virtual-context 0.3.2__tar.gz → 0.3.3__tar.gz

{virtual_context-0.3.2 → virtual_context-0.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: virtual-context
-Version: 0.3.2
+Version: 0.3.3
 Summary: OS-style virtual memory for LLM session context management
 Project-URL: Homepage, https://virtual-context.com
 Project-URL: Repository, https://github.com/virtual-context/virtual-context

{virtual_context-0.3.2 → virtual_context-0.3.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "virtual-context"
-version = "0.3.2"
+version = "0.3.3"
 description = "OS-style virtual memory for LLM session context management"
 readme = "README.md"
 license = "AGPL-3.0-or-later"
@@ -71,6 +71,7 @@ exclude = [
 artifacts = [
     "virtual_context/proxy/static/*",
     "virtual_context/data/anthropic-tokenizer/tokenizer.json",
+    "virtual_context/data/models.yaml",
 ]
 
 [tool.hatch.build.targets.wheel]

{virtual_context-0.3.2 → virtual_context-0.3.3}/virtual_context/__init__.py

@@ -14,7 +14,7 @@ from .types import (
     VirtualContextConfig,
 )
 
-__version__ = "0.3.2"
+__version__ = "0.3.3"
 
 __all__ = [
     "VirtualContextEngine",

{virtual_context-0.3.2 → virtual_context-0.3.3}/virtual_context/cli/main.py

@@ -508,6 +508,23 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
     "openrouter": ["qwen/qwen3-30b-a3b", "qwen/qwen3-32b", "google/gemini-2.5-flash-preview"],
 }
 
+# Context window sizes for upstream providers — used to set correct defaults
+_UPSTREAM_CONTEXT_WINDOWS: dict[str, int] = {
+    "anthropic": 1_000_000,
+    "openai": 128_000,
+    "gemini": 1_000_000,
+    "ollama": 128_000,
+    "openrouter": 200_000,
+    "custom": 128_000,
+}
+
+# API key env var names per provider
+_PROVIDER_KEY_ENVS: dict[str, str] = {
+    "anthropic": "ANTHROPIC_API_KEY",
+    "openai": "OPENAI_API_KEY",
+    "openrouter": "OPENROUTER_API_KEY",
+}
+
 
 def _prompt_tagging_provider() -> tuple[str, str]:
     provider = _prompt_choice(
@@ -542,31 +559,43 @@ def _provider_defaults(provider: str) -> tuple[str, str]:
 
 def _write_instance_config(
     base_dir: Path, label: str, provider: str, model: str, inbound_tagger_type: str,
+    *,
+    context_window: int = 200_000,
+    host: str = "127.0.0.1",
+    port: int = 5757,
+    dashboard_token: str = "",
+    tag_api_key_env: str = "",
 ) -> str:
     """Generate a standalone YAML config for one proxy instance.
 
-    Creates isolated storage at ``<base_dir>/.virtualcontext/<label>/store.db``
+    Creates isolated storage at ``~/.virtual-context/<label>/store.db``
     and writes config to ``<base_dir>/virtual-context-proxy-<label>.yaml``.
     Returns the path to the written file.
     """
-    storage_root = f".virtualcontext/{label}"
+    storage_root = str(Path.home() / ".virtual-context" / label)
     provider_label, base_url = _provider_defaults(provider)
     provider_block: dict = {}
     if provider == "ollama":
         provider_block = {provider_label: {"type": "generic_openai", "base_url": base_url + "/v1"}}
     elif provider == "anthropic":
-        provider_block = {provider_label: {"type": "anthropic"}}
+        prov = {"type": "anthropic"}
+        if tag_api_key_env:
+            prov["api_key_env"] = tag_api_key_env
+        provider_block = {provider_label: prov}
     elif provider == "openai":
-        provider_block = {provider_label: {"type": "generic_openai", "base_url": base_url}}
+        prov = {"type": "generic_openai", "base_url": base_url}
+        if tag_api_key_env:
+            prov["api_key_env"] = tag_api_key_env
+        provider_block = {provider_label: prov}
     elif provider == "openrouter":
-        provider_block = {provider_label: {"type": "openrouter", "base_url": base_url, "api_key_env": "OPENROUTER_API_KEY"}}
+        provider_block = {provider_label: {"type": "openrouter", "base_url": base_url, "api_key_env": tag_api_key_env or "OPENROUTER_API_KEY"}}
     else:
         provider_block = {provider_label: {"type": "generic_openai", "base_url": base_url}}
 
     cfg: dict = {
         "version": "0.2",
         "storage_root": storage_root,
-        "context_window": 120_000,
+        "context_window": context_window,
         "token_counter": "estimate",
         "tag_generator": {
             "type": "llm" if provider != "keyword" else "keyword",
@@ -596,6 +625,23 @@ def _write_instance_config(
     return str(out_path)
 
 
+def _check_provider_reachable(provider: str) -> bool:
+    """Quick check if a tagging provider is reachable."""
+    if provider == "ollama":
+        try:
+            import urllib.request
+            urllib.request.urlopen("http://127.0.0.1:11434/api/tags", timeout=3)
+            return True
+        except Exception:
+            return False
+    return True  # Remote providers — can't cheaply validate without a key
+
+
+def _generate_dashboard_token() -> str:
+    import secrets
+    return secrets.token_urlsafe(24)
+
+
 def _run_instance_wizard() -> tuple[list[dict], list[str]]:
     """Interactive wizard for proxy instance setup.
 
@@ -606,13 +652,47 @@ def _run_instance_wizard() -> tuple[list[dict], list[str]]:
     print("\n--- Tagging & Summarization ---")
     tag_provider, tag_model = _prompt_tagging_provider()
 
+    # Validate tagging provider is reachable
+    if not _check_provider_reachable(tag_provider):
+        print(f"\n  WARNING: {tag_provider} does not appear to be running.")
+        if tag_provider == "ollama":
+            print("  Start Ollama first: ollama serve")
+        if not _prompt_yes_no("Continue anyway?", default_yes=False):
+            sys.exit(0)
+
+    # API key for tagging provider
+    tag_api_key_env = ""
+    tag_key_env = _PROVIDER_KEY_ENVS.get(tag_provider)
+    if tag_key_env:
+        existing = os.environ.get(tag_key_env, "")
+        if existing:
+            print(f"\n  Found {tag_key_env} in environment.")
+            tag_api_key_env = tag_key_env
+        else:
+            print(f"\n  The {tag_provider} tagger needs an API key.")
+            print(f"  Set {tag_key_env} in your environment before starting the proxy.")
+            tag_api_key_env = tag_key_env
+
     inbound_tagger_type = _prompt_choice(
         "Inbound tagging mode:",
         ["embedding", "llm", "keyword"],
         default="embedding",
     )
 
-    multi = _prompt_yes_no("Configure multiple proxy instances?", default_yes=False)
+    # Client selection — determines post-wizard instructions
+    client = _prompt_choice(
+        "What client will you use with this proxy?",
+        ["Claude Code", "Claude Desktop", "Other / REST API"],
+        default="Claude Code",
+    )
+
+    # Dashboard token
+    print("\n--- Security ---")
+    dashboard_token = _generate_dashboard_token()
+    print(f"  Generated dashboard token: {dashboard_token}")
+    print("  (protects dashboard mutating endpoints)")
+
+    multi = _prompt_yes_no("\nConfigure multiple proxy instances?", default_yes=False)
     count = 1
     if multi:
         count = _prompt_int("How many proxy instances?", default=2, min_value=1)
@@ -620,7 +700,8 @@ def _run_instance_wizard() -> tuple[list[dict], list[str]]:
     instances: list[dict] = []
     config_paths: list[str] = []
     used_ports: set[int] = set()
-    base_dir = Path.cwd()
+    base_dir = Path.home() / ".virtual-context"
+    base_dir.mkdir(parents=True, exist_ok=True)
 
     for i in range(count):
         print(f"\n--- Instance {i + 1}/{count} ---")
@@ -634,6 +715,10 @@ def _run_instance_wizard() -> tuple[list[dict], list[str]]:
         label = _prompt("Label", default_label)
         upstream = _prompt("Upstream URL", default_upstream)
 
+        # Context window — default based on upstream provider
+        default_cw = _UPSTREAM_CONTEXT_WINDOWS.get(provider, 200_000)
+        context_window = _prompt_int("Context window (tokens)", default=default_cw, min_value=8_000)
+
         default_port = 5757 + i
         while True:
             port = _prompt_int("Port", default=default_port, min_value=1)
@@ -647,19 +732,26 @@ def _run_instance_wizard() -> tuple[list[dict], list[str]]:
 
         # Per-instance tagger override
         use_different_tagger = False
+        inst_tag_provider = tag_provider
+        inst_tag_model = tag_model
+        inst_tag_key_env = tag_api_key_env
         if count > 1:
             use_different_tagger = _prompt_yes_no(
                 f"Use a different tagger provider for '{label}'?", default_yes=False,
             )
 
-        inst_tag_provider = tag_provider
-        inst_tag_model = tag_model
         if use_different_tagger:
             inst_tag_provider, inst_tag_model = _prompt_tagging_provider()
+            inst_tag_key_env = _PROVIDER_KEY_ENVS.get(inst_tag_provider, "")
 
         # Write per-instance config
         cfg_path = _write_instance_config(
             base_dir, label, inst_tag_provider, inst_tag_model, inbound_tagger_type,
+            context_window=context_window,
+            host=host,
+            port=port,
+            dashboard_token=dashboard_token,
+            tag_api_key_env=inst_tag_key_env,
         )
         config_paths.append(cfg_path)
 
@@ -670,13 +762,16 @@ def _run_instance_wizard() -> tuple[list[dict], list[str]]:
                 "label": label,
                 "host": host,
                 "config": cfg_path,
+                "context_window": context_window,
+                "dashboard_token": dashboard_token,
             }
         )
 
     # Review screen
     print("\n--- Review ---")
     for inst in instances:
-        print(f"  [{inst['label']}] :{inst['port']} -> {inst['upstream']}")
+        print(f"  [{inst['label']}] {inst['host']}:{inst['port']} -> {inst['upstream']}")
+        print(f"    context window: {inst['context_window']:,} tokens")
         print(f"    config: {inst['config']}")
     print()
 
@@ -684,6 +779,48 @@ def _run_instance_wizard() -> tuple[list[dict], list[str]]:
         print("Aborted.")
         sys.exit(0)
 
+    # Post-wizard summary
+    print("\n" + "=" * 60)
+    print("SETUP COMPLETE")
+    print("=" * 60)
+    vc_home = Path.home() / ".virtual-context"
+    print(f"\n  Config files:  {vc_home}/")
+    print(f"  Storage:       {vc_home}/<label>/")
+    print(f"  Dashboard:     http://{instances[0]['host']}:{instances[0]['port']}/dashboard")
+
+    # API key reminders
+    needed_keys = set()
+    if tag_key_env and not os.environ.get(tag_key_env, ""):
+        needed_keys.add(tag_key_env)
+    if needed_keys:
+        print("\n  API keys needed (set before starting proxy):")
+        for k in sorted(needed_keys):
+            print(f"    export {k}=\"your-key-here\"")
+
+    # Client-specific instructions
+    inst = instances[0]
+    if "Claude Code" in client:
+        print(f"\n  Claude Code setup:")
+        print(f"    Set this in your environment or .claude/settings.json:")
+        print(f"    ANTHROPIC_BASE_URL=http://{inst['host']}:{inst['port']}/")
+        print(f"\n    Disable auto-compact in Claude Code:")
+        print(f"    Type /config -> Auto-compact -> disable")
+        print(f"\n    Set VC_DASHBOARD_TOKEN in your environment:")
+        print(f"    export VC_DASHBOARD_TOKEN=\"{dashboard_token}\"")
+    elif "Desktop" in client:
+        print(f"\n  Claude Desktop setup:")
+        print(f"    In Claude Desktop settings, set the API base URL to:")
+        print(f"    http://{inst['host']}:{inst['port']}/")
+    else:
+        print(f"\n  REST API / Custom client:")
+        print(f"    Proxy endpoint: http://{inst['host']}:{inst['port']}/")
+        print(f"    Forward your LLM API requests through this proxy.")
+
+    print(f"\n  Logs:")
+    print(f"    Proxy: tail -f ~/Library/Logs/virtual-context.log")
+    print(f"    Errors: tail -f ~/Library/Logs/virtual-context.err.log")
+    print()
+
     return instances, config_paths
 
 
@@ -703,7 +840,10 @@ def _proxy_command(config_path: Path, upstream: str | None) -> str:
     return cmd
 
 
-def _install_launchd_daemon(config_path: Path, upstream: str | None, start: bool) -> None:
+def _install_launchd_daemon(
+    config_path: Path, upstream: str | None, start: bool,
+    *, dashboard_token: str = "",
+) -> None:
     from xml.sax.saxutils import escape as _xml_escape
 
     launch_agents = Path.home() / "Library" / "LaunchAgents"
@@ -712,6 +852,17 @@ def _install_launchd_daemon(config_path: Path, upstream: str | None, start: bool
     cmd = _proxy_command(config_path, upstream)
     log_path = Path.home() / "Library" / "Logs" / "virtual-context.log"
     err_path = Path.home() / "Library" / "Logs" / "virtual-context.err.log"
+
+    # Build optional EnvironmentVariables block
+    env_block = ""
+    if dashboard_token:
+        env_block = f"""  <key>EnvironmentVariables</key>
+  <dict>
+    <key>VC_DASHBOARD_TOKEN</key>
+    <string>{_xml_escape(dashboard_token)}</string>
+  </dict>
+"""
+
     # Escape XML special characters to prevent XML injection
     plist = f"""<?xml version="1.0" encoding="UTF-8"?>
 <!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
@@ -733,7 +884,7 @@ def _install_launchd_daemon(config_path: Path, upstream: str | None, start: bool
   <string>{_xml_escape(str(log_path))}</string>
   <key>StandardErrorPath</key>
   <string>{_xml_escape(str(err_path))}</string>
-</dict>
+{env_block}</dict>
 </plist>
 """
     plist_path.write_text(plist)
@@ -829,7 +980,9 @@ def _install_windows_task_daemon(config_path: Path, upstream: str | None, start:
 
 
 def cmd_onboard(args):
-    config_path = Path(args.config) if args.config else Path.cwd() / "virtual-context.yaml"
+    vc_home = Path.home() / ".virtual-context"
+    vc_home.mkdir(parents=True, exist_ok=True)
+    config_path = Path(args.config) if args.config else vc_home / "virtual-context.yaml"
 
     if not config_path.exists():
         preset = get_preset(args.preset)
@@ -863,12 +1016,20 @@ def cmd_onboard(args):
     interactive = bool(args.wizard and sys.stdin.isatty())
     selected_upstream = args.upstream
 
+    _wizard_dashboard_token = ""
     if interactive:
         instances, _config_paths = _run_instance_wizard()
         _apply_proxy_instances(config_path, instances)
         print(f"Updated {config_path} with {len(instances)} proxy instance(s).")
-        # In multi-instance mode, upstream comes from config; daemon command should omit it.
-        selected_upstream = None
+        # In wizard mode, use the first instance's config for daemon install
+        if _config_paths:
+            config_path = Path(_config_paths[0])
+        # Extract upstream and dashboard token from first instance
+        if instances:
+            selected_upstream = instances[0].get("upstream")
+            _wizard_dashboard_token = instances[0].get("dashboard_token", "")
+        else:
+            selected_upstream = None
         if not daemon_requested:
             daemon_requested = _prompt_yes_no("Install daemon/service now?", default_yes=True)
 
@@ -884,7 +1045,10 @@ def cmd_onboard(args):
     system = platform.system().lower()
     try:
         if system == "darwin":
-            _install_launchd_daemon(config_path, selected_upstream, start=not args.no_start)
+            _install_launchd_daemon(
+                config_path, selected_upstream, start=not args.no_start,
+                dashboard_token=_wizard_dashboard_token,
+            )
         elif system == "linux":
             _install_systemd_user_daemon(config_path, selected_upstream, start=not args.no_start)
         elif system == "windows":

{virtual_context-0.3.2 → virtual_context-0.3.3}/virtual_context/core/model_catalog.py

@@ -30,11 +30,23 @@ class ModelCatalog:
 
     @classmethod
     def default(cls) -> ModelCatalog:
+        # Search order:
+        # 1. ~/.virtual-context/models.yaml (user override)
+        # 2. Bundled inside the package (virtual_context/data/models.yaml)
+        # 3. Project root (dev checkout)
+        user_path = os.path.join(os.path.expanduser("~"), ".virtual-context", "models.yaml")
+        if os.path.exists(user_path):
+            return cls(user_path)
+
         here = os.path.dirname(os.path.abspath(__file__))
-        # core/ -> virtual_context/ -> project root
+        pkg_path = os.path.join(os.path.dirname(here), "data", "models.yaml")
+        if os.path.exists(pkg_path):
+            return cls(pkg_path)
+
+        # Dev checkout: core/ -> virtual_context/ -> project root
         root = os.path.dirname(os.path.dirname(here))
-        path = os.path.join(root, "models.yaml")
-        return cls(path)
+        root_path = os.path.join(root, "models.yaml")
+        return cls(root_path)
 
     def _load(self, path: str) -> None:
         if not os.path.exists(path):

virtual_context-0.3.3/virtual_context/data/models.yaml

@@ -0,0 +1,207 @@
+# Model pricing catalog for Virtual Context telemetry.
+# Prices in USD per million tokens. Updated 2026-03-15.
+
+models:
+  # ---------------------------------------------------------------------------
+  # Anthropic
+  # ---------------------------------------------------------------------------
+  claude-haiku-4-5-20251001:
+    provider: anthropic
+    input_per_mtok: 1.00
+    output_per_mtok: 5.00
+    context_window: 200000
+    aliases: [haiku, claude-haiku, claude-haiku-4-5]
+
+  claude-sonnet-4-5-20250929:
+    provider: anthropic
+    input_per_mtok: 3.00
+    output_per_mtok: 15.00
+    context_window: 200000
+    aliases: [claude-sonnet-4-5, claude-sonnet-4-20250514]
+
+  claude-sonnet-4-6:
+    provider: anthropic
+    input_per_mtok: 3.00
+    output_per_mtok: 15.00
+    context_window: 200000
+    aliases: [sonnet, claude-sonnet]
+
+  claude-opus-4-5:
+    provider: anthropic
+    input_per_mtok: 5.00
+    output_per_mtok: 25.00
+    context_window: 200000
+    aliases: [claude-opus-4-20250514]
+
+  claude-opus-4-6:
+    provider: anthropic
+    input_per_mtok: 5.00
+    output_per_mtok: 25.00
+    context_window: 200000
+    aliases: [opus, claude-opus]
+
+  # ---------------------------------------------------------------------------
+  # OpenAI
+  # ---------------------------------------------------------------------------
+  gpt-5.2:
+    provider: openai
+    input_per_mtok: 1.75
+    output_per_mtok: 14.00
+    context_window: 128000
+    aliases: [gpt5.2]
+
+  gpt-5.2-pro:
+    provider: openai
+    input_per_mtok: 21.00
+    output_per_mtok: 168.00
+    context_window: 128000
+    aliases: [gpt5.2-pro]
+
+  gpt-5.3-codex:
+    provider: openai
+    input_per_mtok: 1.75
+    output_per_mtok: 14.00
+    context_window: 128000
+    aliases: [gpt5-codex, codex]
+
+  gpt-5.4:
+    provider: openai
+    input_per_mtok: 2.50
+    output_per_mtok: 15.00
+    context_window: 1050000
+    aliases: [gpt5.4]
+
+  gpt-5-mini:
+    provider: openai
+    input_per_mtok: 0.25
+    output_per_mtok: 2.00
+    context_window: 128000
+    aliases: [gpt5-mini]
+
+  gpt-5-nano:
+    provider: openai
+    input_per_mtok: 0.05
+    output_per_mtok: 0.40
+    context_window: 128000
+    aliases: [gpt5-nano]
+
+  gpt-4.1-nano:
+    provider: openai
+    input_per_mtok: 0.10
+    output_per_mtok: 0.40
+    context_window: 128000
+    aliases: [gpt4-nano]
+
+  gpt-4.1-mini:
+    provider: openai
+    input_per_mtok: 0.40
+    output_per_mtok: 1.60
+    context_window: 128000
+    aliases: [gpt4-mini]
+
+  gpt-4o:
+    provider: openai
+    input_per_mtok: 2.50
+    output_per_mtok: 10.00
+    context_window: 128000
+    aliases: [gpt4o]
+
+  gpt-4o-mini:
+    provider: openai
+    input_per_mtok: 0.15
+    output_per_mtok: 0.60
+    context_window: 128000
+    aliases: [gpt4o-mini]
+
+  gpt-4.1:
+    provider: openai
+    input_per_mtok: 2.00
+    output_per_mtok: 8.00
+    context_window: 1050000
+    aliases: [gpt4.1]
+
+  gpt-5:
+    provider: openai
+    input_per_mtok: 1.25
+    output_per_mtok: 10.00
+    context_window: 128000
+    aliases: [gpt5]
+
+  o1:
+    provider: openai
+    input_per_mtok: 15.00
+    output_per_mtok: 60.00
+    context_window: 200000
+    aliases: []
+
+  o3:
+    provider: openai
+    input_per_mtok: 2.00
+    output_per_mtok: 8.00
+    context_window: 200000
+    aliases: []
+
+  o4-mini:
+    provider: openai
+    input_per_mtok: 1.10
+    output_per_mtok: 4.40
+    context_window: 200000
+    aliases: []
+
+  # ---------------------------------------------------------------------------
+  # Google Gemini
+  # ---------------------------------------------------------------------------
+  gemini-2.5-pro:
+    provider: google
+    input_per_mtok: 1.25
+    output_per_mtok: 10.00
+    context_window: 1000000
+    aliases: [gemini-pro, gemini-2.5-pro-preview-03-25]
+
+  gemini-2.5-flash:
+    provider: google
+    input_per_mtok: 0.30
+    output_per_mtok: 2.50
+    context_window: 1000000
+    aliases: [gemini-flash, gemini-2.5-flash-preview-04-17]
+
+  gemini-3.1-pro-preview:
+    provider: google
+    input_per_mtok: 2.00
+    output_per_mtok: 12.00
+    context_window: 1000000
+    aliases: [gemini-3.1-pro]
+
+  gemini-3-pro-preview:
+    provider: google
+    input_per_mtok: 2.00
+    output_per_mtok: 12.00
+    context_window: 1000000
+    aliases: [gemini-3-pro]
+
+  gemini-2.0-flash:
+    provider: google
+    input_per_mtok: 0.10
+    output_per_mtok: 0.40
+    context_window: 1000000
+    aliases: [gemini-2.0-flash-001]
+
+  # ---------------------------------------------------------------------------
+  # Xiaomi MiMo
+  # ---------------------------------------------------------------------------
+  mimo-v2-flash:
+    provider: xiaomi
+    input_per_mtok: 0.09
+    output_per_mtok: 0.29
+    context_window: 262000
+    aliases: [mimo, mimo-flash]
+
+  # ---------------------------------------------------------------------------
+  # Local / Ollama (free)
+  # ---------------------------------------------------------------------------
+  qwen3:4b-instruct-2507-fp16:
+    provider: ollama
+    input_per_mtok: 0.00
+    output_per_mtok: 0.00
+    context_window: 32768
+    aliases: [qwen3-4b]