vantageai-agent 0.1.0__tar.gz

vantageai_agent-0.1.0/PKG-INFO

@@ -0,0 +1,167 @@
+Metadata-Version: 2.4
+Name: vantageai-agent
+Version: 0.1.0
+Summary: AI coding agent with per-tool permissions, cost tracking, and token optimization
+Author-email: Aman Jain <aman@vantageaiops.com>
+License: MIT
+Project-URL: Homepage, https://vantageaiops.com
+Project-URL: Documentation, https://vantageaiops.com/docs
+Project-URL: Repository, https://github.com/amanjain/vantageai
+Project-URL: Bug Tracker, https://github.com/amanjain/vantageai/issues
+Keywords: ai,agent,cost-tracking,claude,codex,gemini,llm
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Classifier: Topic :: Utilities
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+Requires-Dist: anthropic>=0.40.0
+Requires-Dist: rich>=13.0.0
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Requires-Dist: pytest-asyncio>=0.21; extra == "dev"
+
+# Vantage Agent
+
+AI coding agent with per-tool permissions, cost tracking, prompt optimization, and anomaly detection. Works across Claude (API), Claude Code, Codex CLI, and Gemini CLI.
+
+## Install
+
+```bash
+pip install vantageai-agent
+```
+
+Requires Python 3.9+.
+
+## Quick Start
+
+### Interactive REPL
+
+```bash
+# Set your API key, then launch
+export ANTHROPIC_API_KEY=sk-...
+vantageai-agent
+```
+
+### One-shot prompt
+
+```bash
+vantageai-agent "explain the main function in cli.py"
+```
+
+### Pipe mode
+
+```bash
+echo "summarize this file" | vantageai-agent
+```
+
+## Backends
+
+Vantage auto-detects the best backend. You can override with `--backend`:
+
+| Backend | Description | Requires |
+|---------|-------------|---------|
+| `api` | Direct Anthropic API (exact token counts) | `ANTHROPIC_API_KEY` |
+| `claude` | Claude Code subprocess (free tier / subscription) | `claude` binary |
+| `codex` | OpenAI Codex CLI subprocess | `codex` binary |
+| `gemini` | Gemini CLI subprocess | `gemini` binary |
+
+```bash
+vantageai-agent --backend claude "refactor this function"
+vantageai-agent --backend codex "write unit tests for auth.py"
+```
+
+Auto-detect priority: `VANTAGE_BACKEND` env → `ANTHROPIC_API_KEY` → `claude` binary → `codex` binary → `gemini` binary.
+
+## Session Management
+
+Sessions are persisted to `~/.vantage/sessions/`. Resume any previous session:
+
+```bash
+# List sessions and costs
+vantageai-agent summary
+
+# Resume by session ID
+vantageai-agent --resume abc12345
+```
+
+## REPL Commands
+
+| Command | Description |
+|---------|-------------|
+| `/help` | Show all commands |
+| `/tools` | Show tool approval status |
+| `/allow Bash,Write` | Approve specific tools |
+| `/allow all` | Approve all tools for this session |
+| `/cost` | Show current session cost |
+| `/optimize on\|off` | Toggle prompt optimization |
+| `/model claude-opus-4-6` | Switch model mid-session |
+| `/reset` | Clear history, permissions, cost |
+| `/quit` | Exit (shows final cost summary) |
+
+## Cost Tracking
+
+Every turn shows token usage and cost. Exit shows a full session summary:
+
+```
+  +----- Cost Summary -----+
+  Model:             claude-sonnet-4-6
+  Input tokens:      1,234
+  Output tokens:     456
+  Cost:              $0.0123
+  Session total:     $0.0456
+  Prompts:           3
+  +-------------------------+
+```
+
+For Claude Code / Codex / Gemini (subscription or free tier), costs are labeled `~$0.00 (free tier)` or `~$0.0123 (estimated)`.
+
+## VantageAI Dashboard
+
+Send telemetry to your VantageAI dashboard for cross-session cost analysis:
+
+```bash
+export VANTAGE_API_KEY=vnt_...
+vantageai-agent
+```
+
+Or pass inline: `vantageai-agent --vantage-key vnt_...`
+
+## All Flags
+
+```
+vantageai-agent [OPTIONS] [PROMPT]
+
+  PROMPT              One-shot prompt (omit for interactive REPL)
+
+  --backend           api | claude | codex | gemini (auto-detected)
+  --model             Model ID (default: claude-sonnet-4-6)
+  --max-tokens        Max output tokens (default: 16384)
+  --resume SESSION_ID Resume a previous session
+  --api-key           Anthropic API key (or ANTHROPIC_API_KEY env)
+  --vantage-key       VantageAI dashboard API key (or VANTAGE_API_KEY env)
+  --system            Custom system prompt
+  --no-optimize       Disable prompt optimization
+  --cwd               Set working directory
+  --debug             Enable debug output
+```
+
+## Environment Variables
+
+| Variable | Description |
+|----------|-------------|
+| `ANTHROPIC_API_KEY` | Anthropic API key |
+| `VANTAGE_API_KEY` | VantageAI dashboard key |
+| `VANTAGE_BACKEND` | Force backend (api/claude/codex/gemini) |
+| `VANTAGE_MODEL` | Default model override |
+
+## License
+
+MIT

vantageai_agent-0.1.0/README.md

@@ -0,0 +1,137 @@
+# Vantage Agent
+
+AI coding agent with per-tool permissions, cost tracking, prompt optimization, and anomaly detection. Works across Claude (API), Claude Code, Codex CLI, and Gemini CLI.
+
+## Install
+
+```bash
+pip install vantageai-agent
+```
+
+Requires Python 3.9+.
+
+## Quick Start
+
+### Interactive REPL
+
+```bash
+# Set your API key, then launch
+export ANTHROPIC_API_KEY=sk-...
+vantageai-agent
+```
+
+### One-shot prompt
+
+```bash
+vantageai-agent "explain the main function in cli.py"
+```
+
+### Pipe mode
+
+```bash
+echo "summarize this file" | vantageai-agent
+```
+
+## Backends
+
+Vantage auto-detects the best backend. You can override with `--backend`:
+
+| Backend | Description | Requires |
+|---------|-------------|---------|
+| `api` | Direct Anthropic API (exact token counts) | `ANTHROPIC_API_KEY` |
+| `claude` | Claude Code subprocess (free tier / subscription) | `claude` binary |
+| `codex` | OpenAI Codex CLI subprocess | `codex` binary |
+| `gemini` | Gemini CLI subprocess | `gemini` binary |
+
+```bash
+vantageai-agent --backend claude "refactor this function"
+vantageai-agent --backend codex "write unit tests for auth.py"
+```
+
+Auto-detect priority: `VANTAGE_BACKEND` env → `ANTHROPIC_API_KEY` → `claude` binary → `codex` binary → `gemini` binary.
+
+## Session Management
+
+Sessions are persisted to `~/.vantage/sessions/`. Resume any previous session:
+
+```bash
+# List sessions and costs
+vantageai-agent summary
+
+# Resume by session ID
+vantageai-agent --resume abc12345
+```
+
+## REPL Commands
+
+| Command | Description |
+|---------|-------------|
+| `/help` | Show all commands |
+| `/tools` | Show tool approval status |
+| `/allow Bash,Write` | Approve specific tools |
+| `/allow all` | Approve all tools for this session |
+| `/cost` | Show current session cost |
+| `/optimize on\|off` | Toggle prompt optimization |
+| `/model claude-opus-4-6` | Switch model mid-session |
+| `/reset` | Clear history, permissions, cost |
+| `/quit` | Exit (shows final cost summary) |
+
+## Cost Tracking
+
+Every turn shows token usage and cost. Exit shows a full session summary:
+
+```
+  +----- Cost Summary -----+
+  Model:             claude-sonnet-4-6
+  Input tokens:      1,234
+  Output tokens:     456
+  Cost:              $0.0123
+  Session total:     $0.0456
+  Prompts:           3
+  +-------------------------+
+```
+
+For Claude Code / Codex / Gemini (subscription or free tier), costs are labeled `~$0.00 (free tier)` or `~$0.0123 (estimated)`.
+
+## VantageAI Dashboard
+
+Send telemetry to your VantageAI dashboard for cross-session cost analysis:
+
+```bash
+export VANTAGE_API_KEY=vnt_...
+vantageai-agent
+```
+
+Or pass inline: `vantageai-agent --vantage-key vnt_...`
+
+## All Flags
+
+```
+vantageai-agent [OPTIONS] [PROMPT]
+
+  PROMPT              One-shot prompt (omit for interactive REPL)
+
+  --backend           api | claude | codex | gemini (auto-detected)
+  --model             Model ID (default: claude-sonnet-4-6)
+  --max-tokens        Max output tokens (default: 16384)
+  --resume SESSION_ID Resume a previous session
+  --api-key           Anthropic API key (or ANTHROPIC_API_KEY env)
+  --vantage-key       VantageAI dashboard API key (or VANTAGE_API_KEY env)
+  --system            Custom system prompt
+  --no-optimize       Disable prompt optimization
+  --cwd               Set working directory
+  --debug             Enable debug output
+```
+
+## Environment Variables
+
+| Variable | Description |
+|----------|-------------|
+| `ANTHROPIC_API_KEY` | Anthropic API key |
+| `VANTAGE_API_KEY` | VantageAI dashboard key |
+| `VANTAGE_BACKEND` | Force backend (api/claude/codex/gemini) |
+| `VANTAGE_MODEL` | Default model override |
+
+## License
+
+MIT

vantageai_agent-0.1.0/pyproject.toml

@@ -0,0 +1,45 @@
+[build-system]
+requires = ["setuptools>=42", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "vantageai-agent"
+version = "0.1.0"
+description = "AI coding agent with per-tool permissions, cost tracking, and token optimization"
+readme = "README.md"
+license = { text = "MIT" }
+requires-python = ">=3.9"
+authors = [{ name = "Aman Jain", email = "aman@vantageaiops.com" }]
+keywords = ["ai", "agent", "cost-tracking", "claude", "codex", "gemini", "llm"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Environment :: Console",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.9",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Software Development :: Libraries :: Python Modules",
+    "Topic :: Utilities",
+]
+dependencies = [
+    "anthropic>=0.40.0",
+    "rich>=13.0.0",
+]
+
+[project.urls]
+Homepage = "https://vantageaiops.com"
+Documentation = "https://vantageaiops.com/docs"
+Repository = "https://github.com/amanjain/vantageai"
+"Bug Tracker" = "https://github.com/amanjain/vantageai/issues"
+
+[project.optional-dependencies]
+dev = ["pytest>=7.0", "pytest-asyncio>=0.21"]
+
+[project.scripts]
+vantageai-agent = "vantage_agent.cli:main"
+
+[tool.setuptools.packages.find]
+include = ["vantage_agent*"]

vantageai_agent-0.1.0/setup.cfg

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

vantageai_agent-0.1.0/tests/test_anomaly.py

@@ -0,0 +1,31 @@
+"""Tests for anomaly.py — cost anomaly detection."""
+import pytest
+from vantage_agent.anomaly import check_cost_anomaly
+
+
+class TestCheckCostAnomaly:
+    def test_no_anomaly_with_few_priors(self):
+        assert check_cost_anomaly(0.10, 0.05, 1) is False
+
+    def test_no_anomaly_zero_prior(self):
+        assert check_cost_anomaly(0.10, 0.0, 5) is False
+
+    def test_no_anomaly_normal_cost(self):
+        # avg = 0.10/5 = 0.02, current 0.03 < 0.06 (3x)
+        assert check_cost_anomaly(0.03, 0.10, 5) is False
+
+    def test_anomaly_detected(self):
+        # avg = 0.10/5 = 0.02, current 0.10 > 0.06 (3x)
+        assert check_cost_anomaly(0.10, 0.10, 5) is True
+
+    def test_no_anomaly_below_min_avg(self):
+        # avg = 0.001/5 = 0.0002 < MIN_AVG_COST
+        assert check_cost_anomaly(0.01, 0.001, 5) is False
+
+    def test_exactly_3x_not_anomaly(self):
+        # avg = 0.10/5 = 0.02, current 0.06 == 3x (not >)
+        assert check_cost_anomaly(0.06, 0.10, 5) is False
+
+    def test_just_over_3x_is_anomaly(self):
+        # avg = 0.10/5 = 0.02, current 0.0601 > 0.06
+        assert check_cost_anomaly(0.0601, 0.10, 5) is True

vantageai_agent-0.1.0/tests/test_backends.py

@@ -0,0 +1,89 @@
+"""Tests for Backend implementations, auto-detect logic, and ToolRegistry."""
+from __future__ import annotations
+
+import os
+from unittest.mock import patch
+
+import pytest
+
+from vantage_agent.backends import create_backend, auto_detect_backend
+from vantage_agent.backends.base import BackendCapabilities, BackendResult
+
+
+def test_auto_detect_api_key_returns_api():
+    with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "sk-test", "VANTAGE_BACKEND": ""}):
+        name = auto_detect_backend()
+    assert name == "api"
+
+
+def test_auto_detect_claude_binary_returns_claude():
+    with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "", "VANTAGE_BACKEND": ""}):
+        with patch("shutil.which", side_effect=lambda b: "/usr/bin/claude" if b == "claude" else None):
+            name = auto_detect_backend()
+    assert name == "claude"
+
+
+def test_auto_detect_no_binaries_raises_error():
+    with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "", "VANTAGE_BACKEND": ""}):
+        with patch("shutil.which", return_value=None):
+            with pytest.raises(RuntimeError, match="No backend found"):
+                auto_detect_backend()
+
+
+def test_explicit_env_var_overrides_autodetect():
+    with patch.dict(os.environ, {"VANTAGE_BACKEND": "gemini", "ANTHROPIC_API_KEY": "sk-test"}):
+        name = auto_detect_backend()
+    assert name == "gemini"
+
+
+def test_backend_capabilities_declared():
+    """Every backend must declare all 4 capability fields."""
+    for name in ("api", "claude", "codex", "gemini"):
+        backend = create_backend(name)
+        caps = backend.capabilities
+        assert isinstance(caps.supports_process, bool)
+        assert caps.token_count in ("exact", "estimated", "free_tier")
+        assert caps.tool_format in ("anthropic", "openai", "google")
+
+
+def test_api_backend_not_supports_process():
+    backend = create_backend("api")
+    assert backend.capabilities.supports_process is False
+    assert backend.capabilities.token_count == "exact"
+
+
+def test_cli_backends_support_process():
+    for name in ("claude", "codex", "gemini"):
+        backend = create_backend(name)
+        assert backend.capabilities.supports_process is True
+        assert backend.capabilities.token_count == "estimated"
+
+
+def test_unknown_backend_raises():
+    with pytest.raises(ValueError, match="Unknown backend"):
+        create_backend("llama")
+
+
+def test_tool_registry_translates_anthropic_to_openai():
+    from vantage_agent.tool_registry import ToolRegistry
+    tools = ToolRegistry.for_format("openai")
+    assert len(tools) > 0
+    for tool in tools:
+        assert "function" in tool, f"OpenAI format missing 'function' key: {tool}"
+        assert tool["type"] == "function"
+
+
+def test_tool_registry_translates_anthropic_to_google():
+    from vantage_agent.tool_registry import ToolRegistry
+    tools = ToolRegistry.for_format("google")
+    assert len(tools) > 0
+    for tool in tools:
+        assert "name" in tool
+        assert "description" in tool
+
+
+def test_tool_registry_anthropic_passthrough():
+    from vantage_agent.tool_registry import ToolRegistry
+    from vantage_agent.tools import TOOL_DEFINITIONS
+    tools = ToolRegistry.for_format("anthropic")
+    assert tools == list(TOOL_DEFINITIONS)

vantageai_agent-0.1.0/tests/test_bugs.py

@@ -0,0 +1,127 @@
+"""Regression tests for 5 pre-existing bugs."""
+from __future__ import annotations
+
+import hashlib
+from unittest.mock import patch
+
+import httpx
+import pytest
+
+from vantage_agent.optimizer import optimize_prompt
+from vantage_agent.tracker import Tracker, TrackerConfig
+
+
+# ---------------------------------------------------------------------------
+# Bug 1: Semantic corruption — "whether or not"
+# ---------------------------------------------------------------------------
+
+def test_whether_or_not_preserves_question_intent():
+    """'whether or not' must NOT be stripped — changes question to directive."""
+    original = "Tell me whether or not to delete the file"
+    result = optimize_prompt(original)
+    optimized = result.optimized if hasattr(result, "optimized") else str(result)
+    assert "whether" in optimized.lower(), (
+        f"'whether or not' was incorrectly stripped. Got: {optimized!r}"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Bug 2: anonymized privacy mode not implemented
+# ---------------------------------------------------------------------------
+
+def _make_tracker(privacy: str) -> Tracker:
+    cfg = TrackerConfig(api_key="test-key", privacy=privacy)
+    return Tracker(cfg)
+
+
+def test_anonymized_mode_strips_agent_name_and_hashes_event_id():
+    """anonymized mode must strip agent_name/team and hash event_id."""
+    tracker = _make_tracker("anonymized")
+    tracker.record(
+        model="claude-sonnet-4-6",
+        input_tokens=100,
+        output_tokens=50,
+        cost_usd=0.001,
+        latency_ms=200,
+        agent_name="my-team-agent",
+    )
+    with tracker._lock:
+        event = tracker._queue[0]
+
+    assert event.agent_name == "", (
+        f"agent_name not stripped in anonymized mode: {event.agent_name!r}"
+    )
+    assert event.team == "", (
+        f"team not stripped in anonymized mode: {event.team!r}"
+    )
+    assert len(event.event_id) == 64 and all(
+        c in "0123456789abcdef" for c in event.event_id
+    ), f"event_id not hashed in anonymized mode: {event.event_id!r}"
+
+
+# ---------------------------------------------------------------------------
+# Bug 3: Unknown agent defaults to "anthropic" provider
+# ---------------------------------------------------------------------------
+
+def test_unknown_agent_provider_defaults_to_unknown_not_anthropic():
+    """Unrecognised agent_name must map to 'unknown', not 'anthropic'."""
+    tracker = _make_tracker("full")
+    tracker.record(
+        model="gpt-4o",
+        input_tokens=100,
+        output_tokens=50,
+        cost_usd=0.002,
+        latency_ms=300,
+        agent_name="some-random-gpt-tool",
+    )
+    with tracker._lock:
+        event = tracker._queue[0]
+    assert event.provider == "unknown", (
+        f"Unknown agent mapped to provider {event.provider!r} instead of 'unknown'"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Bug 4: Pricing dictionaries inconsistent
+# ---------------------------------------------------------------------------
+
+def test_pricing_dictionaries_consistent():
+    """cost_tracker.py and pricing.py must use the same model key for claude-haiku-4-5."""
+    from vantage_agent.pricing import MODEL_PRICES
+    from vantage_agent.cost_tracker import MODEL_PRICING
+
+    haiku_key_in_pricing = any("haiku-4-5" in k for k in MODEL_PRICES)
+    haiku_key_in_tracker = any("haiku-4-5" in k for k in MODEL_PRICING)
+    assert haiku_key_in_pricing, "claude-haiku-4-5 not found in pricing.MODEL_PRICES"
+    assert haiku_key_in_tracker, "claude-haiku-4-5 not found in cost_tracker.MODEL_PRICING"
+
+    pricing_key = next(k for k in MODEL_PRICES if "haiku-4-5" in k)
+    tracker_key = next(k for k in MODEL_PRICING if "haiku-4-5" in k)
+    assert pricing_key == tracker_key, (
+        f"Key mismatch: pricing.py uses {pricing_key!r}, cost_tracker.py uses {tracker_key!r}"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Bug 5: Flush-before-success loses events
+# ---------------------------------------------------------------------------
+
+def test_flush_retains_events_on_network_error():
+    """Events must NOT be cleared from queue if the HTTP POST fails."""
+    tracker = _make_tracker("full")
+    tracker.record(
+        model="claude-sonnet-4-6",
+        input_tokens=100,
+        output_tokens=50,
+        cost_usd=0.001,
+        latency_ms=100,
+        agent_name="vantage-agent",
+    )
+    assert len(tracker._queue) == 1
+
+    with patch("httpx.post", side_effect=httpx.ConnectError("timeout")):
+        tracker.flush()
+
+    assert len(tracker._queue) == 1, (
+        "Events were lost from queue after a failed HTTP POST"
+    )