@workweave/router 0.2.4 → 0.2.6

package/README.md

@@ -84,6 +84,18 @@ for the full reference.
 
 ## Why npx
 
-`curl -fsSL https://weave.ai/cc/install.sh | sh` still works and is fine.
-`npx @workweave/router` adds: Windows support via Git Bash, painless version
-pinning, no `curl | sh` aversion, and discoverability via the npm registry.
+`npx @workweave/router` gives Windows support via Git Bash, painless version
+pinning, and discoverability via the npm registry.
+
+## Older npm
+
+On npm ≤ 6 the bundled `npx` treats an undeclared `-y` as consuming the next
+token, so `npx -y @workweave/router --claude` silently drops the package name
+and resolves the following argument as the command instead. Either upgrade
+(`npm i -g npm@latest`) or name the binary explicitly:
+
+```bash
+npx --package @workweave/router -y -- weave-router --claude
+```
+
+That form is correct on every npm version.

package/cc-statusline.sh

@@ -130,14 +130,19 @@ normalize_model() {
 prices='{
   "input": {
     "claude-fable-5":                   0.01,
-    "claude-haiku-4-5":                 0.0008,
+    "claude-haiku-4-5":                 0.001,
+    "claude-opus-4-0":                  0.015,
+    "claude-opus-4-1":                  0.015,
+    "claude-opus-4-5":                  0.005,
     "claude-opus-4-6":                  0.005,
     "claude-opus-4-7":                  0.005,
     "claude-opus-4-8":                  0.005,
+    "claude-opus-5":                    0.005,
     "claude-sonnet-4-5":                0.003,
     "claude-sonnet-4-6":                0.003,
-    "deepseek/deepseek-v4-flash":       0.00014,
-    "deepseek/deepseek-v4-pro":         0.00174,
+    "claude-sonnet-5":                  0.003,
+    "deepseek/deepseek-v4-flash":       0.0001134,
+    "deepseek/deepseek-v4-pro":         0.001318,
     "gemini-2.0-flash":                 0.0001,
     "gemini-2.0-flash-lite":            0.000075,
     "gemini-2.5-flash":                 0.0003,
@@ -148,6 +153,9 @@ prices='{
     "gemini-3.1-flash-lite-preview":    0.0001,
     "gemini-3.1-pro-preview":           0.002,
     "gemini-3.5-flash":                 0.0015,
+    "gemini-3.5-flash-lite":            0.0003,
+    "gemini-3.6-flash":                 0.0015,
+    "google/gemma-4-26b-a4b-it":        0.00015,
     "gpt-4.1":                          0.002,
     "gpt-4.1-mini":                     0.0004,
     "gpt-4.1-nano":                     0.0001,
@@ -157,43 +165,53 @@ prices='{
     "gpt-5-chat":                       0.0025,
     "gpt-5-mini":                       0.0005,
     "gpt-5-nano":                       0.0001,
-    "gpt-5.4":                          0.003,
-    "gpt-5.4-mini":                     0.0004,
-    "gpt-5.4-nano":                     0.0001,
-    "gpt-5.4-pro":                      0.02,
+    "gpt-5.4":                          0.0025,
+    "gpt-5.4-mini":                     0.00075,
+    "gpt-5.4-nano":                     0.0002,
+    "gpt-5.4-pro":                      0.03,
     "gpt-5.5":                          0.005,
     "gpt-5.5-mini":                     0.0005,
     "gpt-5.5-nano":                     0.00015,
     "gpt-5.5-pro":                      0.03,
+    "gpt-5.6-luna":                     0.001,
+    "gpt-5.6-sol":                      0.005,
+    "gpt-5.6-terra":                    0.0025,
+    "grok-4.5":                         0.002,
     "minimax/minimax-m2.7":             0.0003,
     "minimax/minimax-m3":               0.0003,
     "mistralai/mistral-small-2603":     0.0002,
     "moonshotai/kimi-k2.5":             0.0006,
     "moonshotai/kimi-k2.6":             0.00095,
     "moonshotai/kimi-k2.7":             0.00095,
+    "moonshotai/kimi-k3":               0.003,
     "qwen/qwen3-235b-a22b-2507":        0.0002266,
     "qwen/qwen3-30b-a3b-instruct-2507": 0.00015,
     "qwen/qwen3-coder":                 0.0009,
     "qwen/qwen3-coder-next":            0.0005,
     "qwen/qwen3-next-80b-a3b-instruct": 0.00015,
     "qwen/qwen3.5-flash-02-23":         0.00005,
-    "qwen/qwen3.6-35b-a3b":             0.00015,
+    "qwen/qwen3.6-35b-a3b":             0.000172,
     "qwen/qwen3.7-plus":                0.0004,
     "xiaomi/mimo-v2.5-pro":             0.001,
-    "z-ai/glm-5":                       0.0006,
-    "z-ai/glm-5.1":                     0.00105,
+    "z-ai/glm-5":                       0.001,
+    "z-ai/glm-5.1":                     0.0014,
     "z-ai/glm-5.2":                     0.0014
   },
   "output": {
     "claude-fable-5":                   0.05,
-    "claude-haiku-4-5":                 0.004,
+    "claude-haiku-4-5":                 0.005,
+    "claude-opus-4-0":                  0.075,
+    "claude-opus-4-1":                  0.075,
+    "claude-opus-4-5":                  0.025,
     "claude-opus-4-6":                  0.025,
     "claude-opus-4-7":                  0.025,
     "claude-opus-4-8":                  0.025,
+    "claude-opus-5":                    0.025,
     "claude-sonnet-4-5":                0.015,
     "claude-sonnet-4-6":                0.015,
-    "deepseek/deepseek-v4-flash":       0.00028,
-    "deepseek/deepseek-v4-pro":         0.00348,
+    "claude-sonnet-5":                  0.015,
+    "deepseek/deepseek-v4-flash":       0.0002791,
+    "deepseek/deepseek-v4-pro":         0.0026361,
     "gemini-2.0-flash":                 0.0004,
     "gemini-2.0-flash-lite":            0.0003,
     "gemini-2.5-flash":                 0.0012,
@@ -204,6 +222,9 @@ prices='{
     "gemini-3.1-flash-lite-preview":    0.0004,
     "gemini-3.1-pro-preview":           0.008,
     "gemini-3.5-flash":                 0.009,
+    "gemini-3.5-flash-lite":            0.0025,
+    "gemini-3.6-flash":                 0.0075,
+    "google/gemma-4-26b-a4b-it":        0.0006,
     "gpt-4.1":                          0.008,
     "gpt-4.1-mini":                     0.0016,
     "gpt-4.1-nano":                     0.0004,
@@ -213,31 +234,36 @@ prices='{
     "gpt-5-chat":                       0.01,
     "gpt-5-mini":                       0.002,
     "gpt-5-nano":                       0.0004,
-    "gpt-5.4":                          0.012,
-    "gpt-5.4-mini":                     0.0016,
-    "gpt-5.4-nano":                     0.0004,
-    "gpt-5.4-pro":                      0.08,
-    "gpt-5.5":                          0.04,
+    "gpt-5.4":                          0.015,
+    "gpt-5.4-mini":                     0.0045,
+    "gpt-5.4-nano":                     0.00125,
+    "gpt-5.4-pro":                      0.18,
+    "gpt-5.5":                          0.03,
     "gpt-5.5-mini":                     0.0025,
     "gpt-5.5-nano":                     0.0006,
-    "gpt-5.5-pro":                      0.12,
+    "gpt-5.5-pro":                      0.18,
+    "gpt-5.6-luna":                     0.006,
+    "gpt-5.6-sol":                      0.03,
+    "gpt-5.6-terra":                    0.015,
+    "grok-4.5":                         0.006,
     "minimax/minimax-m2.7":             0.0012,
     "minimax/minimax-m3":               0.0012,
     "mistralai/mistral-small-2603":     0.0006,
     "moonshotai/kimi-k2.5":             0.003,
     "moonshotai/kimi-k2.6":             0.004,
     "moonshotai/kimi-k2.7":             0.004,
+    "moonshotai/kimi-k3":               0.015,
     "qwen/qwen3-235b-a22b-2507":        0.0009064,
     "qwen/qwen3-30b-a3b-instruct-2507": 0.0006,
     "qwen/qwen3-coder":                 0.0027,
     "qwen/qwen3-coder-next":            0.0012,
     "qwen/qwen3-next-80b-a3b-instruct": 0.0012,
     "qwen/qwen3.5-flash-02-23":         0.00015,
-    "qwen/qwen3.6-35b-a3b":             0.00095,
+    "qwen/qwen3.6-35b-a3b":             0.0012002,
     "qwen/qwen3.7-plus":                0.0016,
     "xiaomi/mimo-v2.5-pro":             0.003,
-    "z-ai/glm-5":                       0.00208,
-    "z-ai/glm-5.1":                     0.0035,
+    "z-ai/glm-5":                       0.0032,
+    "z-ai/glm-5.1":                     0.0044,
     "z-ai/glm-5.2":                     0.0044
   }
 }'
@@ -427,20 +453,23 @@ elif [[ "$routed" == "failure" ]]; then
   # swap or compute savings against a non-model.
   printf '%s — %s%s' "$brand" "$routed" "$tokens_clause"
 elif [[ -n "$routed" ]]; then
-  # Show the savings clause only when the session is genuinely net-saving.
+  # Always show the savings clause, flooring a non-positive total at $0.00.
   # session_savings is "0.0000" on fresh sessions or sessions where every
-  # turn routed back to the selected model; it can also go negative when
-  # sticky routing forces a haiku-tagged side-call up to a cached
-  # sonnet/opus decision. In both cases the word "saved" would mislead,
-  # so drop the savings clause but keep the token totals.
-  has_savings="false"
-  if [[ -n "$session_savings" ]] \
-     && awk -v v="$session_savings" 'BEGIN{exit !(v+0 > 0.005)}'; then
-    has_savings="true"
+  # turn routed back to the selected model, and can go negative when routing
+  # upgrades a turn to a pricier model (e.g. a sticky/hard-pinned side-call,
+  # or a hard turn escalated to opus). "saved -$X" would mislead, so clamp
+  # the display to $0.00 rather than dropping the clause — a $0.00 readout
+  # tells the user the router ran and simply didn't beat their selection.
+  # When the CC selection is unknown ("?" or empty) there's nothing to
+  # compare against, so drop the "← selection" arrow and just show routed.
+  display_savings="$session_savings"
+  if [[ -z "$display_savings" ]] \
+     || awk -v v="$display_savings" 'BEGIN{exit !(v+0 < 0)}'; then
+    display_savings="0"
   fi
-  if [[ "$has_savings" == "true" ]]; then
+  if [[ -n "$selected_display" && "$selected_display" != "?" ]]; then
     printf '%s — %s ← %s · saved %s%s' \
-      "$brand" "$routed" "$selected_display" "$(fmt_money "$session_savings")" "$tokens_clause"
+      "$brand" "$routed" "$selected_display" "$(fmt_money "$display_savings")" "$tokens_clause"
   else
     printf '%s — %s%s' "$brand" "$routed" "$tokens_clause"
   fi

package/commands/router-session.md

@@ -0,0 +1,15 @@
+---
+description: Print the session id that correlates this session in router telemetry and logs.
+allowed-tools: Bash(echo:*)
+---
+
+Print the session id used by the Weave Router to correlate this session's
+telemetry, logs, and feedback submissions. Also works as a transcript
+filename key for manual lookups in the router dashboard.
+
+The router carries this id as `X-Claude-Code-Session-Id` on every request and
+stores it in `model_router_users.SessionID` for analytics joins.
+
+Run `echo "$CLAUDE_CODE_SESSION_ID"`.
+
+Then tell me the session id in one line, formatted as inline code so I can copy it.

package/install.sh

@@ -127,7 +127,10 @@ skip() { printf "%s⊙%s %s%s%s\n" "$C_DIM" "$C_RESET" "$C_DIM" "$*" "$C_RESET";
 # print after a successful install is copy-paste-correct. Kept in sync with
 # uninstall.sh's flag surface.
 uninstall_cmd() {
-  local cmd="npx -y @workweave/router --uninstall"
+  # --package + `--` is load-bearing: npm <= 6's bundled npx treats an
+  # undeclared `-y` as consuming the NEXT token, so `npx -y @workweave/router`
+  # loses the package name and resolves whatever follows as the command.
+  local cmd="npx --package @workweave/router -y -- weave-router --uninstall"
   case "$target" in
     codex)    cmd="$cmd --codex" ;;
     opencode) cmd="$cmd --opencode" ;;
@@ -577,11 +580,10 @@ write_opencode_config() {
     | (if $name  != "" then . + {"X-Weave-User-Name":  $name } else . end)
   ')"
 
-  # Headline models we surface in opencode's picker. The router re-routes
-  # each request anyway, so this list is mostly UX — what shows up when the
-  # user runs /models inside opencode. We list a mix of GPT + Claude so the
-  # picker reflects that the router spans families; whichever model serves the
-  # turn, its matching subscription (if connected) pays.
+  # Surface one Auto choice in opencode's picker. The router chooses the
+  # upstream model for every request, so presenting pinned model names would
+  # imply a choice that the router intentionally does not honor. Whichever
+  # model serves a turn uses its matching subscription when one is connected.
   #
   # npm is @ai-sdk/openai and baseURL KEEPS its /v1 here: opencode's
   # @ai-sdk/openai provider appends /responses, yielding the router's
@@ -601,11 +603,7 @@ write_opencode_config() {
       name: "Weave Router",
       options: { apiKey: $key, baseURL: $url, headers: $headers },
       models: {
-        "claude-sonnet-4-6": { name: "Claude Sonnet 4.6 (via Weave Router)" },
-        "claude-opus-4-8":   { name: "Claude Opus 4.8 (via Weave Router)" },
-        "claude-haiku-4-5":  { name: "Claude Haiku 4.5 (via Weave Router)" },
-        "gpt-5.5":           { name: "GPT-5.5 (via Weave Router)" },
-        "gpt-5.4":           { name: "GPT-5.4 (via Weave Router)" }
+        auto: { name: "Auto" }
       }
     }
   ')"
@@ -655,8 +653,9 @@ write_opencode_config() {
 
   # Merge into any existing opencode.json. We always overwrite provider.weave
   # so re-install reflects the latest key/identity, but we leave the rest of the
-  # file (other providers, mcp, agent settings) untouched. Top-level `model` is
-  # only set when the user hasn't already picked one.
+  # file (other providers, mcp, agent settings) untouched. A previously
+  # installed `weave/*` model is migrated to `weave/auto`; unrelated provider
+  # choices stay untouched.
   #
   # The weave-claude login provider AND the plugin entry are written together
   # only when the bundled plugin was present and copied ($plugin non-empty): the
@@ -684,12 +683,12 @@ write_opencode_config() {
                    then (.plugin -= [$pluginspec]) | (if (.plugin | length) == 0 then del(.plugin) else . end)
                    else . end)
          end)
-      | (if (.model // "") == "" then .model = "weave/claude-sonnet-4-6" else . end)
-      # Reset the default model if it points at a provider we just removed: the
-      # retired weave-codex, or weave-claude on a plugin-less re-install (it has
-      # no models anyway). Otherwise opencode boots with a dangling model.
-      | (if (.model // "" | tostring | startswith("weave-codex/")) then .model = "weave/claude-sonnet-4-6" else . end)
-      | (if $plugin == "" and (.model // "" | tostring | startswith("weave-claude/")) then .model = "weave/claude-sonnet-4-6" else . end)
+      | (if (.model // "") == "" then .model = "weave/auto" else . end)
+      # Replace any legacy Weave model choice with the single auto-routing
+      # choice. Models from unrelated providers remain unchanged.
+      | (if (.model // "" | tostring) as $model
+           | ($model | startswith("weave/") or startswith("weave-codex/") or startswith("weave-claude/"))
+           then .model = "weave/auto" else . end)
       | (.["$schema"] //= "https://opencode.ai/config.json")
     ' "$config_file")"
   else
@@ -699,7 +698,7 @@ write_opencode_config() {
       --arg plugin "$plugin_arg" '
       {
         "$schema": "https://opencode.ai/config.json",
-        model: "weave/claude-sonnet-4-6",
+        model: "weave/auto",
         provider: { weave: $block }
       }
       | (if $plugin != "" then .provider["weave-claude"] = $claude | .plugin = [$plugin] else . end)
@@ -770,10 +769,12 @@ write_pi_models_config() {
       headers: $headers,
       models: [
         { id: "claude-fable-5",    name: "Claude Fable 5 (via Weave Router)",    reasoning: true, input: ["text","image"], contextWindow: 1000000, maxTokens: 128000 },
-        { id: "claude-opus-4-8",   name: "Claude Opus 4.8 (via Weave Router)",   reasoning: true, input: ["text","image"], contextWindow: 200000, maxTokens: 64000 },
+        { id: "claude-opus-5",     name: "Claude Opus 5 (via Weave Router)",     reasoning: true, input: ["text","image"], contextWindow: 1000000, maxTokens: 128000 },
         { id: "claude-opus-4-7",   name: "Claude Opus 4.7 (via Weave Router)",   reasoning: true, input: ["text","image"], contextWindow: 200000, maxTokens: 64000 },
         { id: "claude-sonnet-4-6", name: "Claude Sonnet 4.6 (via Weave Router)", reasoning: true, input: ["text","image"], contextWindow: 200000, maxTokens: 64000 },
-        { id: "claude-haiku-4-5",  name: "Claude Haiku 4.5 (via Weave Router)",  reasoning: true, input: ["text","image"], contextWindow: 200000, maxTokens: 32000 }
+        { id: "claude-haiku-4-5",  name: "Claude Haiku 4.5 (via Weave Router)",  reasoning: true, input: ["text","image"], contextWindow: 200000, maxTokens: 32000 },
+        { id: "gpt-5.6-sol",       name: "GPT-5.6 Sol (via Weave Router)",       reasoning: true, input: ["text","image"], contextWindow: 1050000, maxTokens: 128000 },
+        { id: "grok-4.5",          name: "Grok 4.5 (via Weave Router)",          reasoning: true, input: ["text","image"], contextWindow: 500000, maxTokens: 131072 }
       ]
     }
   ')"
@@ -1402,9 +1403,9 @@ toggle_claude() {
       else
         # Park just the router-owned env keys, then strip them so Claude Code
         # falls back to its own Anthropic login.
-        jq '{env: ((.env // {}) | {ANTHROPIC_BASE_URL, ANTHROPIC_CUSTOM_HEADERS} | with_entries(select(.value != null)))}' "$settings_file" >"$parked"
+        jq '{env: ((.env // {}) | {ANTHROPIC_BASE_URL, ANTHROPIC_CUSTOM_HEADERS, ENABLE_TOOL_SEARCH} | with_entries(select(.value != null)))}' "$settings_file" >"$parked"
         chmod 600 "$parked"
-        merged="$(jq '(.env // {}) |= del(.ANTHROPIC_BASE_URL, .ANTHROPIC_CUSTOM_HEADERS)
+        merged="$(jq '(.env // {}) |= del(.ANTHROPIC_BASE_URL, .ANTHROPIC_CUSTOM_HEADERS, .ENABLE_TOOL_SEARCH)
                       | (if (.env // {} | length) == 0 then del(.env) else . end)' "$settings_file")"
         printf '%s\n' "$merged" >"$settings_file"
       fi
@@ -1572,30 +1573,34 @@ toggle_opencode() {
       ;;
     on)
       if [ "$on" = "true" ]; then ok "opencode is already on — nothing to do."; return 0; fi
-      restore_model="weave/claude-sonnet-4-6"
+      restore_model="weave/auto"
       if [ "$parked_present" = "true" ]; then
-        restore_model="$(jq -r '.model // "weave/claude-sonnet-4-6"' "$parked")"
+        restore_model="$(jq -r '.model // "weave/auto"' "$parked")"
       elif [ "$has_weave" != "true" ]; then
         warn "opencode isn't configured for the router. Run the installer first."; return 0
       else
         # No parked model (sidecar deleted by hand). Derive the default from the
         # installed provider.weave.models block rather than a hardcoded literal
-        # that silently diverges when the installer's default changes — prefer a
-        # sonnet entry, else the first model the installer registered.
+        # that silently diverges when the installer's default changes — prefer
+        # the auto-routing entry, else the first model the installer registered.
         restore_model="$(jq -r '
           (.provider.weave.models // {} | keys) as $k
-          | (([$k[] | select(test("sonnet"))] | first) // $k[0] // "claude-sonnet-4-6")
+          | (([$k[] | select(. == "auto")] | first) // $k[0] // "auto")
           | "weave/" + .
-        ' "$f" 2>/dev/null || echo "weave/claude-sonnet-4-6")"
+        ' "$f" 2>/dev/null || echo "weave/auto")"
       fi
-      # Never restore a legacy weave-codex model whose provider is no longer
-      # registered (every current install strips weave-codex). Fall back to the
-      # `weave` default so `on` can't leave opencode pointing at a missing
-      # provider.
+      # Restore only a registered `weave` model. Legacy installations may have
+      # parked a pinned model, but a current install exposes only `weave/auto`.
       case "$restore_model" in
+        weave/*)
+          local model_id="${restore_model#weave/}"
+          if ! jq -e --arg id "$model_id" '(.provider.weave.models // {}) | has($id)' "$f" >/dev/null 2>&1; then
+            restore_model="weave/auto"
+          fi
+          ;;
         weave-codex/*)
           if [ "$(jq -r '((.provider // {}) | has("weave-codex"))' "$f" 2>/dev/null || true)" != "true" ]; then
-            restore_model="weave/claude-sonnet-4-6"
+            restore_model="weave/auto"
           fi
           ;;
       esac
@@ -1688,6 +1693,7 @@ fi
 # Layout:
 #   Claude:  <settings_dir>/commands/{force-model,unforce-model}.md  → /force-model
 #   Codex:   <codex_dir>/prompts/{force-model,unforce-model}.md      → /prompts:force-model
+#   opencode: <commands_dir>/{force-model,unforce-model}.md          → /force-model
 #
 # Files come from install/commands/ in the repo (or the colocated commands/
 # directory the npm package ships alongside install.sh).
@@ -1718,8 +1724,8 @@ install_slash_commands() {
   installed="force-model, unforce-model, router-feedback, fm, ufm, rf"
   cmds="force-model unforce-model router-feedback fm ufm rf"
   if [ "$target" = "claude" ]; then
-    cmds="$cmds router-off router-on router-status"
-    installed="$installed, router-off, router-on, router-status"
+    cmds="$cmds router-off router-on router-status router-session"
+    installed="$installed, router-off, router-on, router-status, router-session"
   fi
 
   # Bake the same scope selector the toggle needs to find this install: --dir
@@ -1829,6 +1835,26 @@ if [ "$target" = "opencode" ]; then
     ok "Updated $gitignore (ignored opencode.json, .weave/)"
   fi
 
+  # Slash command wrappers: opencode discovers commands in *.md files under
+  # ~/.config/opencode/commands/ (user scope) and .opencode/commands/ (project
+  # scope). Install command wrappers for /rf (±), /force-model, /unforce-model
+  # so typing /rf + in the TUI expands to /router-feedback + and reaches the
+  # router's server-side feedback interceptor, same as Claude Code + Codex.
+  #
+  # Router on/off/status are not installed — they run npx shell commands
+  # specific to the Claude Code settings model and don't apply to opencode.
+  if [ "$scope" = "project" ]; then
+    opencode_commands_dir="$opencode_dir/.opencode/commands"
+  else
+    # User scope and --dir installs: use the global commands path that opencode
+    # discovers regardless of working directory.
+    opencode_commands_dir="${XDG_CONFIG_HOME:-$HOME/.config}/opencode/commands"
+  fi
+  install_slash_commands "$opencode_commands_dir"
+  if [ "$scope" = "user" ] || [ -n "$install_dir" ]; then
+    info "opencode restart required for commands to take effect."
+  fi
+
   # Post-install verification: same probes the Claude/Codex paths run.
   if [ "$quiet" != "true" ]; then
     if ! spin "Pinging $base_url/health" curl -fsS --max-time 5 "$base_url/health"; then
@@ -2070,14 +2096,19 @@ normalize_model() {
 prices='{
   "input": {
     "claude-fable-5":                   0.01,
-    "claude-haiku-4-5":                 0.0008,
+    "claude-haiku-4-5":                 0.001,
+    "claude-opus-4-0":                  0.015,
+    "claude-opus-4-1":                  0.015,
+    "claude-opus-4-5":                  0.005,
     "claude-opus-4-6":                  0.005,
     "claude-opus-4-7":                  0.005,
     "claude-opus-4-8":                  0.005,
+    "claude-opus-5":                    0.005,
     "claude-sonnet-4-5":                0.003,
     "claude-sonnet-4-6":                0.003,
-    "deepseek/deepseek-v4-flash":       0.00014,
-    "deepseek/deepseek-v4-pro":         0.00174,
+    "claude-sonnet-5":                  0.003,
+    "deepseek/deepseek-v4-flash":       0.0001134,
+    "deepseek/deepseek-v4-pro":         0.001318,
     "gemini-2.0-flash":                 0.0001,
     "gemini-2.0-flash-lite":            0.000075,
     "gemini-2.5-flash":                 0.0003,
@@ -2088,6 +2119,9 @@ prices='{
     "gemini-3.1-flash-lite-preview":    0.0001,
     "gemini-3.1-pro-preview":           0.002,
     "gemini-3.5-flash":                 0.0015,
+    "gemini-3.5-flash-lite":            0.0003,
+    "gemini-3.6-flash":                 0.0015,
+    "google/gemma-4-26b-a4b-it":        0.00015,
     "gpt-4.1":                          0.002,
     "gpt-4.1-mini":                     0.0004,
     "gpt-4.1-nano":                     0.0001,
@@ -2097,43 +2131,53 @@ prices='{
     "gpt-5-chat":                       0.0025,
     "gpt-5-mini":                       0.0005,
     "gpt-5-nano":                       0.0001,
-    "gpt-5.4":                          0.003,
-    "gpt-5.4-mini":                     0.0004,
-    "gpt-5.4-nano":                     0.0001,
-    "gpt-5.4-pro":                      0.02,
+    "gpt-5.4":                          0.0025,
+    "gpt-5.4-mini":                     0.00075,
+    "gpt-5.4-nano":                     0.0002,
+    "gpt-5.4-pro":                      0.03,
     "gpt-5.5":                          0.005,
     "gpt-5.5-mini":                     0.0005,
     "gpt-5.5-nano":                     0.00015,
     "gpt-5.5-pro":                      0.03,
+    "gpt-5.6-luna":                     0.001,
+    "gpt-5.6-sol":                      0.005,
+    "gpt-5.6-terra":                    0.0025,
+    "grok-4.5":                         0.002,
     "minimax/minimax-m2.7":             0.0003,
     "minimax/minimax-m3":               0.0003,
     "mistralai/mistral-small-2603":     0.0002,
     "moonshotai/kimi-k2.5":             0.0006,
     "moonshotai/kimi-k2.6":             0.00095,
     "moonshotai/kimi-k2.7":             0.00095,
+    "moonshotai/kimi-k3":               0.003,
     "qwen/qwen3-235b-a22b-2507":        0.0002266,
     "qwen/qwen3-30b-a3b-instruct-2507": 0.00015,
     "qwen/qwen3-coder":                 0.0009,
     "qwen/qwen3-coder-next":            0.0005,
     "qwen/qwen3-next-80b-a3b-instruct": 0.00015,
     "qwen/qwen3.5-flash-02-23":         0.00005,
-    "qwen/qwen3.6-35b-a3b":             0.00015,
+    "qwen/qwen3.6-35b-a3b":             0.000172,
     "qwen/qwen3.7-plus":                0.0004,
     "xiaomi/mimo-v2.5-pro":             0.001,
-    "z-ai/glm-5":                       0.0006,
-    "z-ai/glm-5.1":                     0.00105,
+    "z-ai/glm-5":                       0.001,
+    "z-ai/glm-5.1":                     0.0014,
     "z-ai/glm-5.2":                     0.0014
   },
   "output": {
     "claude-fable-5":                   0.05,
-    "claude-haiku-4-5":                 0.004,
+    "claude-haiku-4-5":                 0.005,
+    "claude-opus-4-0":                  0.075,
+    "claude-opus-4-1":                  0.075,
+    "claude-opus-4-5":                  0.025,
     "claude-opus-4-6":                  0.025,
     "claude-opus-4-7":                  0.025,
     "claude-opus-4-8":                  0.025,
+    "claude-opus-5":                    0.025,
     "claude-sonnet-4-5":                0.015,
     "claude-sonnet-4-6":                0.015,
-    "deepseek/deepseek-v4-flash":       0.00028,
-    "deepseek/deepseek-v4-pro":         0.00348,
+    "claude-sonnet-5":                  0.015,
+    "deepseek/deepseek-v4-flash":       0.0002791,
+    "deepseek/deepseek-v4-pro":         0.0026361,
     "gemini-2.0-flash":                 0.0004,
     "gemini-2.0-flash-lite":            0.0003,
     "gemini-2.5-flash":                 0.0012,
@@ -2144,6 +2188,9 @@ prices='{
     "gemini-3.1-flash-lite-preview":    0.0004,
     "gemini-3.1-pro-preview":           0.008,
     "gemini-3.5-flash":                 0.009,
+    "gemini-3.5-flash-lite":            0.0025,
+    "gemini-3.6-flash":                 0.0075,
+    "google/gemma-4-26b-a4b-it":        0.0006,
     "gpt-4.1":                          0.008,
     "gpt-4.1-mini":                     0.0016,
     "gpt-4.1-nano":                     0.0004,
@@ -2153,31 +2200,36 @@ prices='{
     "gpt-5-chat":                       0.01,
     "gpt-5-mini":                       0.002,
     "gpt-5-nano":                       0.0004,
-    "gpt-5.4":                          0.012,
-    "gpt-5.4-mini":                     0.0016,
-    "gpt-5.4-nano":                     0.0004,
-    "gpt-5.4-pro":                      0.08,
-    "gpt-5.5":                          0.04,
+    "gpt-5.4":                          0.015,
+    "gpt-5.4-mini":                     0.0045,
+    "gpt-5.4-nano":                     0.00125,
+    "gpt-5.4-pro":                      0.18,
+    "gpt-5.5":                          0.03,
     "gpt-5.5-mini":                     0.0025,
     "gpt-5.5-nano":                     0.0006,
-    "gpt-5.5-pro":                      0.12,
+    "gpt-5.5-pro":                      0.18,
+    "gpt-5.6-luna":                     0.006,
+    "gpt-5.6-sol":                      0.03,
+    "gpt-5.6-terra":                    0.015,
+    "grok-4.5":                         0.006,
     "minimax/minimax-m2.7":             0.0012,
     "minimax/minimax-m3":               0.0012,
     "mistralai/mistral-small-2603":     0.0006,
     "moonshotai/kimi-k2.5":             0.003,
     "moonshotai/kimi-k2.6":             0.004,
     "moonshotai/kimi-k2.7":             0.004,
+    "moonshotai/kimi-k3":               0.015,
     "qwen/qwen3-235b-a22b-2507":        0.0009064,
     "qwen/qwen3-30b-a3b-instruct-2507": 0.0006,
     "qwen/qwen3-coder":                 0.0027,
     "qwen/qwen3-coder-next":            0.0012,
     "qwen/qwen3-next-80b-a3b-instruct": 0.0012,
     "qwen/qwen3.5-flash-02-23":         0.00015,
-    "qwen/qwen3.6-35b-a3b":             0.00095,
+    "qwen/qwen3.6-35b-a3b":             0.0012002,
     "qwen/qwen3.7-plus":                0.0016,
     "xiaomi/mimo-v2.5-pro":             0.003,
-    "z-ai/glm-5":                       0.00208,
-    "z-ai/glm-5.1":                     0.0035,
+    "z-ai/glm-5":                       0.0032,
+    "z-ai/glm-5.1":                     0.0044,
     "z-ai/glm-5.2":                     0.0044
   }
 }'
@@ -2367,20 +2419,23 @@ elif [[ "$routed" == "failure" ]]; then
   # swap or compute savings against a non-model.
   printf '%s — %s%s' "$brand" "$routed" "$tokens_clause"
 elif [[ -n "$routed" ]]; then
-  # Show the savings clause only when the session is genuinely net-saving.
+  # Always show the savings clause, flooring a non-positive total at $0.00.
   # session_savings is "0.0000" on fresh sessions or sessions where every
-  # turn routed back to the selected model; it can also go negative when
-  # sticky routing forces a haiku-tagged side-call up to a cached
-  # sonnet/opus decision. In both cases the word "saved" would mislead,
-  # so drop the savings clause but keep the token totals.
-  has_savings="false"
-  if [[ -n "$session_savings" ]] \
-     && awk -v v="$session_savings" 'BEGIN{exit !(v+0 > 0.005)}'; then
-    has_savings="true"
-  fi
-  if [[ "$has_savings" == "true" ]]; then
+  # turn routed back to the selected model, and can go negative when routing
+  # upgrades a turn to a pricier model (e.g. a sticky/hard-pinned side-call,
+  # or a hard turn escalated to opus). "saved -$X" would mislead, so clamp
+  # the display to $0.00 rather than dropping the clause — a $0.00 readout
+  # tells the user the router ran and simply didn't beat their selection.
+  # When the CC selection is unknown ("?" or empty) there's nothing to
+  # compare against, so drop the "← selection" arrow and just show routed.
+  display_savings="$session_savings"
+  if [[ -z "$display_savings" ]] \
+     || awk -v v="$display_savings" 'BEGIN{exit !(v+0 < 0)}'; then
+    display_savings="0"
+  fi
+  if [[ -n "$selected_display" && "$selected_display" != "?" ]]; then
     printf '%s — %s ← %s · saved %s%s' \
-      "$brand" "$routed" "$selected_display" "$(fmt_money "$session_savings")" "$tokens_clause"
+      "$brand" "$routed" "$selected_display" "$(fmt_money "$display_savings")" "$tokens_clause"
   else
     printf '%s — %s%s' "$brand" "$routed" "$tokens_clause"
   fi
@@ -2417,14 +2472,19 @@ if [ -n "$user_name" ]; then
 fi
 custom_headers="$custom_headers"$'\n'"X-App: claude-code"
 
+# Setting ANTHROPIC_BASE_URL makes Claude Code treat us as non-first-party and
+# disable MCP tool-search deferral, inlining every tool schema into every request
+# — a large uncompactable prefix that can push a session into an autocompact
+# thrash loop. ENABLE_TOOL_SEARCH=auto restores on-demand loading (Claude Code's
+# own first-party default).
 if [ "$scope" = "project" ] && [ -z "$install_dir" ]; then
   jq -n --arg url "$base_url" --arg sl "$statusline_path_for_settings" '{
-    env: { ANTHROPIC_BASE_URL: $url },
+    env: { ANTHROPIC_BASE_URL: $url, ENABLE_TOOL_SEARCH: "auto" },
     statusLine: { type: "command", command: $sl }
   }' >"$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 },
+    env: { ANTHROPIC_BASE_URL: $url, ANTHROPIC_CUSTOM_HEADERS: $header, ENABLE_TOOL_SEARCH: "auto" },
     statusLine: { type: "command", command: $sl }
   }' >"$tmp_patch"
 fi

package/opencode-weave/README.md

@@ -3,7 +3,7 @@
 Lets a caller's own **AI subscriptions** pay for their **opencode** turns, routed
 through the Weave Router. A subscription is a **credential scoped to the model
 family it can pay for**, not a provider you pick: you connect your ChatGPT
-(Codex) and/or Claude (Pro/Max) plan once, the router routes every turn to the
+(Codex) and/or Claude (Pro/Max) plan once, the router routes every request to the
 best model, and bills the plan that matches the model it served — ChatGPT pays
 for GPT/Codex turns, Claude pays for Claude turns, your Weave key pays for
 everything else.
@@ -34,9 +34,9 @@ router's dedicated headers:
 | `X-Weave-Anthropic-Subscription: <sk-ant-oat token>` | pays Claude turns, refreshed on expiry |
 | `X-Weave-Router-Key: rk_…` | from `opencode.json` `options.headers` — the router authenticates off this |
 
-The router routes the turn across every model the caller's subs + key can pay
+The router routes each request across every model the caller's subs + key can pay
 for and resolves the subscription matching the chosen provider, so a sub is
-never billed for a turn outside its family.
+never billed for a request outside its family.
 
 ## Two storage slots, one request provider
 

package/package.json

@@ -1,9 +1,10 @@
 {
   "name": "@workweave/router",
-  "version": "0.2.4",
+  "version": "0.2.6",
   "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"
+    "weave-router": "bin.js",
+    "router": "bin.js"
   },
   "pi": {
     "extensions": [

package/pi-router/README.md

@@ -9,8 +9,8 @@ LLM proxy that picks the most cost-efficient model that still solves each task.
 Installed automatically by the Weave Router installer:
 
 ```bash
-WEAVE_ROUTER_KEY=rk_… npx -y @workweave/router --pi          # user scope
-WEAVE_ROUTER_KEY=rk_… npx -y @workweave/router --pi --local  # local router (http://localhost:8080)
+WEAVE_ROUTER_KEY=rk_… npx --package @workweave/router -y -- weave-router --pi
+WEAVE_ROUTER_KEY=rk_… npx --package @workweave/router -y -- weave-router --pi --local  # local router
 ```
 
 That writes `~/.pi/agent/models.json` (the `weave` provider), adds

package/uninstall.sh

@@ -213,6 +213,41 @@ if [ "$target" = "opencode" ]; then
     ok "Removed $opencode_parked"
   fi
 
+  # Remove slash command wrapper files this installer owns. Install mirrors
+  # this split: project scope uses <repo>/.opencode/commands/, while user
+  # scope—including a user-style --dir install—uses the global XDG commands
+  # directory so opencode discovers the wrappers from any working directory.
+  remove_opencode_command_dir() {
+    local opencode_cmds_dir="$1" cmd cmd_file
+    if [ -d "$opencode_cmds_dir" ]; then
+      refuse_if_symlink "$opencode_cmds_dir"
+      for cmd in force-model unforce-model router-feedback fm ufm rf; do
+        cmd_file="$opencode_cmds_dir/$cmd.md"
+        if [ -f "$cmd_file" ]; then
+          refuse_if_symlink "$cmd_file"
+          rm -f "$cmd_file"
+          ok "Removed $cmd_file"
+        fi
+      done
+      rmdir "$opencode_cmds_dir" 2>/dev/null || true
+    fi
+  }
+
+  if [ "$scope" = "project" ]; then
+    opencode_commands_dir="$opencode_dir/.opencode/commands"
+  else
+    opencode_commands_dir="${XDG_CONFIG_HOME:-$HOME/.config}/opencode/commands"
+  fi
+  remove_opencode_command_dir "$opencode_commands_dir"
+
+  # Also clean the two paths used by earlier installer revisions, but avoid
+  # visiting the canonical path twice. This keeps uninstall idempotent for
+  # existing project and --dir installations.
+  for legacy_dir in "$opencode_dir/commands" "$opencode_dir/.opencode/commands"; do
+    [ "$legacy_dir" = "$opencode_commands_dir" ] && continue
+    remove_opencode_command_dir "$legacy_dir"
+  done
+
   if [ -n "$install_dir" ]; then
     ok "Weave Router uninstalled from $install_dir (opencode)."
   else
@@ -475,13 +510,13 @@ else
 fi
 
 if [ -f "$settings_file" ]; then
-  # Only remove keys we actually installed: scrub our two env vars, and only
+  # Only remove keys we actually installed: scrub our env vars, and only
   # delete `statusLine` / `apiKeyHelper` when they point at scripts this
   # installer used in older versions. Otherwise an unrelated user-configured
   # statusLine or apiKeyHelper would be silently clobbered.
   cleaned="$(jq '
     if .env then
-      .env |= (del(.ANTHROPIC_BASE_URL, .ANTHROPIC_AUTH_TOKEN, .ANTHROPIC_CUSTOM_HEADERS))
+      .env |= (del(.ANTHROPIC_BASE_URL, .ANTHROPIC_AUTH_TOKEN, .ANTHROPIC_CUSTOM_HEADERS, .ENABLE_TOOL_SEARCH))
       | (if (.env | length) == 0 then del(.env) else . end)
     else . end
     | (if (.statusLine.command // "" | tostring | endswith("cc-statusline.sh"))
@@ -501,7 +536,7 @@ if [ -n "$local_settings_file" ] && [ -f "$local_settings_file" ]; then
   # uninstall fully reverts a toggled-off install.
   cleaned="$(jq '
     if .env then
-      .env |= (del(.ANTHROPIC_BASE_URL, .ANTHROPIC_AUTH_TOKEN, .ANTHROPIC_CUSTOM_HEADERS))
+      .env |= (del(.ANTHROPIC_BASE_URL, .ANTHROPIC_AUTH_TOKEN, .ANTHROPIC_CUSTOM_HEADERS, .ENABLE_TOOL_SEARCH))
       | (if (.env | length) == 0 then del(.env) else . end)
     else . end
     | del(.apiKeyHelper)
@@ -531,7 +566,7 @@ done
 commands_dir="$(dirname "$settings_file")/commands"
 if [ -d "$commands_dir" ]; then
   refuse_if_symlink "$commands_dir"
-  for cmd in force-model unforce-model router-feedback fm ufm rf router-off router-on router-status; do
+  for cmd in force-model unforce-model router-feedback fm ufm rf router-off router-on router-status router-session; do
     cmd_file="$commands_dir/$cmd.md"
     if [ -f "$cmd_file" ]; then
       refuse_if_symlink "$cmd_file"