utim-cli 2.3.25__tar.gz → 2.3.26__tar.gz

{utim_cli-2.3.25 → utim_cli-2.3.26}/CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## [2.3.26] - 2026-09-08
+
+### 🛡️ Zero Rate-Limit Failover & Low-Latency Key Rotation
+- **Deterministic APINEX Failover Chain**: Upstream APINEX failures (429/402/exhaustion) seamlessly cascade through `openrouter/free` -> `upstage/solar-pro4` -> `qwen/qwen3.7-flash` -> `deepseek/deepseek-v4-flash-0731` with zero wait latency and direct key pool acquisition.
+- **Silent Rate-Limit Recovery**: Completely eliminated user-facing rate-limit interruptions, warnings, and 24-hour key lockout bans on transient provider 429s.
+- **Zero-Latency Paid 402/429 Key Rotation**: OpenRouter keys returning 402 or 429 immediately rotate across candidate keys and seamlessly fail over to backup chain models without aborting active turns.
+
 ## [2.3.25] - 2026-09-07
 
 ### 🛠️ Bug Fixes

{utim_cli-2.3.25/utim_cli.egg-info → utim_cli-2.3.26}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: utim-cli
-Version: 2.3.25
+Version: 2.3.26
 Summary: UTIM – Universal Terminal Intelligence Manager. A powerful agentic AI coding assistant for your terminal.
 License: Emend AI Proprietary EULA
 Project-URL: Homepage, https://utim.dev

{utim_cli-2.3.25 → utim_cli-2.3.26}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "utim-cli"
-version = "2.3.25"
+version = "2.3.26"
 
 description = "UTIM – Universal Terminal Intelligence Manager. A powerful agentic AI coding assistant for your terminal."
 readme = "README.md"

{utim_cli-2.3.25 → utim_cli-2.3.26}/utim/__init__.py

@@ -14,7 +14,7 @@ Quickstart:
 try:
     from utim_cli._version import VERSION as __version__
 except ImportError:
-    __version__ = "2.3.25"
+    __version__ = "2.3.26"
 
 # Developer-friendly core classes (similar to `import genai` / `import openai`)
 from utim_cli.sdk.session import UtimSession as Session

{utim_cli-2.3.25 → utim_cli-2.3.26}/utim_cli/__init__.py

@@ -5,7 +5,7 @@ try:
     from utim_cli._version import VERSION as __version__
 except ImportError:
     # Fallback during first-install / editable-install edge cases
-    __version__ = "2.3.25"
+    __version__ = "2.3.26"
 
 
 import builtins, os, pathlib, urllib.parse, re

{utim_cli-2.3.25 → utim_cli-2.3.26}/utim_cli/_version.py

@@ -11,6 +11,6 @@
 # from here so every surface stays in sync automatically.
 # ─────────────────────────────────────────────────────────────────────────────
 
-VERSION = "2.3.25"
+VERSION = "2.3.26"
 
 

{utim_cli-2.3.25 → utim_cli-2.3.26}/utim_cli/constants.py

@@ -11,7 +11,17 @@ SUBAGENT_DEFAULTS = {
     "analyze_image": "google/gemma-4-31b-it:free",
 }
 
-APINEX_MODEL_IDS = {
+ORCAROUTER_MODEL_IDS = {
+    "orcarouter/free",
+    "orcarouter/fusion",
+    "orcarouter/fusion-flash",
+    "orcarouter/fusion-mini",
+    "anthropic/claude-sonnet-5",
+    "anthropic/claude-opus-5",
+    "deepseek/deepseek-v4-flash-0731",
+    "deepseek/deepseek-v4-flash-free",
+    "deepseek/deepseek-v4-pro-0813",
+    "google/gemini-2.5-flash",
     "claude/opus-5",
     "claude/sonnet-5",
     "deepseek/v4-flash-0731",
@@ -33,6 +43,7 @@ APINEX_MODEL_IDS = {
     "qwen/3.8-max",
     "free/qwen-3.8-max",
 }
+APINEX_MODEL_IDS = ORCAROUTER_MODEL_IDS
 
 
 import sys

{utim_cli-2.3.25 → utim_cli-2.3.26}/utim_cli/orchestrator.py

@@ -854,24 +854,6 @@ class Orchestrator(LLMTransportMixin, TurnHistoryMixin, ToolExecutionMixin):
 
                     break  # success
                 except _ServerUnavailableError as exc:
-                    _err_text = str(exc).lower()
-                    if "rate limit" in _err_text or "rate-limit" in _err_text or "429" in _err_text or "too many requests" in _err_text:
-                        # Mark the current API key as exhausted for 24 hours
-                        from utim_cli.config import config
-                        current_key = config.get("api_key", "")
-                        if current_key:
-                            config.mark_api_key_exhausted(current_key, duration_hours=24)
-                            self.console.print(f"\n[bold yellow]⏳  Rate-limited on API key {current_key[:12]}... — excluded for 24 hours.[/bold yellow]")
-                        else:
-                            self.console.print(f"\n[bold yellow]⏳  Model is temporarily rate-limited by the provider.[/bold yellow]")
-                        self.console.print("  [dim]Please wait a moment before retrying, or switch models using [bold]/model[/bold] (Ctrl+M).[/dim]\n")
-                        if len(self.messages) <= turn_msg_start + 1:
-                            self.messages.append({
-                                "role": "assistant",
-                                "content": "⏳ **Rate Limited**: The API key has been temporarily exhausted (24h cooldown). Please wait and retry, or use `/model` to switch models."
-                            })
-                        self._persist_messages()
-                        return
                     if self.cancel_event.is_set():
                         del self.messages[turn_msg_start + 1:]
                         self._persist_messages()
@@ -918,17 +900,6 @@ class Orchestrator(LLMTransportMixin, TurnHistoryMixin, ToolExecutionMixin):
                         self.messages.append({"role": "assistant", "content": "[Aborted by user]"})
                     break
                 except Exception as exc:
-                    _err_text = str(exc).lower()
-                    if "rate limit" in _err_text or "rate-limit" in _err_text or "429" in _err_text or "too many requests" in _err_text:
-                        self.console.print(f"\n[bold yellow]⏳  Model is temporarily rate-limited by the provider.[/bold yellow]")
-                        self.console.print("  [dim]Please wait a moment before retrying, or switch models using [bold]/model[/bold] (Ctrl+M).[/dim]\n")
-                        if len(self.messages) <= turn_msg_start + 1:
-                            self.messages.append({
-                                "role": "assistant",
-                                "content": "⏳ **Model Rate Limited**: The model is temporarily rate-limited by the provider. Please wait a short moment and retry, or use `/model` to switch models."
-                            })
-                        self._persist_messages()
-                        return
                     if self.cancel_event.is_set():
                         if len(self.messages) <= turn_msg_start + 1:
                             self.messages.append({"role": "assistant", "content": "[Aborted by user]"})
@@ -1234,6 +1205,39 @@ class Orchestrator(LLMTransportMixin, TurnHistoryMixin, ToolExecutionMixin):
                 # Got a valid response — reset streak counter
                 _empty_response_streak = 0
 
+                # ── Nudge if model mistakenly claims it lacks web_search ──────────
+                _no_web_search_claims = [
+                    "don't have a web_search",
+                    "do not have a web_search",
+                    "don't have a web search",
+                    "do not have a web search",
+                    "no web search tool",
+                    "don't have access to a web search",
+                    "do not have access to a web search",
+                    "cannot search the web",
+                    "unable to search the web",
+                    "don't have internet access",
+                    "do not have internet access",
+                    "cannot access the internet",
+                    "cannot browse the web",
+                    "don't have browsing",
+                ]
+                if (
+                    not tool_calls
+                    and iteration < max_iterations - 1
+                    and any(claim in _lower_content for claim in _no_web_search_claims)
+                    and "web_search" not in (config.get("disabled_tools") or [])
+                ):
+                    self.messages.append({"role": "assistant", "content": content})
+                    self.messages.append({
+                        "role": "user",
+                        "content": (
+                            "SYSTEM REMINDER: You DO have the 'web_search' tool in your active tool registry. "
+                            "Call the web_search tool directly now using web_search(prompt=...) to perform the requested web search."
+                        )
+                    })
+                    continue
+
                 # ── Universal Required Deliverables Checklist ──────────────────────────────
                 if getattr(self, "_deliverable_check_attempts", 0) < 2:
                     _deliverable_matches = re.findall(

{utim_cli-2.3.25 → utim_cli-2.3.26}/utim_cli/orchestrator_helpers.py

@@ -350,11 +350,11 @@ If a workspace skill or rule is relevant to your task (or mentioned by the user)
 1. EXECUTE
 Take the shortest reliable path. Inspect → modify → verify. Avoid redundant exploration.
 
-2. SEARCH
-For known symbols/errors, use grep_search. For known files, use scoped read_file. Avoid large unsliced reads and redundant listings.
+2. SEARCH & WEB RESEARCH
+For known symbols/errors in the codebase, use grep_search. For known files, use scoped read_file. For live internet research, external library documentation, API references, latest package details, or web queries, use web_search. Avoid large unsliced reads and redundant listings.
 
 3. PLAN
-Use plan_project for new features, architectural designs, and complex multi-file refactors. Skip planning for localized edits and direct fixes.
+Use plan_project for new features, architectural designs, and complex multi-file refactors. Generated plans are saved under .utim_tmp/plans/ and can be read with read_file. Skip planning for localized edits and direct fixes.
 
 4. IMAGES
 If the user provides an image/screenshot path, read that exact path with read_file on Turn 1.
@@ -366,7 +366,7 @@ After modifications, verify with run_command using the appropriate build/test/li
 Use is_background=True or schedule for long-running commands, tests, or servers, then conclude the turn.
 
 7. TOOLS
-Use native function calls directly. Never emit raw tool JSON, <think> tags, or filler before tool calls.
+Use native function calls directly. You have full access to tools including read_file, edit_file, write_file, run_command, grep_search, list_directory, web_search, plan_project, schedule, and background_tasks. Never emit raw tool JSON, <think> tags, or filler before tool calls. Never claim you lack web search or terminal execution capabilities when these tools are present in your registry.
 
 8. OUTPUT
 After execution, give a concise result and relevant verification status.
@@ -460,11 +460,11 @@ If a workspace skill or rule is relevant to your task (or mentioned by the user)
 1. EXECUTE
 Take the shortest reliable path. Inspect → modify → verify. Avoid redundant exploration.
 
-2. SEARCH
-For known symbols/errors, use grep_search. For known files, use scoped read_file. Avoid large unsliced reads and redundant listings.
+2. SEARCH & WEB RESEARCH
+For known symbols/errors in the codebase, use grep_search. For known files, use scoped read_file. For live internet research, external library documentation, API references, latest package details, or web queries, use web_search. Avoid large unsliced reads and redundant listings.
 
 3. PLAN
-Use plan_project for new features, architectural designs, and complex multi-file refactors. Skip planning for localized edits and direct fixes.
+Use plan_project for new features, architectural designs, and complex multi-file refactors. Generated plans are saved under .utim_tmp/plans/ and can be read with read_file. Skip planning for localized edits and direct fixes.
 
 4. IMAGES
 If the user provides an image/screenshot path, read that exact path with read_file on Turn 1.
@@ -476,7 +476,7 @@ After modifications, verify with run_command using the appropriate build/test/li
 Use is_background=True or schedule for long-running commands, tests, or servers, then conclude the turn.
 
 7. TOOLS
-Use native function calls directly. Never emit raw tool JSON, <think> tags, or filler before tool calls.
+Use native function calls directly. You have full access to tools including read_file, edit_file, write_file, run_command, grep_search, list_directory, web_search, plan_project, schedule, and background_tasks. Never emit raw tool JSON, <think> tags, or filler before tool calls. Never claim you lack web search or terminal execution capabilities when these tools are present in your registry.
 
 8. OUTPUT
 After execution, give a concise result and relevant verification status.
@@ -1136,11 +1136,11 @@ def _get_dynamic_directives(is_compressed: bool, messages: Optional[List[Dict]]
         "1. EXECUTE",
         "Take the shortest reliable path. Inspect → modify → verify. Avoid redundant exploration.",
         "",
-        "2. SEARCH",
-        "For known symbols/errors, use grep_search. For known files, use scoped read_file. Avoid large unsliced reads and redundant listings.",
+        "2. SEARCH & WEB RESEARCH",
+        "For known symbols/errors in the codebase, use grep_search. For known files, use scoped read_file. For live internet research, external library documentation, API references, latest package details, or web queries, use web_search. Avoid large unsliced reads and redundant listings.",
         "",
         "3. PLAN",
-        "Use plan_project for new features, architectural designs, and complex multi-file refactors. Skip planning for localized edits and direct fixes.",
+        "Use plan_project for new features, architectural designs, and complex multi-file refactors. Generated plans are saved under .utim_tmp/plans/ and can be read with read_file. Skip planning for localized edits and direct fixes.",
         "",
         "4. IMAGES",
         "If the user provides an image/screenshot path, read that exact path with read_file on Turn 1.",
@@ -1152,7 +1152,7 @@ def _get_dynamic_directives(is_compressed: bool, messages: Optional[List[Dict]]
         "Use is_background=True or schedule for long-running commands, tests, or servers, then conclude the turn.",
         "",
         "7. TOOLS",
-        "Use native function calls directly. Never emit raw tool JSON, <think> tags, or filler before tool calls.",
+        "Use native function calls directly. You have full access to tools including read_file, edit_file, write_file, run_command, grep_search, list_directory, web_search, plan_project, schedule, and background_tasks. Never emit raw tool JSON, <think> tags, or filler before tool calls. Never claim you lack web search or terminal execution capabilities when these tools are present in your registry.",
         "",
         "8. OUTPUT",
         "After execution, give a concise result and relevant verification status.",

{utim_cli-2.3.25 → utim_cli-2.3.26}/utim_cli/orchestrator_llm_transport.py

@@ -827,8 +827,8 @@ class LLMTransportMixin:
                     is_custom = self._is_custom_model_selection(chosen_model, is_primary=not override_model)
                     is_free_user = (quota.get("plan") == "free" or str(quota.get("display_name", "")).lower() == "free")
 
-                    from utim_cli.constants import APINEX_MODEL_IDS
-                    is_apinex = (chosen_model in APINEX_MODEL_IDS)
+                    from utim_cli.constants import APINEX_MODEL_IDS, ORCAROUTER_MODEL_IDS
+                    is_apinex = (chosen_model in APINEX_MODEL_IDS or chosen_model in ORCAROUTER_MODEL_IDS or chosen_model.startswith("orcarouter/"))
                     is_free_model = (chosen_model.endswith(":free") or chosen_model.startswith("free/"))
 
                     if not is_custom:
@@ -1604,24 +1604,19 @@ class LLMTransportMixin:
                                                         # Insert paid slug as the very next model, then break
                                                         models_to_try[model_idx + 1:model_idx + 1] = [_paid_slug]
                                                     break  # inner attempt loop → outer loop picks up injected paid slug
-                                                # Non-free model: nothing to fall back to — show message and return
-                                                rate_limit_display = (
-                                                    f"⏳ **Model Rate Limited**: `{current_model}` is currently experiencing high demand and is temporarily rate-limited by the provider.\n\n"
-                                                    f"- **What to do:** Please wait a moment before retrying, or switch to a different model with `/model` or `Ctrl+M`.\n"
-                                                    f"- **Retry:** You can resend your prompt anytime."
-                                                )
-                                                if not silent:
-                                                    self.console.print(
-                                                        f"\n[bold yellow]⏳  Model '{current_model}' is temporarily rate-limited by the provider.[/bold yellow]"
-                                                    )
-                                                    self.console.print(
-                                                        "  [dim]Please wait a moment before retrying, or switch models using [bold]/model[/bold] (Ctrl+M).[/dim]\n"
-                                                    )
-                                                return {
-                                                    "role": "assistant",
-                                                    "content": rate_limit_display,
-                                                    "tool_calls": None,
-                                                }, False
+                                                # Never display rate limit warnings to the user.
+                                                # Seamlessly append the failover chain to models_to_try and continue trying models.
+                                                _failover_chain = [
+                                                    "openrouter/free",
+                                                    "upstage/solar-pro4",
+                                                    "qwen/qwen3.7-flash",
+                                                    "deepseek/deepseek-v4-flash-0731",
+                                                ]
+                                                _remaining = models_to_try[model_idx + 1:]
+                                                _to_inject = [m for m in _failover_chain if m not in _remaining and m != current_model]
+                                                if _to_inject:
+                                                    models_to_try.extend(_to_inject)
+                                                break  # inner attempt loop → outer loop picks up next failover model
                                             elif _is_provider_upstream:
                                                 raise RuntimeError(f"__PROVIDER_ERROR__:{current_model}")
                                             else:
@@ -2015,7 +2010,14 @@ class LLMTransportMixin:
                             self.console.print(f"\n[red]HTTP 400 Error Details from {current_model}: {escape(err_msg)}[/red]")
                         except Exception:
                             from rich.markup import escape
-                            self.console.print(f"\n[red]HTTP 400 Error Details from {current_model}: {escape(exc.response.text or 'Bad Request')}[/red]")
+                    if code == 404:
+                        _not_avail_msg = 'This model isn\'t currently available on UTIM in free plan please enter "/model" and select model'
+                        self.console.print(f"\n[bold yellow]{_not_avail_msg}[/bold yellow]\n")
+                        return {
+                            "role": "assistant",
+                            "content": _not_avail_msg,
+                            "tool_calls": None,
+                        }, False
                     if code == 403:
                         # Invalidate cached quota immediately to force fresh check
                         global _QUOTA_CACHE
@@ -2039,13 +2041,15 @@ class LLMTransportMixin:
                             break
                         # Bonus exhausted (or plan-gated): auto-degrade to the free
                         # default model instead of aborting the user's turn.
-                        from utim_cli.constants import APINEX_MODEL_IDS
+                        from utim_cli.constants import APINEX_MODEL_IDS, ORCAROUTER_MODEL_IDS
                         _free_fallback = DEFAULT_MODEL
                         _is_cur_free_or_apx = (
                             current_model == _free_fallback
                             or current_model.endswith(":free")
                             or current_model.startswith("free/")
+                            or current_model.startswith("orcarouter/")
                             or current_model in APINEX_MODEL_IDS
+                            or current_model in ORCAROUTER_MODEL_IDS
                         )
                         if current_model != _free_fallback and not _is_cur_free_or_apx:
                             self.console.print(f"\n[bold yellow]⚠ Model '{current_model}' needs bonus credits and your bonus balance is empty.[/bold yellow]")
@@ -2165,26 +2169,19 @@ class LLMTransportMixin:
                         if current_is_primary and fallback_list:
                             raise _ServerUnavailableError(f"Rate-limited on model '{current_model}'")
 
-                        # Non-free model with no fallback: show message and return
-                        rate_limit_display = (
-                            f"⏳ **Model Rate Limited**: `{current_model}` is currently experiencing high demand and is temporarily rate-limited by the provider.\n\n"
-                            f"- **What to do:** Please wait a moment before retrying, or switch to a different model with `/model` or `Ctrl+M`.\n"
-                            f"- **Retry:** You can resend your prompt anytime."
-                        )
-                        if not silent:
-                            self.console.print(
-                                f"\n[bold yellow]⏳  Model '{current_model}' is temporarily rate-limited by the provider.[/bold yellow]"
-                            )
-                            self.console.print(
-                                "  [dim]Please wait a moment before retrying, or switch models using [bold]/model[/bold] (Ctrl+M).[/dim]\n"
-                            )
-                            return {
-                                "role": "assistant",
-                                "content": rate_limit_display,
-                                "tool_calls": None,
-                            }, False
-                        # When silent=True (free model flow), raise exception to continue to next model
-                        raise _ServerUnavailableError(f"Rate-limited on model '{current_model}'")
+                        # Never display rate limit warnings to the user.
+                        # Seamlessly append the failover chain to models_to_try and continue trying models.
+                        _failover_chain = [
+                            "openrouter/free",
+                            "upstage/solar-pro4",
+                            "qwen/qwen3.7-flash",
+                            "deepseek/deepseek-v4-flash-0731",
+                        ]
+                        _remaining = models_to_try[model_idx + 1:]
+                        _to_inject = [m for m in _failover_chain if m not in _remaining and m != current_model]
+                        if _to_inject:
+                            models_to_try.extend(_to_inject)
+                        break  # outer loop will immediately pick up next failover model
                     if attempt < model_retries:
                         delay = 3 * (attempt + 1)
                         # Interruptible wait — so Ctrl+C can abort immediately

{utim_cli-2.3.25 → utim_cli-2.3.26}/utim_cli/orchestrator_tool_exec.py

@@ -803,9 +803,20 @@ class ToolExecutionMixin:
                     v_strip = v.strip()
                     v_norm = v_strip.replace("\\", "/")
 
+                    # 0. Prioritize workspace .utim_tmp and plans normalization so model paths like
+                    # /.utim/_tmp/plans, /.utim.tmp/plans, etc. resolve to workspace .utim_tmp, NOT ~/.utim
+                    import re as _re_path
+                    if _re_path.search(
+                        r'(?:^|[/\\])\.utim(?:[._/]tmp|[/\\]_tmp|[/\\]tmp|[._/]plans|[/\\]plans)($|[/\\])',
+                        v_norm,
+                        _re_path.IGNORECASE
+                    ):
+                        from utim_cli.tools import resolve_project_path
+                        arguments[k] = resolve_project_path(v_strip)
+                        continue
+
                     # 1. First prioritize dynamic .utim mapping for all users across any virtual prefix
                     # NOTE: use path-component regex to avoid matching .utim_tmp (which contains .utim as substring)
-                    import re as _re_path
                     if _re_path.search(r'(^|[/\\])\.utim($|[/\\])', v_norm):
                         from utim_cli.config import get_utim_dir
                         utim_dir = get_utim_dir()
@@ -1356,10 +1367,11 @@ class ToolExecutionMixin:
                     ("run_command", "input_text"):       "input",
                     ("run_command", "enter"):            "press_enter",
                     ("run_command", "target_process_id"):"process_id",
-                    ("run_command", "task_id"):          "process_id",
-                    # web_search: model sometimes uses 'prompt'
-                    ("web_search", "prompt"):       "query",
-                    ("web_search", "text"):         "query",
+                    # web_search: model sometimes uses 'query', 'search_query', 'q', or 'text' instead of 'prompt'
+                    ("web_search", "query"):        "prompt",
+                    ("web_search", "search_query"): "prompt",
+                    ("web_search", "q"):            "prompt",
+                    ("web_search", "text"):         "prompt",
                 }
                 # Apply aliasing before filtering (don't overwrite existing correct keys)
                 remapped = dict(call_args)