coderouter-cli 2.5.5__tar.gz → 2.6.1__tar.gz

{coderouter_cli-2.5.5 → coderouter_cli-2.6.1}/.gitignore

@@ -36,6 +36,7 @@ ENV/
 # _OUTPUTS
 # ============================================================
 _OUTPUTS/
+_run/
 
 # uv
 # Note: uv.lock SHOULD be committed (per plan.md §5.4)

{coderouter_cli-2.5.5 → coderouter_cli-2.6.1}/CHANGELOG.md

@@ -6,6 +6,122 @@ versioning follows [SemVer](https://semver.org/).
 
 ---
 
+## [v2.6.1] — 2026-06-28 (Token-savings accounting)
+
+Patch release: surfaces **token-savings accounting** in the metrics layer
+and dashboard. The figure is owned by core, so it appears even when no
+plugin is installed (trim savings from the context-budget guard), and the
+optional `compress` plugin adds to the same total via a neutral
+`tokens-saved` log event. **No new core dependency**, **no behavioral
+change** to existing paths, and **fully backward compatible** — existing
+counters and events are untouched; the new buckets are additive.
+
+### Added
+
+- **Token-savings buckets in `MetricsCollector`**
+  (`coderouter/metrics/collector.py`). `tokens_saved_total` plus a
+  per-mechanism breakdown (`tokens_saved_by_mechanism`). Two feeds
+  aggregate under one schema: `trim` (derived from the existing
+  `context-budget-trimmed` event's before/after token estimate) and
+  `compress` (the neutral `tokens-saved` event emitted by the compress
+  plugin — no core import). Wired through `snapshot()`, `save_state()`,
+  `load_state()`, and `reset()`.
+- **Dashboard Token Savings tiles**
+  (`coderouter/ingress/dashboard_routes.py`). Three tiles — trim /
+  compress / total — in the "Cost & Language Tax" panel, zero-filled so a
+  fresh or local-only deployment renders cleanly.
+
+### Docs / examples
+
+- Reorganized provider samples under `examples/` with a category index.
+- Added the language-tax guide (JA/EN) and a Claude Code +
+  llama.cpp/vllm backend guide.
+
+### Tests
+
+- `tests/test_tokens_saved_metric.py` — trim/compress accounting,
+  negative-delta clamping, combined aggregate, persistence round-trip,
+  and reset. Full suite: **1263 passed, 1 skipped**.
+
+### Companion
+
+- `coderouter-plugin-compress` emits the `tokens-saved` event (plugin
+  branch `feat/tokens-saved-emit`, v0.2.0). CodeRouter core works
+  standalone — trim savings show without the plugin.
+
+---
+
+## [v2.6.0] — 2026-06-20 (Language Tax: measure, route, visualize)
+
+Minor release: makes the CJK **"language tax"** — cloud tokenizers bill
+Japanese/Chinese/Korean text ~1.2–1.5× more tokens per character than
+English, while local models are unaffected — measurable, routable, and
+visible. Built entirely on existing infrastructure; **no new core
+dependency** (the accurate tokenizer is the existing optional `accuracy`
+extra), **no network** (local `tokenizer.json` only), and **fully
+backward compatible** — the feature is inert until a provider declares
+`tokenizer_path`.
+
+### Added
+
+- **Language-tax measurement (`coderouter/language_tax.py`).** A leaf
+  module exposing `cjk_char_ratio`, `estimate_language_tax`,
+  `LanguageTaxBreakdown`, and `language_tax_usd`. CJK detection is
+  stdlib-only (Unicode range checks); the accurate token count is
+  delegated to the optional `accuracy` (`tokenizers`) backend with a
+  char/4 fallback. The tax multiplier is `tokens_accurate /
+  tokens_heuristic` — ~1.0 for English/code, ~2.0–4.0 for pure CJK.
+
+- **End-to-end cost integration.** `CostBreakdown` gains
+  `language_tax_multiplier` / `language_tax_usd`;
+  `compute_cost_for_attempt` accepts an optional `language_tax=`. Both
+  `cache-observed` emit sites in `routing/fallback.py` (streaming +
+  non-streaming) build a `LanguageTaxBreakdown` **only when the provider
+  declares `tokenizer_path`**, so the hot path is untouched by default.
+  The `cache-observed` log line now carries `language_tax_usd` /
+  `language_tax_multiplier`, and `MetricsCollector` aggregates per-provider
+  + total language-tax spend (mirroring the cost-savings aggregation).
+
+- **`ProviderConfig.tokenizer_path`** — optional path to a local
+  `tokenizer.json` for accurate (language-tax) token counting. Local-file
+  only; never contacts the HuggingFace Hub. Inert when unset.
+
+- **`cjk_ratio_min` auto-route matcher.** A new `RuleMatcher` variant that
+  routes turns whose latest user message CJK ratio ≥ threshold to a
+  (typically local, tax-free) profile, while ASCII/code turns fall through
+  to the cloud chain. Per-turn property mirroring `code_fence_ratio_min`.
+
+  ```yaml
+  auto_router:
+    rules:
+      - match: { cjk_ratio_min: 0.3 }   # JA-heavy turns → local
+        profile: local
+      - match: { has_tools: true }
+        profile: cloud
+    default_rule_profile: cloud
+  ```
+
+- **Dashboard "Cost & Language Tax" panel** on `/dashboard`: total spend,
+  cache savings, and CJK language-tax spend (aggregate + per-provider).
+  Also surfaces the previously-hidden cost aggregates.
+
+- **`token_estimation.extract_text_from_anthropic_request()`** — pulls the
+  concatenated request text for the accurate tokenizer leg.
+
+### Security
+
+- **Bump starlette 1.0.1 → 1.3.1**, clearing four advisories
+  (CVE-2026-48817 / CVE-2026-48818 / CVE-2026-54282 / CVE-2026-54283) that
+  failed the `cve-audit` CI job (`pip-audit --strict`).
+
+### Notes
+
+- 38 new tests (`test_language_tax`, `test_language_tax_integration`,
+  `test_auto_router_cjk`, extended dashboard contract). Full suite:
+  **1250 passed, 8 skipped**. ruff clean. The 5-deps invariant is intact.
+
+---
+
 ## [v2.5.5] — 2026-06-06 (Claude Code >= 2.1.154 `system` role normalization)
 
 Patch release: ingress-side workaround for a Claude Code CLI regression.

{coderouter_cli-2.5.5 → coderouter_cli-2.6.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: coderouter-cli
-Version: 2.5.5
+Version: 2.6.1
 Summary: Local-first, free-first, fallback-built-in LLM router. Claude Code / OpenAI compatible.
 Project-URL: Homepage, https://github.com/zephel01/CodeRouter
 Project-URL: Repository, https://github.com/zephel01/CodeRouter
@@ -161,6 +161,36 @@ ANTHROPIC_BASE_URL=http://localhost:8088 ANTHROPIC_AUTH_TOKEN=dummy claude
 | **`coderouter replay`** | provider 切替の効果を統計比較 (A/B 分析) / `--suggest-rules` でルール最適化提案 |
 | **Continuous Probe** | idle 時も定期的に backend を監視 |
 
+### 言語税トラッキング — v2.6.0
+
+日本語などの CJK テキストは、クラウドのトークナイザだと「同じ意味の英語」より多くのトークンを消費します（**実測: GPT-4o 系 o200k で平均 1.6 倍、GPT-4 系 cl100k で平均 2.0 倍**）。ローカル LLM は課金されないので、この「言語税」はクラウド利用時だけ効いてきます。CodeRouter v2.6.0 はこれを **計測・ルーティング回避・可視化** します。
+
+| 機能 | 何をしてくれるか |
+|---|---|
+| **言語税の計測** | プロバイダに `tokenizer_path`（ローカルの `tokenizer.json`）を指定すると、char/4 ヒューリスティック比の実トークン倍率と割増 USD を算出（ネットワーク不要・未設定なら無効） |
+| **`cjk_ratio_min` ルーティング** | CJK 比率が高いリクエストを自動でローカル LLM（課金ゼロ）へ。コードや英語はクラウドへ |
+| **ダッシュボード可視化** | `/dashboard` の「Cost & Language Tax」パネルで総支出・キャッシュ節約・言語税をリアルタイム表示 |
+
+```yaml
+# providers.yaml — CJK 多めのターンはローカルへ自動回避
+auto_router:
+  rules:
+    - match: { cjk_ratio_min: 0.3 }   # 日本語が3割以上 → ローカル
+      profile: local
+    - match: { has_tools: true }      # ツール使用 → クラウド
+      profile: cloud
+  default_rule_profile: cloud
+
+providers:
+  - name: cloud-sonnet
+    kind: anthropic
+    base_url: https://api.anthropic.com
+    model: claude-sonnet-4-6
+    tokenizer_path: ~/.coderouter/tokenizers/sonnet.json   # 言語税の正確計測（任意）
+```
+
+詳細 → [言語税ガイド](./docs/guides/language-tax.md)
+
 ### Launcher — llama.cpp / vllm 起動 UI
 
 `http://localhost:8088/launcher` で開けるブラウザ UI。llama.cpp や vllm を GUI で起動・管理できます。
@@ -229,6 +259,7 @@ providers:
 | 使いこなす | [利用ガイド](./docs/guides/usage-guide.md) |
 | 無料で回す | [無料枠ガイド](./docs/guides/free-tier-guide.md) |
 | llama.cpp / vllm を GUI で起動 | [Launcher ガイド](./docs/backends/launcher.md) |
+| 言語税を計測・回避する | [言語税ガイド](./docs/guides/language-tax.md) |
 | 詰まった | [トラブルシューティング](./docs/guides/troubleshooting.md) |
 | 設計を知りたい | [アーキテクチャ詳細](./docs/concepts/architecture.md) |
 | 全リリース履歴 | [CHANGELOG](./CHANGELOG.md) |

{coderouter_cli-2.5.5 → coderouter_cli-2.6.1}/README.en.md

@@ -116,6 +116,36 @@ Full decision matrix → [Do I need CodeRouter?](./docs/start/when-do-i-need-cod
 | **`coderouter replay`** | Compare providers statistically (A/B analysis) / `--suggest-rules` for automated rule suggestions |
 | **Continuous Probe** | Background health monitoring even during idle |
 
+### Language Tax tracking — v2.6.0
+
+CJK text (Japanese / Chinese / Korean) costs more tokens on cloud tokenizers than the same meaning in English (**measured: ~1.6× on average with GPT-4o-era o200k, ~2.0× with GPT-4-era cl100k**). Local models bill nothing per token, so this "language tax" only bites on the cloud leg. CodeRouter v2.6.0 makes it **measurable, routable, and visible**.
+
+| Feature | What it does |
+|---|---|
+| **Language-tax measurement** | Set a provider's `tokenizer_path` (a local `tokenizer.json`) to compute the real token multiplier vs the char/4 heuristic and the extra USD. No network; inert when unset. |
+| **`cjk_ratio_min` routing** | Auto-route CJK-heavy turns to a local (tax-free) model; code/English falls through to the cloud chain. |
+| **Dashboard panel** | The `/dashboard` "Cost & Language Tax" panel shows total spend, cache savings, and language-tax spend live. |
+
+```yaml
+# providers.yaml — steer CJK-heavy turns to local
+auto_router:
+  rules:
+    - match: { cjk_ratio_min: 0.3 }   # >=30% CJK chars -> local
+      profile: local
+    - match: { has_tools: true }      # tool use -> cloud
+      profile: cloud
+  default_rule_profile: cloud
+
+providers:
+  - name: cloud-sonnet
+    kind: anthropic
+    base_url: https://api.anthropic.com
+    model: claude-sonnet-4-6
+    tokenizer_path: ~/.coderouter/tokenizers/sonnet.json   # accurate language-tax (optional)
+```
+
+Details → [Language Tax guide](./docs/guides/language-tax.en.md)
+
 ### Launcher — llama.cpp / vllm GUI
 
 Browser UI at `http://localhost:8088/launcher` for starting and managing local inference backends.
@@ -184,6 +214,7 @@ More detail → [Usage guide](./docs/guides/usage-guide.en.md) · [Architecture]
 | Use it well | [Usage guide](./docs/guides/usage-guide.en.md) |
 | Run for free | [Free-tier guide](./docs/guides/free-tier-guide.en.md) |
 | Launch llama.cpp / vllm via GUI | [Launcher guide](./docs/backends/launcher.md) |
+| Measure & avoid the language tax | [Language Tax guide](./docs/guides/language-tax.en.md) |
 | Stuck? | [Troubleshooting](./docs/guides/troubleshooting.en.md) |
 | Understand the design | [Architecture](./docs/concepts/architecture.md) |
 | Full release history | [CHANGELOG](./CHANGELOG.md) |

{coderouter_cli-2.5.5 → coderouter_cli-2.6.1}/README.md

@@ -116,6 +116,36 @@ ANTHROPIC_BASE_URL=http://localhost:8088 ANTHROPIC_AUTH_TOKEN=dummy claude
 | **`coderouter replay`** | provider 切替の効果を統計比較 (A/B 分析) / `--suggest-rules` でルール最適化提案 |
 | **Continuous Probe** | idle 時も定期的に backend を監視 |
 
+### 言語税トラッキング — v2.6.0
+
+日本語などの CJK テキストは、クラウドのトークナイザだと「同じ意味の英語」より多くのトークンを消費します（**実測: GPT-4o 系 o200k で平均 1.6 倍、GPT-4 系 cl100k で平均 2.0 倍**）。ローカル LLM は課金されないので、この「言語税」はクラウド利用時だけ効いてきます。CodeRouter v2.6.0 はこれを **計測・ルーティング回避・可視化** します。
+
+| 機能 | 何をしてくれるか |
+|---|---|
+| **言語税の計測** | プロバイダに `tokenizer_path`（ローカルの `tokenizer.json`）を指定すると、char/4 ヒューリスティック比の実トークン倍率と割増 USD を算出（ネットワーク不要・未設定なら無効） |
+| **`cjk_ratio_min` ルーティング** | CJK 比率が高いリクエストを自動でローカル LLM（課金ゼロ）へ。コードや英語はクラウドへ |
+| **ダッシュボード可視化** | `/dashboard` の「Cost & Language Tax」パネルで総支出・キャッシュ節約・言語税をリアルタイム表示 |
+
+```yaml
+# providers.yaml — CJK 多めのターンはローカルへ自動回避
+auto_router:
+  rules:
+    - match: { cjk_ratio_min: 0.3 }   # 日本語が3割以上 → ローカル
+      profile: local
+    - match: { has_tools: true }      # ツール使用 → クラウド
+      profile: cloud
+  default_rule_profile: cloud
+
+providers:
+  - name: cloud-sonnet
+    kind: anthropic
+    base_url: https://api.anthropic.com
+    model: claude-sonnet-4-6
+    tokenizer_path: ~/.coderouter/tokenizers/sonnet.json   # 言語税の正確計測（任意）
+```
+
+詳細 → [言語税ガイド](./docs/guides/language-tax.md)
+
 ### Launcher — llama.cpp / vllm 起動 UI
 
 `http://localhost:8088/launcher` で開けるブラウザ UI。llama.cpp や vllm を GUI で起動・管理できます。
@@ -184,6 +214,7 @@ providers:
 | 使いこなす | [利用ガイド](./docs/guides/usage-guide.md) |
 | 無料で回す | [無料枠ガイド](./docs/guides/free-tier-guide.md) |
 | llama.cpp / vllm を GUI で起動 | [Launcher ガイド](./docs/backends/launcher.md) |
+| 言語税を計測・回避する | [言語税ガイド](./docs/guides/language-tax.md) |
 | 詰まった | [トラブルシューティング](./docs/guides/troubleshooting.md) |
 | 設計を知りたい | [アーキテクチャ詳細](./docs/concepts/architecture.md) |
 | 全リリース履歴 | [CHANGELOG](./CHANGELOG.md) |

{coderouter_cli-2.5.5 → coderouter_cli-2.6.1}/coderouter/config/schemas.py

@@ -185,6 +185,19 @@ class ProviderConfig(BaseModel):
     )
     timeout_s: float = Field(default=30.0, ge=1.0, le=600.0)
 
+    # v2.6 language-tax track: path to a LOCAL ``tokenizer.json`` for this
+    # provider's model, used to measure the CJK over-count vs the char/4
+    # baseline (see ``coderouter.language_tax``). Loaded local-file-only —
+    # never contacts the HuggingFace Hub. When unset, language-tax falls
+    # back to char/4 (multiplier 1.0) and the feature is silently inert.
+    tokenizer_path: str | None = Field(
+        default=None,
+        description=(
+            "Local tokenizer.json for accurate (language-tax) token "
+            "counting. No network access. Requires the 'accuracy' extra."
+        ),
+    )
+
     # Provider-specific extras merged into the outbound request body.
     # Use for non-standard fields like Ollama's `think: false`, `keep_alive`,
     # `options.num_ctx`, or any vendor-specific toggle. User-supplied request
@@ -763,6 +776,16 @@ class RuleMatcher(BaseModel):
       ``request.tools`` set). The ``has_tools`` matcher is the
       profile-level lever for steering tool-laden traffic to the right
       chain entirely.
+
+    Variants (v2.6 / language-tax routing):
+
+    - ``cjk_ratio_min: 0.3`` — CJK character ratio of the latest user
+      message is ``>=`` this threshold. Routes CJK-heavy turns (which
+      pay the cloud "language tax" of ~1.2-1.5x more tokens) to a local
+      model that bills nothing per token, while ASCII/code turns fall
+      through to the cloud chain. Per-turn property like
+      ``code_fence_ratio_min``; see
+      :func:`coderouter.language_tax.cjk_char_ratio`.
     """
 
     model_config = ConfigDict(extra="forbid")
@@ -773,6 +796,13 @@ class RuleMatcher(BaseModel):
     content_regex: str | None = None
     model_pattern: str | None = None
     content_token_count_min: int | None = Field(default=None, ge=1)
+    # v2.6 language-tax routing: CJK character ratio of the latest user
+    # message >= this threshold. Lets operators steer CJK-heavy traffic
+    # (which carries the cloud language tax) to a local model that bills
+    # nothing per token. Operates on the latest user message like
+    # ``code_fence_ratio_min`` (a per-turn property), not the whole
+    # request. See ``coderouter.language_tax.cjk_char_ratio``.
+    cjk_ratio_min: float | None = Field(default=None, ge=0.0, le=1.0)
     # [Unreleased]: tool-aware routing (OpenClaw + Raspberry Pi 由来).
     # See class docstring "Variants ([Unreleased] / tool-aware routing)"
     # above for the full rationale. Boolean shape mirrors ``has_image`` —
@@ -789,6 +819,7 @@ class RuleMatcher(BaseModel):
         "model_pattern",
         "content_token_count_min",
         "has_tools",
+        "cjk_ratio_min",
     )
 
     @model_validator(mode="after")

{coderouter_cli-2.5.5 → coderouter_cli-2.6.1}/coderouter/cost.py

@@ -58,9 +58,13 @@ in the cost calc.
 from __future__ import annotations
 
 from dataclasses import dataclass
+from typing import TYPE_CHECKING
 
 from coderouter.config.schemas import CostConfig
 
+if TYPE_CHECKING:  # avoid an import cycle at runtime; used only for typing
+    from coderouter.language_tax import LanguageTaxBreakdown
+
 
 @dataclass(frozen=True)
 class CostBreakdown:
@@ -82,6 +86,12 @@ class CostBreakdown:
             chart. ``input_usd`` is "fresh input only" (does not
             include cache buckets); cache_read_usd / cache_creation_usd
             are the post-discount / post-premium values.
+        language_tax_multiplier: ``tokens_accurate / tokens_heuristic``
+            for the request text (v2.6 language-tax track). 1.0 when no
+            tax is measurable (English/code, or no accurate tokenizer).
+        language_tax_usd: USD share of ``total_usd`` attributable to the
+            CJK over-count vs CodeRouter's char/4 English baseline.
+            0.0 for free / local providers. See :mod:`coderouter.language_tax`.
     """
 
     total_usd: float = 0.0
@@ -90,6 +100,10 @@ class CostBreakdown:
     output_usd: float = 0.0
     cache_read_usd: float = 0.0
     cache_creation_usd: float = 0.0
+    # v2.6 language-tax track (additive; defaults keep pre-v2.6 behaviour
+    # and equality with a bare ``CostBreakdown()``).
+    language_tax_multiplier: float = 1.0
+    language_tax_usd: float = 0.0
 
 
 _PER_MILLION: float = 1_000_000.0
@@ -102,6 +116,7 @@ def compute_cost_for_attempt(
     output_tokens: int,
     cache_read_input_tokens: int,
     cache_creation_input_tokens: int,
+    language_tax: LanguageTaxBreakdown | None = None,
 ) -> CostBreakdown:
     """Translate per-attempt token counts into a USD :class:`CostBreakdown`.
 
@@ -144,6 +159,21 @@ def compute_cost_for_attempt(
     full_rate_for_cache_read = safe_read * input_rate
     savings_usd = full_rate_for_cache_read - cache_read_usd
 
+    # v2.6 language tax: the share of fresh-input spend attributable to
+    # the CJK over-count vs the char/4 English baseline. Defaults to a
+    # 1.0 multiplier / $0 when no LanguageTaxBreakdown is supplied, so
+    # the pre-v2.6 call shape is unchanged.
+    lt_multiplier = 1.0
+    lt_usd = 0.0
+    if language_tax is not None:
+        lt_multiplier = language_tax.tax_multiplier
+        from coderouter.language_tax import language_tax_usd
+
+        lt_usd = language_tax_usd(
+            language_tax.extra_tokens,
+            input_tokens_per_million=cost_config.input_tokens_per_million,
+        )
+
     return CostBreakdown(
         total_usd=total_usd,
         savings_usd=max(savings_usd, 0.0),
@@ -151,4 +181,6 @@ def compute_cost_for_attempt(
         output_usd=output_usd,
         cache_read_usd=cache_read_usd,
         cache_creation_usd=cache_creation_usd,
+        language_tax_multiplier=lt_multiplier,
+        language_tax_usd=lt_usd,
     )

{coderouter_cli-2.5.5 → coderouter_cli-2.6.1}/coderouter/ingress/dashboard_routes.py

@@ -165,6 +165,41 @@ _DASHBOARD_HTML = r"""<!doctype html>
   </main>
 
   <footer class="max-w-7xl mx-auto px-4 md:px-6 pb-8">
+    <!-- Panel: Cost & Language Tax (v2.6) -->
+    <section class="bg-slate-900/60 border border-slate-800 rounded-lg p-4 mb-4">
+      <h2 class="text-sm font-semibold uppercase tracking-wider text-slate-400 mb-3">Cost &amp; Language Tax</h2>
+      <div class="grid grid-cols-3 gap-3">
+        <div class="rounded-md bg-slate-800/50 p-3">
+          <div class="text-xs text-slate-400">Total spend</div>
+          <div class="text-2xl font-semibold tabnum" data-bind="cost_total">$0.00</div>
+        </div>
+        <div class="rounded-md bg-slate-800/50 p-3">
+          <div class="text-xs text-slate-400">Cache savings</div>
+          <div class="text-2xl font-semibold tabnum text-green-400" data-bind="cost_savings">$0.00</div>
+        </div>
+        <div class="rounded-md bg-slate-800/50 p-3">
+          <div class="text-xs text-slate-400">Language tax (CJK)</div>
+          <div class="text-2xl font-semibold tabnum text-amber-400" data-bind="language_tax_total">$0.00</div>
+          <div class="text-xs text-slate-500" data-bind="language_tax_hint">no tokenizer configured</div>
+        </div>
+      </div>
+      <div id="language-tax-by-provider" class="text-xs text-slate-400 tabnum mt-3"></div>
+      <!-- Token Savings: trim (core budget guard) + compress (plugin) -->
+      <div class="grid grid-cols-3 gap-3 mt-3">
+        <div class="rounded-md bg-slate-800/50 p-3">
+          <div class="text-xs text-slate-400">Tokens saved &middot; trim</div>
+          <div class="text-2xl font-semibold tabnum text-sky-400" data-bind="tokens_saved_trim">0</div>
+        </div>
+        <div class="rounded-md bg-slate-800/50 p-3">
+          <div class="text-xs text-slate-400">Tokens saved &middot; compress</div>
+          <div class="text-2xl font-semibold tabnum text-sky-400" data-bind="tokens_saved_compress">0</div>
+        </div>
+        <div class="rounded-md bg-slate-800/50 p-3">
+          <div class="text-xs text-slate-400">Tokens saved &middot; total</div>
+          <div class="text-2xl font-semibold tabnum text-green-400" data-bind="tokens_saved_total">0</div>
+        </div>
+      </div>
+    </section>
     <section class="bg-slate-900/60 border border-slate-800 rounded-lg p-4">
       <h2 class="text-sm font-semibold uppercase tracking-wider text-slate-400 mb-3">Usage Mix</h2>
       <div id="usage-bar" class="flex h-3 rounded-full overflow-hidden bg-slate-800" role="img" aria-label="usage mix"></div>
@@ -435,6 +470,35 @@ _DASHBOARD_HTML = r"""<!doctype html>
     {"&": "&amp;", "<": "&lt;", ">": "&gt;", '"': "&quot;", "'": "&#39;"}[c]
   ));
 
+  // v2.6: cost + language-tax panel. The collector zero-fills these, so
+  // a fresh/local-only deployment shows $0.00 across the board.
+  const renderCostTax = (snap) => {
+    const c = snap.counters || {};
+    const usd = (x) => "$" + (Number(x) || 0).toFixed(4);
+    setBind("cost_total", usd(c.cost_total_usd_aggregate));
+    setBind("cost_savings", usd(c.cost_savings_usd_aggregate));
+    const taxTotal = Number(c.language_tax_usd_aggregate) || 0;
+    setBind("language_tax_total", usd(taxTotal));
+    setBind("language_tax_hint",
+      taxTotal > 0 ? "extra paid for CJK vs char/4 baseline"
+                   : "no tax measured (set provider tokenizer_path)");
+    const byProv = c.language_tax_usd || {};
+    const el = document.getElementById("language-tax-by-provider");
+    const rows = Object.entries(byProv).filter(([, v]) => Number(v) > 0);
+    el.innerHTML = rows.length === 0 ? "" :
+      rows.map(([n, v]) =>
+        '<span class="mr-4"><span class="text-slate-500">' + escapeHTML(n) +
+        '</span> ' + usd(v) + '</span>').join("");
+    // token-savings tiles. trim comes from the core budget guard;
+    // compress from the optional plugin. Collector zero-fills, so an
+    // install with neither still renders "0" cleanly.
+    const ts = c.tokens_saved_by_mechanism || {};
+    const intfmt = (x) => (Number(x) || 0).toLocaleString();
+    setBind("tokens_saved_trim", intfmt(ts.trim));
+    setBind("tokens_saved_compress", intfmt(ts.compress));
+    setBind("tokens_saved_total", intfmt(c.tokens_saved_total));
+  };
+
   const renderSnapshot = (snap) => {
     const startup = snap.startup || {};
     const cfg = snap.config || {};
@@ -451,6 +515,7 @@ _DASHBOARD_HTML = r"""<!doctype html>
     renderSparkline(snap);
     renderRecent(snap);
     renderUsageMix(snap);
+    renderCostTax(snap);
   };
 
   const renderError = (msg) => {