hermes-agent-kit 0.2.2__tar.gz

hermes_agent_kit-0.2.2/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 srmdn
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

hermes_agent_kit-0.2.2/PKG-INFO

@@ -0,0 +1,141 @@
+Metadata-Version: 2.4
+Name: hermes-agent-kit
+Version: 0.2.2
+Summary: Production hardening pack for Hermes Agent — per-topic model routing, fallback chains, rate limiting, and cost tracking via gateway hooks
+License: MIT
+Project-URL: Repository, https://github.com/srmdn/hermes-agent-kit
+Project-URL: Homepage, https://github.com/srmdn/hermes-agent-kit#readme
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: pyyaml>=6
+Dynamic: license-file
+
+# hermes-kit
+
+Production hardening pack for [Hermes Agent](https://github.com/NousResearch/hermes-agent).
+
+Self-hosted Hermes gateways are powerful but built for single-user setups. Multi-user deployments hit walls: no per-topic model routing, API failures surface as hard errors, one heavy user can burn your API budget with no alert.
+
+hermes-kit fills these gaps with production-grade hooks.
+
+> ⚠️ **How it works**: hermes-kit monkey-patches Hermes Agent's internal model resolver at runtime. This is intentionally fragile — Hermes Agent updates may break your setup. We're working on an upstream PR to replace the patch with native hook return values. Until then, test after every Hermes upgrade.
+
+## Prerequisites
+
+- Python ≥ 3.11
+- [Hermes Agent](https://github.com/NousResearch/hermes-agent) installed
+- A configured gateway (Telegram, Discord, etc.)
+
+## Install
+
+```bash
+pip install hermes-agent-kit
+```
+
+## Quickstart
+
+```bash
+# Install all hooks in one command
+hermes-kit install router fallback rate-limiter cost-tracker
+
+# Verify
+hermes-kit doctor
+
+# Start gateway with bridge auto-patched
+hermes-kit gateway run --accept-hooks
+
+# If new users get "I don't recognize you":
+GATEWAY_ALLOW_ALL_USERS=true hermes-kit gateway run --accept-hooks
+```
+
+Hooks land in `~/.hermes/hooks/<name>/`. Hermes discovers them on restart.
+
+## Modules
+
+### router — Per-Topic Model Routing
+
+Route Telegram topics to different AI models. Finance chat uses Qwen, coding chat uses DeepSeek, everything else falls back to GPT-4o-mini.
+
+**Via CLI:**
+```bash
+hermes-kit router set-default --model opencode-go/gpt-4o-mini
+hermes-kit router add 42 --model opencode-go/deepseek-v4-pro
+hermes-kit router show
+```
+
+**Via YAML** (`~/.hermes/hooks/router/topic_router.yaml`):
+```yaml
+default:
+  model: "opencode-go/gpt-4o-mini"
+
+topics:
+  "42":
+    model: "opencode-go/deepseek-v4-pro"
+```
+
+**Multi-provider** — route specific topics to native providers:
+```bash
+hermes-kit router add 42 --model gpt-4o --provider openai
+hermes-kit router add 7 --model claude-sonnet-4-20250514 --provider anthropic
+```
+
+Hermes resolves API keys from `~/.hermes/.env` (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, etc.). See [providers guide](docs/providers.md) for all supported providers and model IDs.
+
+### fallback — Automatic Fallback Chain
+
+Define a chain of models to try when the primary fails.
+
+**Via YAML** (`~/.hermes/hooks/fallback/fallback_chain.yaml`):
+```yaml
+chains:
+  global:
+    - "opencode-go/deepseek-v4-pro"     # primary
+    - "opencode-go/claude-sonnet-4"      # fallback
+    - "opencode-go/gpt-4o-mini"          # last resort
+```
+
+After a failure, call `hermes_kit.bridge.retry_with_fallback(session_key)` to advance to the next model.
+
+### rate-limiter — Per-User Rate Limiting
+
+Prevent a single user or chat from draining your API budget.
+
+**Via YAML** (`~/.hermes/hooks/rate-limiter/rate_limits.yaml`):
+```yaml
+limits:
+  global:
+    max_messages_per_window: 100
+    window_seconds: 3600
+  per_user:
+    "123456789":
+      max_messages_per_window: 50
+```
+
+> ⚠️ Rate limiter currently tracks usage but does not block messages. Enforcement is planned for an upcoming release.
+
+### cost-tracker — Real-Time Cost Tracking
+
+Track token costs per session and alert when thresholds are exceeded.
+
+**Via YAML** (`~/.hermes/hooks/cost-tracker/cost_tracker.yaml`):
+```yaml
+alert_threshold_usd: 1.0
+```
+
+Set to `0` to disable alerts but continue tracking.
+
+## Docs
+
+- [Quickstart](docs/quickstart.md) — agent-driven and manual install
+- [Providers](docs/providers.md) — supported AI providers and model lists
+- Manual setup per module:
+  - [Router](docs/manual/router.md) — per-topic model routing
+  - [Fallback](docs/manual/fallback.md) — automatic retry chains
+  - [Rate Limiter](docs/manual/rate-limiter.md) — per-user quotas
+  - [Cost Tracker](docs/manual/cost-tracker.md) — budget alerts
+- [Troubleshooting](docs/troubleshooting.md) — common issues
+
+## License
+
+MIT

hermes_agent_kit-0.2.2/README.md

@@ -0,0 +1,128 @@
+# hermes-kit
+
+Production hardening pack for [Hermes Agent](https://github.com/NousResearch/hermes-agent).
+
+Self-hosted Hermes gateways are powerful but built for single-user setups. Multi-user deployments hit walls: no per-topic model routing, API failures surface as hard errors, one heavy user can burn your API budget with no alert.
+
+hermes-kit fills these gaps with production-grade hooks.
+
+> ⚠️ **How it works**: hermes-kit monkey-patches Hermes Agent's internal model resolver at runtime. This is intentionally fragile — Hermes Agent updates may break your setup. We're working on an upstream PR to replace the patch with native hook return values. Until then, test after every Hermes upgrade.
+
+## Prerequisites
+
+- Python ≥ 3.11
+- [Hermes Agent](https://github.com/NousResearch/hermes-agent) installed
+- A configured gateway (Telegram, Discord, etc.)
+
+## Install
+
+```bash
+pip install hermes-agent-kit
+```
+
+## Quickstart
+
+```bash
+# Install all hooks in one command
+hermes-kit install router fallback rate-limiter cost-tracker
+
+# Verify
+hermes-kit doctor
+
+# Start gateway with bridge auto-patched
+hermes-kit gateway run --accept-hooks
+
+# If new users get "I don't recognize you":
+GATEWAY_ALLOW_ALL_USERS=true hermes-kit gateway run --accept-hooks
+```
+
+Hooks land in `~/.hermes/hooks/<name>/`. Hermes discovers them on restart.
+
+## Modules
+
+### router — Per-Topic Model Routing
+
+Route Telegram topics to different AI models. Finance chat uses Qwen, coding chat uses DeepSeek, everything else falls back to GPT-4o-mini.
+
+**Via CLI:**
+```bash
+hermes-kit router set-default --model opencode-go/gpt-4o-mini
+hermes-kit router add 42 --model opencode-go/deepseek-v4-pro
+hermes-kit router show
+```
+
+**Via YAML** (`~/.hermes/hooks/router/topic_router.yaml`):
+```yaml
+default:
+  model: "opencode-go/gpt-4o-mini"
+
+topics:
+  "42":
+    model: "opencode-go/deepseek-v4-pro"
+```
+
+**Multi-provider** — route specific topics to native providers:
+```bash
+hermes-kit router add 42 --model gpt-4o --provider openai
+hermes-kit router add 7 --model claude-sonnet-4-20250514 --provider anthropic
+```
+
+Hermes resolves API keys from `~/.hermes/.env` (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, etc.). See [providers guide](docs/providers.md) for all supported providers and model IDs.
+
+### fallback — Automatic Fallback Chain
+
+Define a chain of models to try when the primary fails.
+
+**Via YAML** (`~/.hermes/hooks/fallback/fallback_chain.yaml`):
+```yaml
+chains:
+  global:
+    - "opencode-go/deepseek-v4-pro"     # primary
+    - "opencode-go/claude-sonnet-4"      # fallback
+    - "opencode-go/gpt-4o-mini"          # last resort
+```
+
+After a failure, call `hermes_kit.bridge.retry_with_fallback(session_key)` to advance to the next model.
+
+### rate-limiter — Per-User Rate Limiting
+
+Prevent a single user or chat from draining your API budget.
+
+**Via YAML** (`~/.hermes/hooks/rate-limiter/rate_limits.yaml`):
+```yaml
+limits:
+  global:
+    max_messages_per_window: 100
+    window_seconds: 3600
+  per_user:
+    "123456789":
+      max_messages_per_window: 50
+```
+
+> ⚠️ Rate limiter currently tracks usage but does not block messages. Enforcement is planned for an upcoming release.
+
+### cost-tracker — Real-Time Cost Tracking
+
+Track token costs per session and alert when thresholds are exceeded.
+
+**Via YAML** (`~/.hermes/hooks/cost-tracker/cost_tracker.yaml`):
+```yaml
+alert_threshold_usd: 1.0
+```
+
+Set to `0` to disable alerts but continue tracking.
+
+## Docs
+
+- [Quickstart](docs/quickstart.md) — agent-driven and manual install
+- [Providers](docs/providers.md) — supported AI providers and model lists
+- Manual setup per module:
+  - [Router](docs/manual/router.md) — per-topic model routing
+  - [Fallback](docs/manual/fallback.md) — automatic retry chains
+  - [Rate Limiter](docs/manual/rate-limiter.md) — per-user quotas
+  - [Cost Tracker](docs/manual/cost-tracker.md) — budget alerts
+- [Troubleshooting](docs/troubleshooting.md) — common issues
+
+## License
+
+MIT

hermes_agent_kit-0.2.2/pyproject.toml

@@ -0,0 +1,32 @@
+[project]
+name = "hermes-agent-kit"
+version = "0.2.2"
+description = "Production hardening pack for Hermes Agent — per-topic model routing, fallback chains, rate limiting, and cost tracking via gateway hooks"
+readme = "README.md"
+license = {text = "MIT"}
+requires-python = ">=3.11"
+dependencies = ["pyyaml>=6"]
+
+[project.urls]
+Repository = "https://github.com/srmdn/hermes-agent-kit"
+Homepage = "https://github.com/srmdn/hermes-agent-kit#readme"
+
+[dependency-groups]
+dev = ["pytest>=8"]
+
+[project.scripts]
+hermes-kit = "hermes_kit.cli:main"
+
+[build-system]
+requires = ["setuptools>=75"]
+build-backend = "setuptools.build_meta"
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+addopts = "-v"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+"hermes_kit.hooks" = ["*.yaml", "**/*.yaml"]

hermes_agent_kit-0.2.2/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

hermes_agent_kit-0.2.2/src/hermes_agent_kit.egg-info/PKG-INFO

@@ -0,0 +1,141 @@
+Metadata-Version: 2.4
+Name: hermes-agent-kit
+Version: 0.2.2
+Summary: Production hardening pack for Hermes Agent — per-topic model routing, fallback chains, rate limiting, and cost tracking via gateway hooks
+License: MIT
+Project-URL: Repository, https://github.com/srmdn/hermes-agent-kit
+Project-URL: Homepage, https://github.com/srmdn/hermes-agent-kit#readme
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: pyyaml>=6
+Dynamic: license-file
+
+# hermes-kit
+
+Production hardening pack for [Hermes Agent](https://github.com/NousResearch/hermes-agent).
+
+Self-hosted Hermes gateways are powerful but built for single-user setups. Multi-user deployments hit walls: no per-topic model routing, API failures surface as hard errors, one heavy user can burn your API budget with no alert.
+
+hermes-kit fills these gaps with production-grade hooks.
+
+> ⚠️ **How it works**: hermes-kit monkey-patches Hermes Agent's internal model resolver at runtime. This is intentionally fragile — Hermes Agent updates may break your setup. We're working on an upstream PR to replace the patch with native hook return values. Until then, test after every Hermes upgrade.
+
+## Prerequisites
+
+- Python ≥ 3.11
+- [Hermes Agent](https://github.com/NousResearch/hermes-agent) installed
+- A configured gateway (Telegram, Discord, etc.)
+
+## Install
+
+```bash
+pip install hermes-agent-kit
+```
+
+## Quickstart
+
+```bash
+# Install all hooks in one command
+hermes-kit install router fallback rate-limiter cost-tracker
+
+# Verify
+hermes-kit doctor
+
+# Start gateway with bridge auto-patched
+hermes-kit gateway run --accept-hooks
+
+# If new users get "I don't recognize you":
+GATEWAY_ALLOW_ALL_USERS=true hermes-kit gateway run --accept-hooks
+```
+
+Hooks land in `~/.hermes/hooks/<name>/`. Hermes discovers them on restart.
+
+## Modules
+
+### router — Per-Topic Model Routing
+
+Route Telegram topics to different AI models. Finance chat uses Qwen, coding chat uses DeepSeek, everything else falls back to GPT-4o-mini.
+
+**Via CLI:**
+```bash
+hermes-kit router set-default --model opencode-go/gpt-4o-mini
+hermes-kit router add 42 --model opencode-go/deepseek-v4-pro
+hermes-kit router show
+```
+
+**Via YAML** (`~/.hermes/hooks/router/topic_router.yaml`):
+```yaml
+default:
+  model: "opencode-go/gpt-4o-mini"
+
+topics:
+  "42":
+    model: "opencode-go/deepseek-v4-pro"
+```
+
+**Multi-provider** — route specific topics to native providers:
+```bash
+hermes-kit router add 42 --model gpt-4o --provider openai
+hermes-kit router add 7 --model claude-sonnet-4-20250514 --provider anthropic
+```
+
+Hermes resolves API keys from `~/.hermes/.env` (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, etc.). See [providers guide](docs/providers.md) for all supported providers and model IDs.
+
+### fallback — Automatic Fallback Chain
+
+Define a chain of models to try when the primary fails.
+
+**Via YAML** (`~/.hermes/hooks/fallback/fallback_chain.yaml`):
+```yaml
+chains:
+  global:
+    - "opencode-go/deepseek-v4-pro"     # primary
+    - "opencode-go/claude-sonnet-4"      # fallback
+    - "opencode-go/gpt-4o-mini"          # last resort
+```
+
+After a failure, call `hermes_kit.bridge.retry_with_fallback(session_key)` to advance to the next model.
+
+### rate-limiter — Per-User Rate Limiting
+
+Prevent a single user or chat from draining your API budget.
+
+**Via YAML** (`~/.hermes/hooks/rate-limiter/rate_limits.yaml`):
+```yaml
+limits:
+  global:
+    max_messages_per_window: 100
+    window_seconds: 3600
+  per_user:
+    "123456789":
+      max_messages_per_window: 50
+```
+
+> ⚠️ Rate limiter currently tracks usage but does not block messages. Enforcement is planned for an upcoming release.
+
+### cost-tracker — Real-Time Cost Tracking
+
+Track token costs per session and alert when thresholds are exceeded.
+
+**Via YAML** (`~/.hermes/hooks/cost-tracker/cost_tracker.yaml`):
+```yaml
+alert_threshold_usd: 1.0
+```
+
+Set to `0` to disable alerts but continue tracking.
+
+## Docs
+
+- [Quickstart](docs/quickstart.md) — agent-driven and manual install
+- [Providers](docs/providers.md) — supported AI providers and model lists
+- Manual setup per module:
+  - [Router](docs/manual/router.md) — per-topic model routing
+  - [Fallback](docs/manual/fallback.md) — automatic retry chains
+  - [Rate Limiter](docs/manual/rate-limiter.md) — per-user quotas
+  - [Cost Tracker](docs/manual/cost-tracker.md) — budget alerts
+- [Troubleshooting](docs/troubleshooting.md) — common issues
+
+## License
+
+MIT

hermes_agent_kit-0.2.2/src/hermes_agent_kit.egg-info/SOURCES.txt

@@ -0,0 +1,32 @@
+LICENSE
+README.md
+pyproject.toml
+src/hermes_agent_kit.egg-info/PKG-INFO
+src/hermes_agent_kit.egg-info/SOURCES.txt
+src/hermes_agent_kit.egg-info/dependency_links.txt
+src/hermes_agent_kit.egg-info/entry_points.txt
+src/hermes_agent_kit.egg-info/requires.txt
+src/hermes_agent_kit.egg-info/top_level.txt
+src/hermes_kit/__init__.py
+src/hermes_kit/bridge.py
+src/hermes_kit/cli.py
+src/hermes_kit/hooks/cost_tracker/HOOK.yaml
+src/hermes_kit/hooks/cost_tracker/cost_tracker.yaml
+src/hermes_kit/hooks/cost_tracker/handler.py
+src/hermes_kit/hooks/fallback/HOOK.yaml
+src/hermes_kit/hooks/fallback/fallback_chain.yaml
+src/hermes_kit/hooks/fallback/handler.py
+src/hermes_kit/hooks/rate_limiter/HOOK.yaml
+src/hermes_kit/hooks/rate_limiter/handler.py
+src/hermes_kit/hooks/rate_limiter/rate_limits.yaml
+src/hermes_kit/hooks/router/HOOK.yaml
+src/hermes_kit/hooks/router/handler.py
+src/hermes_kit/hooks/router/topic_router.yaml
+src/hermes_kit/lib/__init__.py
+src/hermes_kit/lib/hermes_api.py
+tests/test_cli.py
+tests/test_cost_tracker.py
+tests/test_fallback.py
+tests/test_integration.py
+tests/test_rate_limiter.py
+tests/test_router.py

hermes_agent_kit-0.2.2/src/hermes_agent_kit.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

hermes_agent_kit-0.2.2/src/hermes_agent_kit.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+hermes-kit = hermes_kit.cli:main

hermes_agent_kit-0.2.2/src/hermes_agent_kit.egg-info/requires.txt

@@ -0,0 +1 @@
+pyyaml>=6

hermes_agent_kit-0.2.2/src/hermes_agent_kit.egg-info/top_level.txt

@@ -0,0 +1 @@
+hermes_kit

hermes_agent_kit-0.2.2/src/hermes_kit/__init__.py

@@ -0,0 +1 @@
+"""hermes-kit — Production hardening hooks for Hermes Agent."""

hermes_agent_kit-0.2.2/src/hermes_kit/bridge.py

@@ -0,0 +1,168 @@
+from typing import Optional
+import time
+
+_model_overrides: dict[str, dict[str, Optional[str]]] = {}
+_fallback_chains: dict[str, list[str]] = {}
+_fallback_index: dict[str, int] = {}
+
+_rate_counters: dict[str, int] = {}
+_rate_windows: dict[str, float] = {}
+_rate_limited: set[str] = set()
+
+
+def set_override(session_key: str, model: str, provider: Optional[str] = None) -> None:
+    _model_overrides[session_key] = {
+        "model": model,
+        "provider": provider,
+    }
+
+
+def get_override(session_key: str) -> Optional[dict]:
+    return _model_overrides.get(session_key)
+
+
+def clear_override(session_key: str) -> None:
+    _model_overrides.pop(session_key, None)
+
+
+def set_fallback_chain(session_key: str, chain: list[str]) -> None:
+    _fallback_chains[session_key] = chain
+    _fallback_index[session_key] = 0
+
+
+def get_fallback_chain(session_key: str) -> list[str] | None:
+    return _fallback_chains.get(session_key)
+
+
+def advance_fallback(session_key: str) -> None:
+    if session_key in _fallback_index:
+        _fallback_index[session_key] += 1
+
+
+def get_current_fallback(session_key: str) -> str | None:
+    chain = _fallback_chains.get(session_key)
+    if not chain:
+        return None
+    idx = _fallback_index.get(session_key, 0)
+    if idx < len(chain):
+        return chain[idx]
+    return None
+
+
+def retry_with_fallback(session_key: str) -> str | None:
+    advance_fallback(session_key)
+    model = get_current_fallback(session_key)
+    if model:
+        set_override(session_key, model=model)
+    return model
+
+
+def reset_rate_counter(session_key: str) -> None:
+    _rate_counters[session_key] = 0
+    _rate_windows[session_key] = time.time()
+    _rate_limited.discard(session_key)
+
+
+def increment_rate_counter(session_key: str) -> int:
+    _rate_counters[session_key] = _rate_counters.get(session_key, 0) + 1
+    if session_key not in _rate_windows:
+        _rate_windows[session_key] = time.time()
+    return _rate_counters[session_key]
+
+
+def get_rate_window_start(session_key: str) -> float:
+    return _rate_windows.get(session_key, 0.0)
+
+
+def set_rate_limited(session_key: str) -> None:
+    _rate_limited.add(session_key)
+
+
+def is_rate_limited(session_key: str) -> bool:
+    return session_key in _rate_limited
+
+
+_session_costs: dict[str, dict[str, float]] = {}
+_cost_pricing: dict[str, tuple[float, float]] = {
+    "gpt-4o": (2.50, 10.00),
+    "gpt-4o-mini": (0.15, 0.60),
+    "claude-sonnet-4": (3.00, 15.00),
+    "deepseek-chat": (0.14, 0.28),
+    "qwen-3.6-plus": (0.40, 0.80),
+}
+
+
+def track_cost(session_key: str, model: str, prompt_tokens: int, completion_tokens: int) -> None:
+    if session_key not in _session_costs:
+        _session_costs[session_key] = {}
+    if model not in _session_costs[session_key]:
+        _session_costs[session_key][model] = 0.0
+
+    input_price, output_price = _cost_pricing.get(model, (0.0, 0.0))
+    cost = (prompt_tokens / 1_000_000) * input_price + (completion_tokens / 1_000_000) * output_price
+    _session_costs[session_key][model] += cost
+
+
+def get_session_cost(session_key: str) -> float:
+    if session_key not in _session_costs:
+        return 0.0
+    return sum(_session_costs[session_key].values())
+
+
+def get_session_cost_breakdown(session_key: str) -> dict[str, float]:
+    return _session_costs.get(session_key, {})
+
+
+def reset_session_cost(session_key: str) -> None:
+    _session_costs.pop(session_key, None)
+
+
+def alert_cost_exceeded(session_key: str, total: float, threshold: float) -> None:
+    print(f"[hermes-kit] COST ALERT: session {session_key} total ${total:.4f} exceeds threshold ${threshold:.2f}")
+
+
+_RUNTIME_KEYS = ("provider", "api_key", "base_url", "api_mode")
+
+
+def _apply_override(override: dict, model: str, runtime_kwargs: dict) -> tuple[str, dict]:
+    model = override.get("model", model)
+    for key in _RUNTIME_KEYS:
+        val = override.get(key)
+        if val is not None:
+            runtime_kwargs[key] = val
+    return model, runtime_kwargs
+
+
+def patch_gateway_resolver() -> None:
+    import inspect
+    from gateway.run import GatewayRunner
+
+    original = GatewayRunner._resolve_session_agent_runtime
+    original_is_async = inspect.iscoroutinefunction(original)
+
+    if original_is_async:
+        async def patched_resolver(self, *args, **kwargs):
+            model, runtime_kwargs = await original(self, *args, **kwargs)
+            session_key = kwargs.get("session_key") or (args[0] if args else None)
+            if session_key:
+                override = get_override(session_key)
+                if override:
+                    model, runtime_kwargs = _apply_override(override, model, runtime_kwargs)
+            return model, runtime_kwargs
+    else:
+        def patched_resolver(self, *args, **kwargs):
+            model, runtime_kwargs = original(self, *args, **kwargs)
+            session_key = kwargs.get("session_key") or (args[0] if args else None)
+            if session_key:
+                override = get_override(session_key)
+                if override:
+                    model, runtime_kwargs = _apply_override(override, model, runtime_kwargs)
+            return model, runtime_kwargs
+
+    GatewayRunner._resolve_session_agent_runtime = patched_resolver
+
+
+try:
+    patch_gateway_resolver()
+except ImportError:
+    pass