@workweave/router 0.2.6 → 0.2.7

package/README.md

@@ -1,12 +1,13 @@
 # @workweave/router
 
-One command, anywhere, to point Claude Code, Codex, or opencode at the Weave Router.
+One command, anywhere, to point Claude Code, Codex, opencode, or pi at the Weave Router.
 
 ```bash
-npx @workweave/router                       # interactive: pick Claude Code / Codex / opencode, then scope
+npx @workweave/router                       # interactive: pick Claude Code / Codex / opencode / pi, then scope
 npx @workweave/router --claude              # skip the picker, target Claude Code
 npx @workweave/router --codex               # skip the picker, target the OpenAI Codex CLI
 npx @workweave/router --opencode            # skip the picker, target opencode
+npx @workweave/router --pi                  # skip the picker, target pi + Loom UI
 npx @workweave/router --scope project       # per-repo install, commit settings.json (or .codex/ / opencode.json)
 npx @workweave/router --local               # self-hosted via docker-compose (localhost:8080)
 npx @workweave/router --base-url https://router.acme.internal
@@ -40,6 +41,7 @@ Uninstall:
 npx @workweave/router --uninstall                       # Claude Code, user scope
 npx @workweave/router --uninstall --codex               # Codex, user scope
 npx @workweave/router --uninstall --opencode            # opencode, user scope
+npx @workweave/router --uninstall --pi                  # pi, user scope
 npx @workweave/router --uninstall --scope project       # Claude Code, inside the repo
 npx @workweave/router --uninstall --codex --scope project
 ```
@@ -52,7 +54,7 @@ Node ≥ 18 — no `curl | sh`, no Git clone, no PATH fiddling. Everything the
 shell installer documents (targets, scopes, flags, environment variables)
 works identically here.
 
-Three install targets:
+Four install targets:
 
 - **Claude Code** (default) — patches `~/.claude/settings.json` (or
   `<repo>/.claude/settings.json` with `--scope project`) so `claude` routes
@@ -71,6 +73,12 @@ Three install targets:
   natively, so opencode talks to it unmodified. Re-install rewrites only
   the managed `provider.weave` block; `--uninstall --opencode` strips it
   and leaves your other providers and settings alone.
+- **pi** (`--pi`) — registers the `weave` provider and installs this package as
+  a pi extension. Stock pi then gets the Loom startup header, Wooly's animated
+  mascot, the persistent actual-route display, cumulative session savings,
+  `/fm` + `/ufm` model-pin commands with a `[forced]` status, and the
+  context-isolated `dispatch` tool. There is no forked pi binary and no separate
+  Loom runtime.
 
 See the [main installer docs](https://github.com/workweave/router/tree/main/install)
 for the full reference.
@@ -79,8 +87,8 @@ for the full reference.
 
 - Node ≥ 18 (ships with `npx`)
 - `bash` on PATH (macOS / Linux native; Windows needs Git Bash or WSL)
-- `jq` on PATH — used by the Claude Code status line script and the
-  opencode JSON merge. Not required for the Codex path.
+- `jq` on PATH — used by the Claude Code status line and the opencode/pi JSON
+  merges. Not required for the Codex path.
 
 ## Why npx
 

package/cc-statusline.sh

@@ -33,10 +33,21 @@ set -euo pipefail
 # lives in their per-user cache dir, and on no-content-change days we skip
 # the mv entirely so the repo working tree stays clean. When upstream does
 # change, the first teammate's commit propagates the new version to the rest.
+# A pricing-table miss also triggers a refresh off-schedule — see
+# weave_refresh_on_price_miss below.
 #
 # Opt out entirely with `export WEAVE_STATUSLINE_UPDATE=0`. Override the
 # source with `WEAVE_STATUSLINE_URL=...`, e.g. for self-hosters who fork.
+#
+# $1 is an optional stamp suffix, so a caller refreshing for a reason other than
+# "the interval elapsed" rate-limits on its own clock rather than sharing the
+# periodic check's budget. $2=1 drops the stamp again if the download fails, for
+# callers recovering from a known-bad state where waiting out the full interval
+# on a transient network error is worse than retrying next turn.
 weave_self_refresh() {
+  local stamp_suffix="${1:-}"
+  local retry_on_fail="${2:-0}"
+
   [ "${WEAVE_STATUSLINE_UPDATE:-1}" = "0" ] && return 0
   command -v curl >/dev/null 2>&1 || return 0
 
@@ -53,7 +64,7 @@ weave_self_refresh() {
   mkdir -p "$cache_dir" 2>/dev/null || return 0
   local script_slug
   script_slug="$(printf '%s' "$self" | tr -c 'A-Za-z0-9._-' '_')"
-  local stamp="$cache_dir/checked-at${script_slug}"
+  local stamp="$cache_dir/checked-at${script_slug}${stamp_suffix}"
 
   local now stamp_mtime
   now="$(date +%s 2>/dev/null)" || return 0
@@ -77,7 +88,13 @@ weave_self_refresh() {
   : > "$stamp" 2>/dev/null || return 0
 
   local url="${WEAVE_STATUSLINE_URL:-https://raw.githubusercontent.com/workweave/router/main/install/cc-statusline.sh}"
-  local tmp="${self}.tmp.$$"
+  # $$ alone is not unique: two calls can run in one invocation (the periodic
+  # check and a pricing-miss retry both fire on a cold cache) and would then
+  # curl -o into the same path and mv over each other, installing a truncated
+  # script that can never self-heal. The stamp suffix is the right key — it is
+  # what makes two callers mutually exclusive in the first place, so callers
+  # that could overlap necessarily have different suffixes.
+  local tmp="${self}.tmp.$$${stamp_suffix}"
   (
     # Detach stdin (CC pipes JSON to us) so curl can't accidentally consume
     # it, and silence all output so nothing leaks into the statusline.
@@ -96,6 +113,11 @@ weave_self_refresh() {
       fi
     else
       rm -f "$tmp"
+      # A download that never landed shouldn't spend the caller's whole
+      # interval; drop the stamp so the next turn can try again.
+      if [ "$retry_on_fail" = "1" ]; then
+        rm -f "$stamp"
+      fi
     fi
   ) >/dev/null 2>&1 &
   disown 2>/dev/null || true
@@ -399,6 +421,42 @@ if [[ -n "$transcript_path" && -f "$transcript_path" ]]; then
   ) || true
 fi
 
+# ---------- refresh when a model isn't in the pricing table ----------
+#
+# A model that shipped after this copy of the script has no price entry, so the
+# jq guard above zeroes savings on every turn and the line reads "saved $0.00"
+# — indistinguishable from "the router ran and didn't beat your selection". The
+# periodic check heals that eventually, but a week late, and a model launch is
+# exactly when the number gets looked at. Refresh off-schedule instead.
+#
+# Keyed on its own stamp per unpriced id: a model we never price (self-hosted,
+# unrecognized) then costs one download per interval, not one per turn, and
+# can't starve the periodic check.
+weave_refresh_on_price_miss() {
+  local candidates="" m
+  for m in "$@"; do
+    case "$m" in
+      "" | "?" | failure | weave-router | "<synthetic>") continue ;;
+    esac
+    candidates="${candidates}${m}"$'\n'
+  done
+  [ -n "$candidates" ] || return 0
+
+  # A malformed $prices emits nothing here, which fails closed (no refresh)
+  # rather than re-downloading every turn.
+  local missing
+  missing="$(printf '%s' "$candidates" \
+    | jq -rR --argjson p "$prices" \
+        'select($p.input[.] == null or $p.output[.] == null)' 2>/dev/null \
+    | head -n 1)" || return 0
+  [ -n "$missing" ] || return 0
+
+  local model_slug
+  model_slug="$(printf '%s' "$missing" | tr -c 'A-Za-z0-9._-' '_')"
+  weave_self_refresh ".miss.${model_slug}" 1
+}
+weave_refresh_on_price_miss "$requested_norm" "$routed" 2>/dev/null || true
+
 # Brand color (#FF6C47) on terminals that grok 24-bit truecolor — that's
 # every modern one (iTerm2, Apple Terminal, vscode, ghostty, alacritty,
 # wezterm, kitty). Falls back gracefully on any escape-stripping terminal.

package/install.sh

@@ -1999,10 +1999,21 @@ set -euo pipefail
 # lives in their per-user cache dir, and on no-content-change days we skip
 # the mv entirely so the repo working tree stays clean. When upstream does
 # change, the first teammate's commit propagates the new version to the rest.
+# A pricing-table miss also triggers a refresh off-schedule — see
+# weave_refresh_on_price_miss below.
 #
 # Opt out entirely with `export WEAVE_STATUSLINE_UPDATE=0`. Override the
 # source with `WEAVE_STATUSLINE_URL=...`, e.g. for self-hosters who fork.
+#
+# $1 is an optional stamp suffix, so a caller refreshing for a reason other than
+# "the interval elapsed" rate-limits on its own clock rather than sharing the
+# periodic check's budget. $2=1 drops the stamp again if the download fails, for
+# callers recovering from a known-bad state where waiting out the full interval
+# on a transient network error is worse than retrying next turn.
 weave_self_refresh() {
+  local stamp_suffix="${1:-}"
+  local retry_on_fail="${2:-0}"
+
   [ "${WEAVE_STATUSLINE_UPDATE:-1}" = "0" ] && return 0
   command -v curl >/dev/null 2>&1 || return 0
 
@@ -2019,7 +2030,7 @@ weave_self_refresh() {
   mkdir -p "$cache_dir" 2>/dev/null || return 0
   local script_slug
   script_slug="$(printf '%s' "$self" | tr -c 'A-Za-z0-9._-' '_')"
-  local stamp="$cache_dir/checked-at${script_slug}"
+  local stamp="$cache_dir/checked-at${script_slug}${stamp_suffix}"
 
   local now stamp_mtime
   now="$(date +%s 2>/dev/null)" || return 0
@@ -2043,7 +2054,13 @@ weave_self_refresh() {
   : > "$stamp" 2>/dev/null || return 0
 
   local url="${WEAVE_STATUSLINE_URL:-https://raw.githubusercontent.com/workweave/router/main/install/cc-statusline.sh}"
-  local tmp="${self}.tmp.$$"
+  # $$ alone is not unique: two calls can run in one invocation (the periodic
+  # check and a pricing-miss retry both fire on a cold cache) and would then
+  # curl -o into the same path and mv over each other, installing a truncated
+  # script that can never self-heal. The stamp suffix is the right key — it is
+  # what makes two callers mutually exclusive in the first place, so callers
+  # that could overlap necessarily have different suffixes.
+  local tmp="${self}.tmp.$$${stamp_suffix}"
   (
     # Detach stdin (CC pipes JSON to us) so curl can't accidentally consume
     # it, and silence all output so nothing leaks into the statusline.
@@ -2062,6 +2079,11 @@ weave_self_refresh() {
       fi
     else
       rm -f "$tmp"
+      # A download that never landed shouldn't spend the caller's whole
+      # interval; drop the stamp so the next turn can try again.
+      if [ "$retry_on_fail" = "1" ]; then
+        rm -f "$stamp"
+      fi
     fi
   ) >/dev/null 2>&1 &
   disown 2>/dev/null || true
@@ -2365,6 +2387,42 @@ if [[ -n "$transcript_path" && -f "$transcript_path" ]]; then
   ) || true
 fi
 
+# ---------- refresh when a model isn't in the pricing table ----------
+#
+# A model that shipped after this copy of the script has no price entry, so the
+# jq guard above zeroes savings on every turn and the line reads "saved $0.00"
+# — indistinguishable from "the router ran and didn't beat your selection". The
+# periodic check heals that eventually, but a week late, and a model launch is
+# exactly when the number gets looked at. Refresh off-schedule instead.
+#
+# Keyed on its own stamp per unpriced id: a model we never price (self-hosted,
+# unrecognized) then costs one download per interval, not one per turn, and
+# can't starve the periodic check.
+weave_refresh_on_price_miss() {
+  local candidates="" m
+  for m in "$@"; do
+    case "$m" in
+      "" | "?" | failure | weave-router | "<synthetic>") continue ;;
+    esac
+    candidates="${candidates}${m}"$'\n'
+  done
+  [ -n "$candidates" ] || return 0
+
+  # A malformed $prices emits nothing here, which fails closed (no refresh)
+  # rather than re-downloading every turn.
+  local missing
+  missing="$(printf '%s' "$candidates" \
+    | jq -rR --argjson p "$prices" \
+        'select($p.input[.] == null or $p.output[.] == null)' 2>/dev/null \
+    | head -n 1)" || return 0
+  [ -n "$missing" ] || return 0
+
+  local model_slug
+  model_slug="$(printf '%s' "$missing" | tr -c 'A-Za-z0-9._-' '_')"
+  weave_self_refresh ".miss.${model_slug}" 1
+}
+weave_refresh_on_price_miss "$requested_norm" "$routed" 2>/dev/null || true
+
 # Brand color (#FF6C47) on terminals that grok 24-bit truecolor — that's
 # every modern one (iTerm2, Apple Terminal, vscode, ghostty, alacritty,
 # wezterm, kitty). Falls back gracefully on any escape-stripping terminal.
@@ -2480,12 +2538,20 @@ custom_headers="$custom_headers"$'\n'"X-App: claude-code"
 if [ "$scope" = "project" ] && [ -z "$install_dir" ]; then
   jq -n --arg url "$base_url" --arg sl "$statusline_path_for_settings" '{
     env: { ANTHROPIC_BASE_URL: $url, ENABLE_TOOL_SEARCH: "auto" },
-    statusLine: { type: "command", command: $sl }
+    statusLine: { type: "command", command: $sl },
+    attribution: {
+      commit: "Co-Authored-By: Weave Router <noreply@workweave.ai>",
+      pr: "🤖 Generated with [Weave Router](https://router.workweave.ai)"
+    }
   }' >"$tmp_patch"
 else
   jq -n --arg url "$base_url" --arg header "$custom_headers" --arg sl "$statusline_path_for_settings" '{
     env: { ANTHROPIC_BASE_URL: $url, ANTHROPIC_CUSTOM_HEADERS: $header, ENABLE_TOOL_SEARCH: "auto" },
-    statusLine: { type: "command", command: $sl }
+    statusLine: { type: "command", command: $sl },
+    attribution: {
+      commit: "Co-Authored-By: Weave Router <noreply@workweave.ai>",
+      pr: "🤖 Generated with [Weave Router](https://router.workweave.ai)"
+    }
   }' >"$tmp_patch"
 fi
 
@@ -2501,6 +2567,7 @@ if [ -f "$settings_file" ]; then
     | (if (.env | length) == 0 then del(.env) else . end)
     | del(.apiKeyHelper)
     | (if $b.statusLine then .statusLine = $b.statusLine else . end)
+    | (if $b.attribution then .attribution = $b.attribution else . end)
   ' "$settings_file" "$tmp_patch")"
   printf '%s\n' "$merged" >"$settings_file"
 else

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@workweave/router",
-  "version": "0.2.6",
+  "version": "0.2.7",
   "description": "One-command installer that points Claude Code, Codex, opencode, or pi at the Weave Router. For pi it also ships the routing extension, loaded via pi.extensions.",
   "bin": {
     "weave-router": "bin.js",

package/pi-router/README.md

@@ -20,8 +20,16 @@ from npm on next start and loads this extension via its `pi.extensions` field.
 
 ## What it does
 
+- **Loom experience on stock pi.** Replaces pi's startup header through the
+  public extension API, adds Wooly's responsive orange terminal animation, and
+  keeps pi's own runtime/footer intact. Wooly is visual only: there is no
+  dialogue box, narration, coaching request, or separate Loom runtime.
 - **Automatic model selection.** All pi traffic flows through the router, which
   selects the model per request. You don't pick a model — the router does.
+- **Force-model commands.** `/fm <model>` and `/force-model <model>` pin the
+  current router session; `/ufm` and `/unforce-model` resume automatic routing.
+  The persistent status changes to `WEAVE ROUTER — <model> [forced]` after the
+  router validates and canonicalizes the requested model.
 - **Per-process routing bias.** Static `x-weave-routing-*` knob headers bias the
   router: quality on the main loop, speed + cheap on subagents, cheapest on
   compaction.
@@ -31,10 +39,14 @@ from npm on next start and loads this extension via its `pi.extensions` field.
   natively. `dispatch` spawns child `pi` processes (read-only by default), runs
   them concurrently, and returns only each subagent's final answer — intermediate
   tool output stays in the child, so the main context stays small.
-- **Routed-model display.** Shows the model the router actually chose
-  (`x-router-model`) in the status bar, and opts the request out of the router's
-  in-band routing badge (`X-Weave-Routing-Marker: off`) — pi can't render that
-  separate marker text block inline, and the status bar already conveys the model.
+- **Persistent route + savings display.** Shows
+  `WEAVE ROUTER — <routed> ← <selected> · saved $X.XX` below pi's native footer
+  data. Savings compare the selected and routed catalog prices against the same
+  input/output/cache usage, accumulate across the reachable session branch,
+  and survive resume. Unknown catalog prices are labeled `unpriced` instead of
+  silently contributing zero; costlier routing is labeled `extra`, not savings.
+- **No duplicate in-band badge.** Sets `X-Weave-Routing-Marker: off` because the
+  persistent status already conveys the actual model.
 - **Safety backstop.** Blocks a few catastrophic shell commands (`rm -rf /`,
   `mkfs`, `dd of=/dev/…`, fork bombs, force-push to main). Disable with
   `WEAVE_NO_SAFETY=1`.
@@ -64,6 +76,11 @@ Routing through the router switches pi from Claude **subscription OAuth** to
 **per-token** billing on the router deployment's key (or your BYOK key). BYOK
 skips cross-provider failover; deployment-key billing is the default.
 
+The displayed savings are a client-side estimate from the router's generated
+model-price catalog. Cache writes use 1.25× input price and cache reads use 0.1×
+input price, matching the Claude Code statusline. The ledger stores its catalog
+version with each response so resumed totals remain auditable.
+
 ## Notes
 
 - Cheap compaction is currently a reserved flag — the handler defers to pi's

package/pi-router/src/config.ts

@@ -243,6 +243,8 @@ function model(id: string, name: string, maxTokens: number): ProviderModelConfig
 // ---------- dispatch / misc tunables ----------
 
 export const ROUTED_MODEL_HEADER = (process.env.WEAVE_ROUTED_MODEL_HEADER || "x-router-model").toLowerCase();
+export const ROUTED_PROVIDER_HEADER = "x-router-provider";
+export const ROUTER_DECISION_HEADER = "x-router-decision";
 /** Marker a headless child prints to stderr so the parent dispatch can read its routed model. */
 export const ROUTED_MODEL_STDERR_PREFIX = "weave-routed-model:";
 

package/pi-router/src/force-model.ts

@@ -0,0 +1,126 @@
+/**
+ * Pi consumes slash commands locally, so expose the router's force-model
+ * directives as extension commands and forward one canonical user turn.
+ *
+ * The router remains authoritative for aliases, validation, canonical model
+ * ids, and pin persistence. UI state is reconstructed from command/response
+ * pairs on the reachable Pi branch; this is necessary because Pi records the
+ * selected model handle on assistant messages rather than the response model.
+ */
+
+import type {
+	ExtensionAPI,
+	ExtensionCommandContext,
+	SessionEntry,
+} from "@mariozechner/pi-coding-agent";
+
+export type ForceModelTransition =
+	| { kind: "applied"; model: string }
+	| { kind: "cleared" }
+	| { kind: "noop" };
+
+type ForceModelDirective = "force" | "clear";
+
+function messageText(message: unknown): string | undefined {
+	if (!message || typeof message !== "object" || !("content" in message)) return undefined;
+	const content = message.content;
+	if (typeof content === "string") return content;
+	if (!Array.isArray(content)) return undefined;
+	return content
+		.filter(
+			(block): block is { type: "text"; text: string } =>
+				Boolean(
+					block &&
+						typeof block === "object" &&
+						"type" in block &&
+						block.type === "text" &&
+						"text" in block &&
+						typeof block.text === "string",
+				),
+		)
+		.map((block) => block.text)
+		.join("\n");
+}
+
+export function parseForceModelDirective(text: string): ForceModelDirective | undefined {
+	const command = text.trim();
+	if (/^\/(?:force-model|fm)\s+\S+/i.test(command)) return "force";
+	if (/^\/(?:unforce-model|ufm)$/i.test(command)) return "clear";
+	return undefined;
+}
+
+export function parseForceModelAcknowledgement(text: string): ForceModelTransition | undefined {
+	const applied = /force-model applied:\s+([^\s()]+)/i.exec(text);
+	if (applied) return { kind: "applied", model: applied[1] };
+	if (/force-model cleared/i.test(text)) return { kind: "cleared" };
+	if (/is(?:n't| not) a recognized model/i.test(text)) return { kind: "noop" };
+	return undefined;
+}
+
+function isSyntheticPinClear(text: string): boolean {
+	return /^(?:✦ \*\*Weave Router\*\* →|Weave Router:)\s+(?:Tool-call|Repetition|No-progress) loop detected\b[\s\S]*\bclearing the session pin\b/i.test(
+		text.trim(),
+	);
+}
+
+/** Reconstruct the effective router pin on the currently reachable branch. */
+export function forcedModelFromBranch(entries: readonly SessionEntry[]): string | undefined {
+	let forcedModel: string | undefined;
+	let pendingDirective: ForceModelDirective | undefined;
+
+	for (const entry of entries) {
+		if (entry.type !== "message") continue;
+		if (entry.message.role === "user") {
+			pendingDirective = parseForceModelDirective(messageText(entry.message) ?? "");
+			continue;
+		}
+		if (entry.message.role !== "assistant") continue;
+
+		const text = messageText(entry.message) ?? "";
+		if (isSyntheticPinClear(text)) {
+			forcedModel = undefined;
+			pendingDirective = undefined;
+			continue;
+		}
+		if (!pendingDirective) continue;
+
+		const transition = parseForceModelAcknowledgement(text);
+		if (transition?.kind === "applied" && pendingDirective === "force") forcedModel = transition.model;
+		else if (transition?.kind === "cleared" && pendingDirective === "clear") forcedModel = undefined;
+		// Rejected force-model commands intentionally retain the previous pin.
+		pendingDirective = undefined;
+	}
+
+	return forcedModel;
+}
+
+function sendRouterCommand(pi: ExtensionAPI, command: string, ctx: ExtensionCommandContext): void {
+	pi.sendUserMessage(command, ctx.isIdle() ? undefined : { deliverAs: "followUp" });
+}
+
+export function registerForceModelCommands(pi: ExtensionAPI): void {
+	const forceModel = async (args: string, ctx: ExtensionCommandContext): Promise<void> => {
+		const modelAndPrompt = args.trim();
+		if (!modelAndPrompt) {
+			ctx.ui.notify("Usage: /fm <model-id>", "warning");
+			return;
+		}
+		sendRouterCommand(pi, `/force-model ${modelAndPrompt}`, ctx);
+	};
+	const clearForceModel = async (_args: string, ctx: ExtensionCommandContext): Promise<void> => {
+		sendRouterCommand(pi, "/unforce-model", ctx);
+	};
+
+	for (const name of ["fm", "force-model"]) {
+		pi.registerCommand(name, {
+			description: "Pin this session to a specific model via the Weave Router",
+			handler: forceModel,
+		});
+	}
+	for (const name of ["ufm", "unforce-model"]) {
+		pi.registerCommand(name, {
+			description: "Clear this session's forced Weave Router model",
+			handler: clearForceModel,
+		});
+	}
+}

package/pi-router/src/index.ts

@@ -7,7 +7,7 @@
  *                   the main loop, speed/cheap in subagents).
  *   - metadata:     stamp body.metadata.user_id for sticky sessions + subagent
  *                   detection.
- *   - routed-model: show which model the router actually picked.
+ *   - Loom UI:      branded header, Wooly animation, actual route, and saved $.
  *   - safety:       block catastrophic bash (unless WEAVE_NO_SAFETY=1).
  *   - compaction:   experimental cheap path (only when WEAVE_CHEAP_COMPACTION=1).
  *   - dispatch:     parallel, context-isolated subagents — top-level process
@@ -22,6 +22,7 @@ import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
 import { isSubagent } from "./config.js";
 import { registerCheapCompaction } from "./compaction.js";
 import { registerDispatch } from "./dispatch.js";
+import { registerForceModelCommands } from "./force-model.js";
 import { registerMetadata } from "./metadata.js";
 import { registerRoutedModel } from "./routed-model.js";
 import { registerSafety } from "./safety.js";
@@ -37,6 +38,7 @@ export default function (pi: ExtensionAPI): void {
 	pi.on("session_start", () => registerWeave(pi));
 
 	registerMetadata(pi);
+	registerForceModelCommands(pi);
 	registerRoutedModel(pi);
 
 	if (process.env.WEAVE_NO_SAFETY !== "1") registerSafety(pi);

package/pi-router/src/pricing.generated.ts

@@ -0,0 +1,79 @@
+// Code generated by cmd/genprices; DO NOT EDIT.
+// Source: internal/router/catalog (USD per 1M tokens).
+
+export interface ModelPricing {
+	inputUsdPerMillion: number;
+	outputUsdPerMillion: number;
+}
+
+export const PRICING_VERSION = "catalog-sha256:a13e743aa9694647";
+
+export const MODEL_PRICING: Readonly<Record<string, ModelPricing>> = Object.freeze({
+	"claude-fable-5": { inputUsdPerMillion: 10, outputUsdPerMillion: 50 },
+	"claude-haiku-4-5": { inputUsdPerMillion: 1, outputUsdPerMillion: 5 },
+	"claude-opus-4-0": { inputUsdPerMillion: 15, outputUsdPerMillion: 75 },
+	"claude-opus-4-1": { inputUsdPerMillion: 15, outputUsdPerMillion: 75 },
+	"claude-opus-4-5": { inputUsdPerMillion: 5, outputUsdPerMillion: 25 },
+	"claude-opus-4-6": { inputUsdPerMillion: 5, outputUsdPerMillion: 25 },
+	"claude-opus-4-7": { inputUsdPerMillion: 5, outputUsdPerMillion: 25 },
+	"claude-opus-4-8": { inputUsdPerMillion: 5, outputUsdPerMillion: 25 },
+	"claude-opus-5": { inputUsdPerMillion: 5, outputUsdPerMillion: 25 },
+	"claude-sonnet-4-5": { inputUsdPerMillion: 3, outputUsdPerMillion: 15 },
+	"claude-sonnet-4-6": { inputUsdPerMillion: 3, outputUsdPerMillion: 15 },
+	"claude-sonnet-5": { inputUsdPerMillion: 3, outputUsdPerMillion: 15 },
+	"deepseek/deepseek-v4-flash": { inputUsdPerMillion: 0.1134, outputUsdPerMillion: 0.2791 },
+	"deepseek/deepseek-v4-pro": { inputUsdPerMillion: 1.318, outputUsdPerMillion: 2.6361 },
+	"gemini-2.0-flash": { inputUsdPerMillion: 0.1, outputUsdPerMillion: 0.4 },
+	"gemini-2.0-flash-lite": { inputUsdPerMillion: 0.075, outputUsdPerMillion: 0.3 },
+	"gemini-2.5-flash": { inputUsdPerMillion: 0.3, outputUsdPerMillion: 1.2 },
+	"gemini-2.5-flash-lite": { inputUsdPerMillion: 0.1, outputUsdPerMillion: 0.4 },
+	"gemini-2.5-pro": { inputUsdPerMillion: 1.25, outputUsdPerMillion: 5 },
+	"gemini-3-flash-preview": { inputUsdPerMillion: 0.5, outputUsdPerMillion: 2 },
+	"gemini-3-pro-preview": { inputUsdPerMillion: 2, outputUsdPerMillion: 8 },
+	"gemini-3.1-flash-lite-preview": { inputUsdPerMillion: 0.1, outputUsdPerMillion: 0.4 },
+	"gemini-3.1-pro-preview": { inputUsdPerMillion: 2, outputUsdPerMillion: 8 },
+	"gemini-3.5-flash": { inputUsdPerMillion: 1.5, outputUsdPerMillion: 9 },
+	"gemini-3.5-flash-lite": { inputUsdPerMillion: 0.3, outputUsdPerMillion: 2.5 },
+	"gemini-3.6-flash": { inputUsdPerMillion: 1.5, outputUsdPerMillion: 7.5 },
+	"google/gemma-4-26b-a4b-it": { inputUsdPerMillion: 0.15, outputUsdPerMillion: 0.6 },
+	"gpt-4.1": { inputUsdPerMillion: 2, outputUsdPerMillion: 8 },
+	"gpt-4.1-mini": { inputUsdPerMillion: 0.4, outputUsdPerMillion: 1.6 },
+	"gpt-4.1-nano": { inputUsdPerMillion: 0.1, outputUsdPerMillion: 0.4 },
+	"gpt-4o": { inputUsdPerMillion: 2.5, outputUsdPerMillion: 10 },
+	"gpt-4o-mini": { inputUsdPerMillion: 0.15, outputUsdPerMillion: 0.6 },
+	"gpt-5": { inputUsdPerMillion: 2.5, outputUsdPerMillion: 10 },
+	"gpt-5-chat": { inputUsdPerMillion: 2.5, outputUsdPerMillion: 10 },
+	"gpt-5-mini": { inputUsdPerMillion: 0.5, outputUsdPerMillion: 2 },
+	"gpt-5-nano": { inputUsdPerMillion: 0.1, outputUsdPerMillion: 0.4 },
+	"gpt-5.4": { inputUsdPerMillion: 2.5, outputUsdPerMillion: 15 },
+	"gpt-5.4-mini": { inputUsdPerMillion: 0.75, outputUsdPerMillion: 4.5 },
+	"gpt-5.4-nano": { inputUsdPerMillion: 0.2, outputUsdPerMillion: 1.25 },
+	"gpt-5.4-pro": { inputUsdPerMillion: 30, outputUsdPerMillion: 180 },
+	"gpt-5.5": { inputUsdPerMillion: 5, outputUsdPerMillion: 30 },
+	"gpt-5.5-mini": { inputUsdPerMillion: 0.5, outputUsdPerMillion: 2.5 },
+	"gpt-5.5-nano": { inputUsdPerMillion: 0.15, outputUsdPerMillion: 0.6 },
+	"gpt-5.5-pro": { inputUsdPerMillion: 30, outputUsdPerMillion: 180 },
+	"gpt-5.6-luna": { inputUsdPerMillion: 1, outputUsdPerMillion: 6 },
+	"gpt-5.6-sol": { inputUsdPerMillion: 5, outputUsdPerMillion: 30 },
+	"gpt-5.6-terra": { inputUsdPerMillion: 2.5, outputUsdPerMillion: 15 },
+	"grok-4.5": { inputUsdPerMillion: 2, outputUsdPerMillion: 6 },
+	"minimax/minimax-m2.7": { inputUsdPerMillion: 0.3, outputUsdPerMillion: 1.2 },
+	"minimax/minimax-m3": { inputUsdPerMillion: 0.3, outputUsdPerMillion: 1.2 },
+	"mistralai/mistral-small-2603": { inputUsdPerMillion: 0.2, outputUsdPerMillion: 0.6 },
+	"moonshotai/kimi-k2.5": { inputUsdPerMillion: 0.6, outputUsdPerMillion: 3 },
+	"moonshotai/kimi-k2.6": { inputUsdPerMillion: 0.95, outputUsdPerMillion: 4 },
+	"moonshotai/kimi-k2.7": { inputUsdPerMillion: 0.95, outputUsdPerMillion: 4 },
+	"moonshotai/kimi-k3": { inputUsdPerMillion: 3, outputUsdPerMillion: 15 },
+	"qwen/qwen3-235b-a22b-2507": { inputUsdPerMillion: 0.2266, outputUsdPerMillion: 0.9064 },
+	"qwen/qwen3-30b-a3b-instruct-2507": { inputUsdPerMillion: 0.15, outputUsdPerMillion: 0.6 },
+	"qwen/qwen3-coder": { inputUsdPerMillion: 0.9, outputUsdPerMillion: 2.7 },
+	"qwen/qwen3-coder-next": { inputUsdPerMillion: 0.5, outputUsdPerMillion: 1.2 },
+	"qwen/qwen3-next-80b-a3b-instruct": { inputUsdPerMillion: 0.15, outputUsdPerMillion: 1.2 },
+	"qwen/qwen3.5-flash-02-23": { inputUsdPerMillion: 0.05, outputUsdPerMillion: 0.15 },
+	"qwen/qwen3.6-35b-a3b": { inputUsdPerMillion: 0.172, outputUsdPerMillion: 1.2002 },
+	"qwen/qwen3.7-plus": { inputUsdPerMillion: 0.4, outputUsdPerMillion: 1.6 },
+	"xiaomi/mimo-v2.5-pro": { inputUsdPerMillion: 1, outputUsdPerMillion: 3 },
+	"z-ai/glm-5": { inputUsdPerMillion: 1, outputUsdPerMillion: 3.2 },
+	"z-ai/glm-5.1": { inputUsdPerMillion: 1.4, outputUsdPerMillion: 4.4 },
+	"z-ai/glm-5.2": { inputUsdPerMillion: 1.4, outputUsdPerMillion: 4.4 },
+});