feat: add Grok Build CLI usage adapter - #1447

Check if this issue is valid — if so, understand the root cause and fix it. At rust/crates/ccusage/src/main.rs, line 157:

<comment>The `None` (default `ccusage`) arm does not hydrate `grok_path` from config. Other agent commands (pi, openclaw, grok) call `config.apply_agent_args(...)` with their respective path parameter. If a user has configured `grokPath` in their config file and runs bare `ccusage`, the Grok adapter will not pick up the configured path, potentially causing missed Grok data in the unified report.</comment>

<file context>
@@ -146,12 +147,14 @@ fn main() -> Result<()> {
                 kind: AgentReportKind::Daily,
                 pi_path: None,
                 open_claw_path: None,
+                grok_path: None,
                 codex_speed: cli::CodexSpeed::Auto,
             };
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At rust/crates/ccusage/src/pricing.rs, line 424:

<comment>Candidate-based reports fail to price configured aliases for models available only through models.dev, producing a missing-pricing warning and $0 calculated cost. Resolve each candidate's configured alias before both fallback lookups, matching `find()` behavior.</comment>

<file context>
@@ -387,6 +388,62 @@ impl PricingMap {
+                        models_dev_pricing().and_then(|pricing| {
+                            candidates
+                                .iter()
+                                .find_map(|candidate| pricing.find_entry_or_alias(candidate))
+                        })
+                    })
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At rust/crates/ccusage/src/adapter/grok/parser.rs, line 325:

<comment>Rows with only `totalTokens`, or totals exceeding their breakdown, are omitted or undercounted in reports and cost. Normalize each model row with `apply_total_token_fallback` before this filter, billing the remainder as output-equivalent tokens.</comment>

<file context>
@@ -0,0 +1,736 @@
+        let output_tokens = model_usage.output_tokens;
+        let cache_read = model_usage.cached_read_tokens;
+        let reasoning_tokens = model_usage.reasoning_tokens;
+        if input_tokens == 0 && output_tokens == 0 && cache_read == 0 && reasoning_tokens == 0 {
+            // Skip zero-token rows (including empty totalTokens-only metadata).
+            if model_usage.total_tokens == 0 {
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At rust/crates/ccusage/src/adapter/grok/parser.rs, line 356:

<comment>Distinct no-`eventId` turns for one model in the same timestamp are dropped, underreporting usage and cost. Include the token tuple in this fallback ID so only identical rows dedupe.</comment>

<file context>
@@ -0,0 +1,736 @@
+        let message_id = event_id
+            .as_ref()
+            .map(|id| format!("{id}:{raw_model}"))
+            .or_else(|| Some(format!("{session_id}:{timestamp_text}:{raw_model}")));
+        let data = UsageEntry {
+            session_id: Some(session_id.clone()),
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At rust/crates/ccusage/src/adapter/grok/parser.rs, line 212:

<comment>`summary.info.id` is parsed from `summary.json` but discarded — the comment says to prefer it over the path-derived session ID, but the assignment was never completed. Either complete the override (`session_id = id;`) or remove the dead parsing block and its comment.</comment>

<file context>
@@ -0,0 +1,736 @@
+            .and_then(|info| info.id.clone())
+            .filter(|id| !id.is_empty())
+        {
+            // Prefer explicit summary session id when present.
+            let _ = id;
+        }
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At rust/crates/ccusage/src/adapter/grok/report.rs, line 11:

<comment>Grok report behavior now has a byte-for-byte copy of OpenClaw's report implementation. A shared agent-report helper would prevent future JSON/aggregation fixes from drifting between adapters.</comment>

<file context>
@@ -0,0 +1,73 @@
+    summarize_by_key, summarize_summaries_by_bucket, totals_json,
+};
+
+pub(crate) fn report_from_rows(rows: &[crate::UsageSummary], kind: AgentReportKind) -> Value {
+    let rows_json = rows
+        .iter()
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At rust/crates/ccusage/src/output.rs, line 374:

<comment>Dynamic-router users with a known blended/contract rate are incorrectly told not to configure it unless logs expose the underlying model. An override keyed to `{model}` works; describe the resolved model as an accuracy improvement rather than a prerequisite.</comment>

<file context>
@@ -369,19 +369,42 @@ pub(crate) fn missing_pricing_warnings_for_models<'a>(
+            if is_dynamic_router_model(model) {
                 format!(
-                    "WARN  Missing embedded pricing for {model}; cost excludes this model. Run without --offline or update ccusage after pricing is added."
+                    "WARN  No deterministic token rate for dynamic model router {model}; cost excludes this usage. Configure pricingOverrides only if your logs expose the resolved underlying model."
+                )
+            } else if is_unsupported_or_undetected_model(model) {
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At docs/guide/grok/index.md, line 105:

<comment>The example table's Total Tokens (124,015) exceeds the sum of visible columns (116,944) by 7,071, and the $0.24 cost doesn't match the documented pricing rates. The difference is reasoning tokens, but there's no reasoning column or inline note in the table. Consider either adding a reasoning tokens column, showing a footnote in the table, or making the example numbers balance cleanly (total = input + output + cache_create + cache_read) so users aren't confused by the gap.</comment>

<file context>
@@ -0,0 +1,174 @@
+┌────────────┬──────────────────────────┬───────────┬───────────┬──────────────┬────────────┬──────────────┬──────────────┐
+│ Date       │ Models                   │     Input │    Output │ Cache Create │ Cache Read │ Total Tokens │   Cost (USD) │
+├────────────┼──────────────────────────┼───────────┼───────────┼──────────────┼────────────┼──────────────┼──────────────┤
+│ 2026-07-13 │ - [grok] grok-4.5        │    70,378 │     8,294 │            0 │     38,272 │      124,015 │       $0.24 │
+├────────────┼──────────────────────────┼───────────┼───────────┼──────────────┼────────────┼──────────────┼──────────────┤
+│ Total      │                          │    70,378 │     8,294 │            0 │     38,272 │      124,015 │       $0.24 │
</file context>

Check if this issue is valid — if so, understand the root cause and fix it. At docs/guide/all-reports.md, line 52:

<comment>Same naming inconsistency as in the index.md table: the all-reports summary table says "Grok Build" instead of "Grok Build CLI". Consider aligning to match the product name used in the sidebar, Grok guide, and apps/ccusage README.</comment>

<file context>
@@ -49,6 +49,7 @@ Unified tables include an **Agent** column so you can compare sources in one vie
 | pi-agent     | `pi`       | `ccusage pi monthly`      |
 | Goose        | `goose`    | `ccusage goose daily`     |
 | OpenClaw     | `openclaw` | `ccusage openclaw daily`  |
+| Grok Build   | `grok`     | `ccusage grok daily`      |
 | Kilo         | `kilo`     | `ccusage kilo daily`      |
 | Kimi         | `kimi`     | `ccusage kimi daily`      |
</file context>