-opus-5-5 now get estimated costs and desktop baselines without a custom price file.

The table lists 35 exact model IDs: 15 direct OpenAI IDs, eight prefixed OpenAI IDs, codex-auto-review, seven direct Claude IDs, three OpenRouter Claude IDs, and azure/anthropic/claude-haiku-4-5. The bundled rates match OpenAI's prices, Anthropic's prices, and OpenRouter's model list as of October 21, 2026. codex-auto-review uses the gpt-6-sol rates. IDs with a provider prefix use list rates, so billed charges may differ. IDs must match exactly; unlisted IDs stay unknown.

For GPT-5.6 and newer, cache writes cost 1.25 times the input rate (OpenAI's 30-minute rate). Older OpenAI models charge the input rate for separately counted writes. Claude rates separate 5-minute and 1-hour writes. The table leaves out long-context and service-tier pricing, OpenAI's 12-hour cache writes, tool fees, and cache storage. Bundled rates change only with a release. ChatGPT subscription comparisons are estimates.

To use your own rates, pass --pricing-file PATH to the server or set pricing_file in desktop settings. The TOML or JSON file replaces the whole table. A missing, unreadable, or invalid configured file is an error. Provider-reported charges, including zero, take precedence over estimates. Missing usage or rates leave the estimate unknown, and changing rates does not reprice saved calls. The server loads rates at startup, and the desktop app loads them when it refreshes settings or usage; neither fetches prices over the network. Install creates no price file, and uninstall keeps custom price files. Users must convert or remove older inline [prices] settings by hand.

Testing

A local HTTP stub returned 10 uncached input tokens and 2 output tokens for azure/anthropic/claude-haiku-4-5. With routing logging on and no price file, the server recorded {"usd":0.00002,"source":"estimated"}. TOML and JSON price files that set input to $9 and output to $0 per million tokens both produced $0.00009. The stub checks Switchyard's cost math only; it does not check a provider's bill.

TOML price file used

["azure/anthropic/claude-haiku-4-5"]
input=9
output=0
cache_read=0
cache_write=0

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Repository: NVIDIA-NeMo/Switchyard/.coderabbit.yaml
• Review profile: CHILL
• Plan: Enterprise
• Run ID: c1a62075-d566-4937-9166-068a0661b5ba

📥 Commits

Reviewing files that changed from the base of the PR and between 2e9f94b and ed9cb2d.

📒 Files selected for processing: 6

• crates/switchyard-desktop/prices.toml
• crates/switchyard-desktop/src/bin/switchyard-desktop-install.rs
• crates/switchyard-desktop/src/install.rs
• crates/switchyard-desktop/src/install/windows.rs
• crates/switchyard-desktop/tests/desktop.rs
• scripts/config/prices.json

💤 Files with no reviewable changes: 1

• crates/switchyard-desktop/prices.toml

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 9 remain after this review.

Walkthrough

The desktop installer now bundles pricing in JSON, migrates legacy TOML pricing, and validates pricing files. Generated configuration references the pricing file, and server launch paths use shared settings arguments. Tests cover pricing migration, invalid pricing, backups, and recovery.

Changes

Desktop Pricing Configuration

Layer / File(s)	Summary
Pricing resources and server launch arguments crates/switchyard-desktop/prices.toml, scripts/config/prices.json, crates/switchyard-desktop/src/install.rs, crates/switchyard-desktop/src/bin/switchyard-desktop-install.rs, crates/switchyard-desktop/src/install/windows.rs	The desktop installer uses bundled JSON pricing instead of the deleted TOML pricing file. Generated desktop configuration references prices_file and records a baseline model. Settings::server_args() supplies arguments to both launch paths and includes --pricing-file when the installed file exists.
Legacy pricing reconciliation crates/switchyard-desktop/src/install.rs	Desktop settings reconciliation returns structured changes. When no pricing file exists, it extracts legacy rates and converts them to validated JSON. Conversion defaults cache_read to the input rate and sets cache_write to the input rate.
Installation, backups, and validation crates/switchyard-desktop/src/install.rs, crates/switchyard-desktop/tests/desktop.rs	The installation plan includes the pricing file. Installation validates existing pricing and backs up desktop settings before applying rewrites. Tests cover pricing migration, bundled pricing, backups, invalid pricing rejection, and recovery assertions.

Priority: ➖ Normal

Estimated code review effort: 3 (Moderate) | ~25 minutes

Merge Risk: ⚪ Minimal · up to ed9cb

The installer now creates a shared prices file and starts the server with it. No concrete merge-blocking risk was identified in the supplied changes.

Pre-merge checks: 4 · 1

Failed checks: 1 warning

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 40.74% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 27 functions across 4 files. (1 skipped: …	Write docstrings for the functions missing them to satisfy the coverage threshold.

Passed checks: 4 passed

Check name	Status	Explanation
Description Check		Check skipped - CodeRabbit’s high-level summary is enabled.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.
Title check		The title clearly describes the main change: using bundled model prices by default during desktop installation.

Full details: Docstring Coverage

Explanation

Docstring coverage is 40.74% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 27 functions across 4 files. (1 skipped: 1 unsupported.)

Resolution

Write docstrings for the functions missing them to satisfy the coverage threshold.

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{A rabbit checks the pricing JSON at dawn,
Then hops where legacy TOML once sat.
The server finds its pricing file,
While backups keep the old settings safe.
Bad rates are turned away at the gate,
And the rabbit rests beneath a config leaf.}

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://NVIDIA-NeMo.github.io/Switchyard/pr-preview/pr-1049/
Built to branch gh-pages at 2026-10-21 17:30 UTC. Preview will be ready when the GitHub Pages deployment is complete.

[ayushag-nv]

Should we pull in cost from OpenRouter on the go instead of hardcoding ? It will be difficult to maintain