Merge pull request #177 from KeygraphHQ/feat/model-tiers

feat: add three-tier model system with Bedrock and Vertex AI support
2026-04-01 10:20:53 +02:00 · 2026-03-03 22:40:29 +05:30
parent 98e3446448 6a76df2f4c
commit e29d5b88a0
16 changed files with 347 additions and 193 deletions
--- a/.env.example
+++ b/.env.example
@@ -26,6 +26,45 @@ ANTHROPIC_API_KEY=your-api-key-here
 # OPENROUTER_API_KEY=sk-or-your-openrouter-key
 # ROUTER_DEFAULT=openrouter,google/gemini-3-flash-preview

+# =============================================================================
+# Model Tier Overrides (Anthropic API / OAuth / Bedrock)
+# =============================================================================
+# Override which model is used for each tier. Defaults are used if not set.
+# ANTHROPIC_SMALL_MODEL=...     # Small tier  (default: claude-haiku-4-5-20251001)
+# ANTHROPIC_MEDIUM_MODEL=...    # Medium tier (default: claude-sonnet-4-6)
+# ANTHROPIC_LARGE_MODEL=...     # Large tier  (default: claude-opus-4-6)
+
+# =============================================================================
+# OPTION 3: AWS Bedrock
+# =============================================================================
+# https://aws.amazon.com/blogs/machine-learning/accelerate-ai-development-with-amazon-bedrock-api-keys/
+# Requires the model tier overrides above to be set with Bedrock-specific model IDs.
+# Example Bedrock model IDs for us-east-1:
+#   ANTHROPIC_SMALL_MODEL=us.anthropic.claude-haiku-4-5-20251001-v1:0
+#   ANTHROPIC_MEDIUM_MODEL=us.anthropic.claude-sonnet-4-6
+#   ANTHROPIC_LARGE_MODEL=us.anthropic.claude-opus-4-6
+
+# CLAUDE_CODE_USE_BEDROCK=1
+# AWS_REGION=us-east-1
+# AWS_BEARER_TOKEN_BEDROCK=your-bearer-token
+
+# =============================================================================
+# OPTION 4: Google Vertex AI
+# =============================================================================
+# https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/use-partner-models
+# Requires a GCP service account with roles/aiplatform.user.
+# Download the SA key JSON from GCP Console (IAM > Service Accounts > Keys).
+# Requires the model tier overrides above to be set with Vertex AI model IDs.
+# Example Vertex AI model IDs:
+#   ANTHROPIC_SMALL_MODEL=claude-haiku-4-5@20251001
+#   ANTHROPIC_MEDIUM_MODEL=claude-sonnet-4-6
+#   ANTHROPIC_LARGE_MODEL=claude-opus-4-6
+
+# CLAUDE_CODE_USE_VERTEX=1
+# CLOUD_ML_REGION=us-east5
+# ANTHROPIC_VERTEX_PROJECT_ID=your-gcp-project-id
+# GOOGLE_APPLICATION_CREDENTIALS=./credentials/gcp-sa-key.json
+
 # =============================================================================
 # Available Models
 # =============================================================================
--- a/.gitignore
+++ b/.gitignore
@@ -1,5 +1,6 @@
 node_modules/
 .env
 audit-logs/
+credentials/
 dist/
 repos/
--- a/README.md
+++ b/README.md
@@ -87,6 +87,8 @@ Shannon is available in two editions:
  - [Usage Examples](#usage-examples)
  - [Workspaces and Resuming](#workspaces-and-resuming)
  - [Configuration (Optional)](#configuration-optional)
+  - [AWS Bedrock](#aws-bedrock)
+  - [Google Vertex AI](#google-vertex-ai)
  - [[EXPERIMENTAL - UNSUPPORTED] Router Mode (Alternative Providers)](#experimental---unsupported-router-mode-alternative-providers)
  - [Output and Results](#output-and-results)
 - [Sample Reports](#-sample-reports)
@@ -107,6 +109,8 @@ Shannon is available in two editions:
 - **AI Provider Credentials** (choose one):
  - **Anthropic API key** (recommended) - Get from [Anthropic Console](https://console.anthropic.com)
  - **Claude Code OAuth token**
+  - **AWS Bedrock** - Route through Amazon Bedrock with AWS credentials (see [AWS Bedrock](#aws-bedrock))
+  - **Google Vertex AI** - Route through Google Cloud Vertex AI (see [Google Vertex AI](#google-vertex-ai))
  - **[EXPERIMENTAL - UNSUPPORTED] Alternative providers via Router Mode** - OpenAI or Google Gemini via OpenRouter (see [Router Mode](#experimental---unsupported-router-mode-alternative-providers))

 ### Quick Start
@@ -348,6 +352,70 @@ pipeline:

 `max_concurrent_pipelines` controls how many vulnerability pipelines run simultaneously (1-5, default: 5). Lower values reduce the chance of hitting rate limits but increase wall-clock time.

+### AWS Bedrock
+
+Shannon also supports [Amazon Bedrock](https://aws.amazon.com/bedrock/) instead of using an Anthropic API key.
+
+#### Quick Setup
+
+1. Add your AWS credentials to `.env`:
+
+```bash
+CLAUDE_CODE_USE_BEDROCK=1
+AWS_REGION=us-east-1
+AWS_BEARER_TOKEN_BEDROCK=your-bearer-token
+
+# Set models with Bedrock-specific IDs for your region
+ANTHROPIC_SMALL_MODEL=us.anthropic.claude-haiku-4-5-20251001-v1:0
+ANTHROPIC_MEDIUM_MODEL=us.anthropic.claude-sonnet-4-6
+ANTHROPIC_LARGE_MODEL=us.anthropic.claude-opus-4-6
+```
+
+2. Run Shannon as usual:
+
+```bash
+./shannon start URL=https://example.com REPO=repo-name
+```
+
+Shannon uses three model tiers: **small** (`claude-haiku-4-5-20251001`) for summarization, **medium** (`claude-sonnet-4-6`) for security analysis, and **large** (`claude-opus-4-6`) for deep reasoning. Set `ANTHROPIC_SMALL_MODEL`, `ANTHROPIC_MEDIUM_MODEL`, and `ANTHROPIC_LARGE_MODEL` to the Bedrock model IDs for your region.
+
+### Google Vertex AI
+
+Shannon also supports [Google Vertex AI](https://cloud.google.com/vertex-ai) instead of using an Anthropic API key.
+
+#### Quick Setup
+
+1. Create a service account with the `roles/aiplatform.user` role in the [GCP Console](https://console.cloud.google.com/iam-admin/serviceaccounts), then download a JSON key file.
+
+2. Place the key file in the `./credentials/` directory:
+
+```bash
+mkdir -p ./credentials
+cp /path/to/your-sa-key.json ./credentials/gcp-sa-key.json
+```
+
+3. Add your GCP configuration to `.env`:
+
+```bash
+CLAUDE_CODE_USE_VERTEX=1
+CLOUD_ML_REGION=us-east5
+ANTHROPIC_VERTEX_PROJECT_ID=your-gcp-project-id
+GOOGLE_APPLICATION_CREDENTIALS=./credentials/gcp-sa-key.json
+
+# Set models with Vertex AI model IDs
+ANTHROPIC_SMALL_MODEL=claude-haiku-4-5@20251001
+ANTHROPIC_MEDIUM_MODEL=claude-sonnet-4-6
+ANTHROPIC_LARGE_MODEL=claude-opus-4-6
+```
+
+4. Run Shannon as usual:
+
+```bash
+./shannon start URL=https://example.com REPO=repo-name
+```
+
+Set `CLOUD_ML_REGION=global` for global endpoints, or a specific region like `us-east5`. Some models may not be available on global endpoints — see the [Vertex AI Model Garden](https://console.cloud.google.com/vertex-ai/model-garden) for region availability.
+
 ### [EXPERIMENTAL - UNSUPPORTED] Router Mode (Alternative Providers)

 Shannon can experimentally route requests through alternative AI providers using claude-code-router. This mode is not officially supported and is intended primarily for:
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -24,6 +24,16 @@ services:
      - ANTHROPIC_AUTH_TOKEN=${ANTHROPIC_AUTH_TOKEN:-}  # Auth token for router
      - ROUTER_DEFAULT=${ROUTER_DEFAULT:-}  # Model name when using router (e.g., "gemini,gemini-2.5-pro")
      - CLAUDE_CODE_OAUTH_TOKEN=${CLAUDE_CODE_OAUTH_TOKEN:-}
+      - CLAUDE_CODE_USE_BEDROCK=${CLAUDE_CODE_USE_BEDROCK:-}
+      - AWS_REGION=${AWS_REGION:-}
+      - AWS_BEARER_TOKEN_BEDROCK=${AWS_BEARER_TOKEN_BEDROCK:-}
+      - CLAUDE_CODE_USE_VERTEX=${CLAUDE_CODE_USE_VERTEX:-}
+      - CLOUD_ML_REGION=${CLOUD_ML_REGION:-}
+      - ANTHROPIC_VERTEX_PROJECT_ID=${ANTHROPIC_VERTEX_PROJECT_ID:-}
+      - GOOGLE_APPLICATION_CREDENTIALS=${GOOGLE_APPLICATION_CREDENTIALS:-}
+      - ANTHROPIC_SMALL_MODEL=${ANTHROPIC_SMALL_MODEL:-}
+      - ANTHROPIC_MEDIUM_MODEL=${ANTHROPIC_MEDIUM_MODEL:-}
+      - ANTHROPIC_LARGE_MODEL=${ANTHROPIC_LARGE_MODEL:-}
      - CLAUDE_CODE_MAX_OUTPUT_TOKENS=${CLAUDE_CODE_MAX_OUTPUT_TOKENS:-64000}
    depends_on:
      temporal:
@@ -33,6 +43,7 @@ services:
      - ./prompts:/app/prompts
      - ./audit-logs:/app/audit-logs
      - ${OUTPUT_DIR:-./audit-logs}:/app/output
+      - ./credentials:/app/credentials:ro
      - ./repos:/repos
      - ${BENCHMARKS_BASE:-.}:/benchmarks
    shm_size: 2gb
--- a/prompts/pipeline-testing/vuln-auth.txt
+++ b/prompts/pipeline-testing/vuln-auth.txt
@@ -1,37 +1,13 @@
-## 🧪 Pipeline Testing: MCP Isolation Test for Auth Agent
+Please complete these tasks using your MCP tools:

-**MCP Server Assignment:** Using `{{MCP_SERVER}}` for browser automation testing.
+1. Navigate to https://example.net and take a screenshot:
+   - Use {{MCP_SERVER}}__browser_navigate to go to https://example.net
+   - Use {{MCP_SERVER}}__browser_take_screenshot

-### Step 1: Test Browser Isolation
-1. Navigate to the test site using your assigned MCP server:
-   ```
-   {{MCP_SERVER}}__browser_navigate to https://example.net
-   ```
+2. Save an analysis deliverable:
+   - Use save_deliverable with deliverable_type "AUTH_ANALYSIS"
+   - Content: "# Auth Analysis Report\n\nAnalysis complete. No authentication vulnerabilities identified."

-2. Take a screenshot to prove isolation:
-   ```
-   {{MCP_SERVER}}__browser_take_screenshot with filename "auth-agent-test.png"
-   ```
-
-3. Log the test results:
-   ```
-   console.log("🔵 AUTH AGENT - MCP Server: {{MCP_SERVER}} - Successfully navigated to example.net")
-   ```
-
-### Step 2: Create Required Deliverables
-Use the save_deliverable MCP tool to create both deliverables:
-
-**1. Analysis Report:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "AUTH_ANALYSIS"`
- `content: "# Auth Analysis Complete (Pipeline Test)\n\n**MCP Server Used:** {{MCP_SERVER}}\n**Test Site:** https://example.net\n**Screenshot:** auth-agent-test.png\n**Status:** Browser isolation test successful"`
-
-**2. Exploitation Queue:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "AUTH_QUEUE"`
- `content: {"vulnerabilities": [{"ID": "AUTH-VULN-01", "verdict": "vulnerable", "confidence": "high", "notes": "Test auth vulnerability for pipeline validation (MCP: {{MCP_SERVER}})", "vulnerability_type": "Authentication_Bypass", "endpoint": "/api/login", "flaw_type": "Missing token validation", "exploit_hypothesis": "Empty or invalid token allows access to protected resources", "witness_payload": "Authorization: Bearer invalid_token"}]}`
-
-These tools will automatically create the correct files in `deliverables/`.
-
-### Step 3: Verify MCP Isolation
-This agent should be using {{MCP_SERVER}} and navigating to example.net independently of other parallel agents.
+3. Save a queue deliverable:
+   - Use save_deliverable with deliverable_type "AUTH_QUEUE"
+   - Content: {"vulnerabilities": []}
--- a/prompts/pipeline-testing/vuln-authz.txt
+++ b/prompts/pipeline-testing/vuln-authz.txt
@@ -1,37 +1,13 @@
-## 🧪 Pipeline Testing: MCP Isolation Test for Authorization Agent
+Please complete these tasks using your MCP tools:

-**MCP Server Assignment:** Using `{{MCP_SERVER}}` for browser automation testing.
+1. Navigate to https://jsonplaceholder.typicode.com and take a screenshot:
+   - Use {{MCP_SERVER}}__browser_navigate to go to https://jsonplaceholder.typicode.com
+   - Use {{MCP_SERVER}}__browser_take_screenshot

-### Step 1: Test Browser Isolation
-1. Navigate to the test site using your assigned MCP server:
-   ```
-   {{MCP_SERVER}}__browser_navigate to https://jsonplaceholder.typicode.com
-   ```
+2. Save an analysis deliverable:
+   - Use save_deliverable with deliverable_type "AUTHZ_ANALYSIS"
+   - Content: "# Authorization Analysis Report\n\nAnalysis complete. No authorization vulnerabilities identified."

-2. Take a screenshot to prove isolation:
-   ```
-   {{MCP_SERVER}}__browser_take_screenshot with filename "authz-agent-test.png"
-   ```
-
-3. Log the test results:
-   ```
-   console.log("🟢 AUTHZ AGENT - MCP Server: {{MCP_SERVER}} - Successfully navigated to jsonplaceholder.typicode.com")
-   ```
-
-### Step 2: Create Required Deliverables
-Use the save_deliverable MCP tool to create both deliverables:
-
-**1. Analysis Report:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "AUTHZ_ANALYSIS"`
- `content: "# Authorization Analysis Complete (Pipeline Test)\n\n**MCP Server Used:** {{MCP_SERVER}}\n**Test Site:** https://jsonplaceholder.typicode.com\n**Screenshot:** authz-agent-test.png\n**Status:** Browser isolation test successful"`
-
-**2. Exploitation Queue:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "AUTHZ_QUEUE"`
- `content: {"vulnerabilities": [{"ID": "AUTHZ-VULN-01", "verdict": "vulnerable", "confidence": "high", "notes": "Test authz vulnerability for pipeline validation (MCP: {{MCP_SERVER}})", "vulnerability_type": "Vertical", "endpoint": "/admin/users", "actual_access": "Regular users can access admin functions", "witness_payload": "GET /admin/users with regular user token"}]}`
-
-These tools will automatically create the correct files in `deliverables/`.
-
-### Step 3: Verify MCP Isolation
-This agent should be using {{MCP_SERVER}} and navigating to jsonplaceholder.typicode.com independently of other parallel agents.
+3. Save a queue deliverable:
+   - Use save_deliverable with deliverable_type "AUTHZ_QUEUE"
+   - Content: {"vulnerabilities": []}
--- a/prompts/pipeline-testing/vuln-injection.txt
+++ b/prompts/pipeline-testing/vuln-injection.txt
@@ -1,37 +1,13 @@
-## 🧪 Pipeline Testing: MCP Isolation Test for Injection Agent
+Please complete these tasks using your MCP tools:

-**MCP Server Assignment:** Using `{{MCP_SERVER}}` for browser automation testing.
+1. Navigate to https://example.com and take a screenshot:
+   - Use {{MCP_SERVER}}__browser_navigate to go to https://example.com
+   - Use {{MCP_SERVER}}__browser_take_screenshot

-### Step 1: Test Browser Isolation
-1. Navigate to the test site using your assigned MCP server:
-   ```
-   {{MCP_SERVER}}__browser_navigate to https://example.com
-   ```
+2. Save an analysis deliverable:
+   - Use save_deliverable with deliverable_type "INJECTION_ANALYSIS"
+   - Content: "# Injection Analysis Report\n\nAnalysis complete. No injection vulnerabilities identified."

-2. Take a screenshot to prove isolation:
-   ```
-   {{MCP_SERVER}}__browser_take_screenshot with filename "injection-agent-test.png"
-   ```
-
-3. Log the test results:
-   ```
-   console.log("🔴 INJECTION AGENT - MCP Server: {{MCP_SERVER}} - Successfully navigated to example.com")
-   ```
-
-### Step 2: Create Required Deliverables
-Use the save_deliverable MCP tool to create both deliverables:
-
-**1. Analysis Report:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "INJECTION_ANALYSIS"`
- `content: "# Injection Analysis Complete (Pipeline Test)\n\n**MCP Server Used:** {{MCP_SERVER}}\n**Test Site:** https://example.com\n**Screenshot:** injection-agent-test.png\n**Status:** Browser isolation test successful"`
-
-**2. Exploitation Queue:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "INJECTION_QUEUE"`
- `content: {"vulnerabilities": [{"ID": "INJ-VULN-01", "verdict": "vulnerable", "confidence": "high", "notes": "Test vulnerability for pipeline validation (MCP: {{MCP_SERVER}})", "vulnerability_type": "SQLi", "source": "query parameter at app.js:15", "path": "controller → db.query", "sink_call": "db.query at app.js:20", "slot_type": "SQL-val", "mismatch_reason": "User input directly concatenated into SQL query", "witness_payload": "'; DROP TABLE users; --"}]}`
-
-These tools will automatically create the correct files in `deliverables/`.
-
-### Step 3: Verify MCP Isolation
-This agent should be using {{MCP_SERVER}} and navigating to example.com independently of other parallel agents.
+3. Save a queue deliverable:
+   - Use save_deliverable with deliverable_type "INJECTION_QUEUE"
+   - Content: {"vulnerabilities": []}
--- a/prompts/pipeline-testing/vuln-ssrf.txt
+++ b/prompts/pipeline-testing/vuln-ssrf.txt
@@ -1,37 +1,13 @@
-## 🧪 Pipeline Testing: MCP Isolation Test for SSRF Agent
+Please complete these tasks using your MCP tools:

-**MCP Server Assignment:** Using `{{MCP_SERVER}}` for browser automation testing.
+1. Navigate to https://httpbin.org and take a screenshot:
+   - Use {{MCP_SERVER}}__browser_navigate to go to https://httpbin.org
+   - Use {{MCP_SERVER}}__browser_take_screenshot

-### Step 1: Test Browser Isolation
-1. Navigate to the test site using your assigned MCP server:
-   ```
-   {{MCP_SERVER}}__browser_navigate to https://httpbin.org
-   ```
+2. Save an analysis deliverable:
+   - Use save_deliverable with deliverable_type "SSRF_ANALYSIS"
+   - Content: "# SSRF Analysis Report\n\nAnalysis complete. No SSRF vulnerabilities identified."

-2. Take a screenshot to prove isolation:
-   ```
-   {{MCP_SERVER}}__browser_take_screenshot with filename "ssrf-agent-test.png"
-   ```
-
-3. Log the test results:
-   ```
-   console.log("🟠 SSRF AGENT - MCP Server: {{MCP_SERVER}} - Successfully navigated to httpbin.org")
-   ```
-
-### Step 2: Create Required Deliverables
-Use the save_deliverable MCP tool to create both deliverables:
-
-**1. Analysis Report:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "SSRF_ANALYSIS"`
- `content: "# SSRF Analysis Complete (Pipeline Test)\n\n**MCP Server Used:** {{MCP_SERVER}}\n**Test Site:** https://httpbin.org\n**Screenshot:** ssrf-agent-test.png\n**Status:** Browser isolation test successful"`
-
-**2. Exploitation Queue:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "SSRF_QUEUE"`
- `content: {"vulnerabilities": [{"ID": "SSRF-VULN-01", "verdict": "vulnerable", "confidence": "high", "notes": "Test SSRF vulnerability for pipeline validation (MCP: {{MCP_SERVER}})", "vulnerability_type": "URL_Manipulation", "source": "url parameter in /api/fetch", "outbound_call": "fetch() at api.js:45", "witness_payload": "http://internal.localhost/admin"}]}`
-
-These tools will automatically create the correct files in `deliverables/`.
-
-### Step 3: Verify MCP Isolation
-This agent should be using {{MCP_SERVER}} and navigating to httpbin.org independently of other parallel agents.
+3. Save a queue deliverable:
+   - Use save_deliverable with deliverable_type "SSRF_QUEUE"
+   - Content: {"vulnerabilities": []}
--- a/prompts/pipeline-testing/vuln-xss.txt
+++ b/prompts/pipeline-testing/vuln-xss.txt
@@ -1,37 +1,13 @@
-## 🧪 Pipeline Testing: MCP Isolation Test for XSS Agent
+Please complete these tasks using your MCP tools:

-**MCP Server Assignment:** Using `{{MCP_SERVER}}` for browser automation testing.
+1. Navigate to https://example.org and take a screenshot:
+   - Use {{MCP_SERVER}}__browser_navigate to go to https://example.org
+   - Use {{MCP_SERVER}}__browser_take_screenshot

-### Step 1: Test Browser Isolation
-1. Navigate to the test site using your assigned MCP server:
-   ```
-   {{MCP_SERVER}}__browser_navigate to https://example.org
-   ```
+2. Save an analysis deliverable:
+   - Use save_deliverable with deliverable_type "XSS_ANALYSIS"
+   - Content: "# XSS Analysis Report\n\nAnalysis complete. No XSS vulnerabilities identified."

-2. Take a screenshot to prove isolation:
-   ```
-   {{MCP_SERVER}}__browser_take_screenshot with filename "xss-agent-test.png"
-   ```
-
-3. Log the test results:
-   ```
-   console.log("🟡 XSS AGENT - MCP Server: {{MCP_SERVER}} - Successfully navigated to example.org")
-   ```
-
-### Step 2: Create Required Deliverables
-Use the save_deliverable MCP tool to create both deliverables:
-
-**1. Analysis Report:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "XSS_ANALYSIS"`
- `content: "# XSS Analysis Complete (Pipeline Test)\n\n**MCP Server Used:** {{MCP_SERVER}}\n**Test Site:** https://example.org\n**Screenshot:** xss-agent-test.png\n**Status:** Browser isolation test successful"`
-
-**2. Exploitation Queue:**
-Use `save_deliverable` MCP tool with:
- `deliverable_type: "XSS_QUEUE"`
- `content: {"vulnerabilities": [{"ID": "XSS-VULN-01", "verdict": "vulnerable", "confidence": "high", "notes": "Test XSS vulnerability for pipeline validation (MCP: {{MCP_SERVER}})", "vulnerability_type": "Reflected", "source": "search parameter", "sink_function": "template.render at search.js:25", "render_context": "HTML_BODY", "mismatch_reason": "User input rendered without HTML encoding", "witness_payload": "<script>alert(1)</script>"}]}`
-
-These tools will automatically create the correct files in `deliverables/`.
-
-### Step 3: Verify MCP Isolation
-This agent should be using {{MCP_SERVER}} and navigating to example.org independently of other parallel agents.
+3. Save a queue deliverable:
+   - Use save_deliverable with deliverable_type "XSS_QUEUE"
+   - Content: {"vulnerabilities": []}
--- a/46
+++ b/46
@@ -142,14 +142,52 @@ cmd_start() {
    exit 1
  fi

-  # Check for API key (router mode can use alternative provider API keys)
+  # Check for API key (Bedrock and router modes can bypass this)
  if [ -z "$ANTHROPIC_API_KEY" ] && [ -z "$CLAUDE_CODE_OAUTH_TOKEN" ]; then
-    if [ "$ROUTER" = "true" ] && { [ -n "$OPENAI_API_KEY" ] || [ -n "$OPENROUTER_API_KEY" ]; }; then
+    if [ "$CLAUDE_CODE_USE_BEDROCK" = "1" ]; then
+      # Bedrock mode — validate required AWS credentials
+      MISSING=""
+      [ -z "$AWS_REGION" ] && MISSING="$MISSING AWS_REGION"
+      [ -z "$AWS_BEARER_TOKEN_BEDROCK" ] && MISSING="$MISSING AWS_BEARER_TOKEN_BEDROCK"
+      [ -z "$ANTHROPIC_SMALL_MODEL" ] && MISSING="$MISSING ANTHROPIC_SMALL_MODEL"
+      [ -z "$ANTHROPIC_MEDIUM_MODEL" ] && MISSING="$MISSING ANTHROPIC_MEDIUM_MODEL"
+      [ -z "$ANTHROPIC_LARGE_MODEL" ] && MISSING="$MISSING ANTHROPIC_LARGE_MODEL"
+      if [ -n "$MISSING" ]; then
+        echo "ERROR: Bedrock mode requires the following env vars in .env:$MISSING"
+        exit 1
+      fi
+    elif [ "$CLAUDE_CODE_USE_VERTEX" = "1" ]; then
+      # Vertex AI mode — validate required GCP credentials
+      MISSING=""
+      [ -z "$CLOUD_ML_REGION" ] && MISSING="$MISSING CLOUD_ML_REGION"
+      [ -z "$ANTHROPIC_VERTEX_PROJECT_ID" ] && MISSING="$MISSING ANTHROPIC_VERTEX_PROJECT_ID"
+      [ -z "$ANTHROPIC_SMALL_MODEL" ] && MISSING="$MISSING ANTHROPIC_SMALL_MODEL"
+      [ -z "$ANTHROPIC_MEDIUM_MODEL" ] && MISSING="$MISSING ANTHROPIC_MEDIUM_MODEL"
+      [ -z "$ANTHROPIC_LARGE_MODEL" ] && MISSING="$MISSING ANTHROPIC_LARGE_MODEL"
+      if [ -n "$MISSING" ]; then
+        echo "ERROR: Vertex AI mode requires the following env vars in .env:$MISSING"
+        exit 1
+      fi
+      # Validate service account key file (must be inside ./credentials/ for Docker mount)
+      if [ -z "$GOOGLE_APPLICATION_CREDENTIALS" ]; then
+        echo "ERROR: Vertex AI mode requires GOOGLE_APPLICATION_CREDENTIALS in .env"
+        echo "       Place your service account key in ./credentials/ and set:"
+        echo "       GOOGLE_APPLICATION_CREDENTIALS=./credentials/gcp-sa-key.json"
+        exit 1
+      fi
+      if [ ! -f "$GOOGLE_APPLICATION_CREDENTIALS" ]; then
+        echo "ERROR: Service account key file not found: $GOOGLE_APPLICATION_CREDENTIALS"
+        echo "       Download a key from the GCP Console (IAM > Service Accounts > Keys)"
+        exit 1
+      fi
+    elif [ "$ROUTER" = "true" ] && { [ -n "$OPENAI_API_KEY" ] || [ -n "$OPENROUTER_API_KEY" ]; }; then
      # Router mode with alternative provider - set a placeholder for SDK init
      export ANTHROPIC_API_KEY="router-mode"
    else
      echo "ERROR: Set ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN in .env"
-      echo "       (or use ROUTER=true with OPENAI_API_KEY or OPENROUTER_API_KEY)"
+      echo "       (or use CLAUDE_CODE_USE_BEDROCK=1 for AWS Bedrock,"
+      echo "        CLAUDE_CODE_USE_VERTEX=1 for Google Vertex AI,"
+      echo "        or ROUTER=true with OPENAI_API_KEY or OPENROUTER_API_KEY)"
      exit 1
    fi
  fi
@@ -209,7 +247,7 @@ cmd_start() {
  fi

  # Ensure audit-logs directory exists with write permissions for container user (UID 1001)
-  mkdir -p ./audit-logs
+  mkdir -p ./audit-logs ./credentials
  chmod 777 ./audit-logs

  # Ensure repo deliverables directory is writable by container user (UID 1001)
--- a/src/ai/claude-executor.ts
+++ b/src/ai/claude-executor.ts
@@ -24,6 +24,7 @@ import { detectExecutionContext, formatErrorOutput, formatCompletionMessage } fr
 import { createProgressManager } from './progress-manager.js';
 import { createAuditLogger } from './audit-logger.js';
 import { getActualModelName } from './router-utils.js';
+import { resolveModel, type ModelTier } from './models.js';
 import type { ActivityLogger } from '../types/activity-logger.js';

 declare global {
@@ -202,7 +203,8 @@ export async function runClaudePrompt(
  description: string = 'Claude analysis',
  agentName: string | null = null,
  auditSession: AuditSession | null = null,
-  logger: ActivityLogger
+  logger: ActivityLogger,
+  modelTier: ModelTier = 'medium'
 ): Promise<ClaudePromptResult> {
  // 1. Initialize timing and prompt
  const timer = new Timer(`agent-${description.toLowerCase().replace(/\s+/g, '-')}`);
@@ -225,22 +227,31 @@ export async function runClaudePrompt(
  const sdkEnv: Record<string, string> = {
    CLAUDE_CODE_MAX_OUTPUT_TOKENS: process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS || '64000',
  };
-  if (process.env.ANTHROPIC_API_KEY) {
-    sdkEnv.ANTHROPIC_API_KEY = process.env.ANTHROPIC_API_KEY;
-  }
-  if (process.env.CLAUDE_CODE_OAUTH_TOKEN) {
-    sdkEnv.CLAUDE_CODE_OAUTH_TOKEN = process.env.CLAUDE_CODE_OAUTH_TOKEN;
-  }
-  if (process.env.ANTHROPIC_BASE_URL) {
-    sdkEnv.ANTHROPIC_BASE_URL = process.env.ANTHROPIC_BASE_URL;
-  }
-  if (process.env.ANTHROPIC_AUTH_TOKEN) {
-    sdkEnv.ANTHROPIC_AUTH_TOKEN = process.env.ANTHROPIC_AUTH_TOKEN;
+  const passthroughVars = [
+    'ANTHROPIC_API_KEY',
+    'CLAUDE_CODE_OAUTH_TOKEN',
+    'ANTHROPIC_BASE_URL',
+    'ANTHROPIC_AUTH_TOKEN',
+    'CLAUDE_CODE_USE_BEDROCK',
+    'AWS_REGION',
+    'AWS_BEARER_TOKEN_BEDROCK',
+    'CLAUDE_CODE_USE_VERTEX',
+    'CLOUD_ML_REGION',
+    'ANTHROPIC_VERTEX_PROJECT_ID',
+    'GOOGLE_APPLICATION_CREDENTIALS',
+    'ANTHROPIC_SMALL_MODEL',
+    'ANTHROPIC_MEDIUM_MODEL',
+    'ANTHROPIC_LARGE_MODEL',
+  ];
+  for (const name of passthroughVars) {
+    if (process.env[name]) {
+      sdkEnv[name] = process.env[name]!;
+    }
  }

  // 5. Configure SDK options
  const options = {
-    model: 'claude-sonnet-4-5-20250929',
+    model: resolveModel(modelTier),
    maxTurns: 10_000,
    cwd: sourceDir,
    permissionMode: 'bypassPermissions' as const,
--- a/src/ai/models.ts
+++ b/src/ai/models.ts
@@ -0,0 +1,37 @@
+// Copyright (C) 2025 Keygraph, Inc.
+//
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Affero General Public License version 3
+// as published by the Free Software Foundation.
+
+/**
+ * Model tier definitions and resolution.
+ *
+ * Three tiers mapped to capability levels:
+ * - "small"  (Haiku — summarization, structured extraction)
+ * - "medium" (Sonnet — tool use, general analysis)
+ * - "large"  (Opus — deep reasoning, complex analysis)
+ *
+ * Users override via ANTHROPIC_SMALL_MODEL / ANTHROPIC_MEDIUM_MODEL / ANTHROPIC_LARGE_MODEL,
+ * which works across all providers (direct, Bedrock, Vertex).
+ */
+
+export type ModelTier = 'small' | 'medium' | 'large';
+
+const DEFAULT_MODELS: Readonly<Record<ModelTier, string>> = {
+  small: 'claude-haiku-4-5-20251001',
+  medium: 'claude-sonnet-4-6',
+  large: 'claude-opus-4-6',
+};
+
+/** Resolve a model tier to a concrete model ID. */
+export function resolveModel(tier: ModelTier = 'medium'): string {
+  switch (tier) {
+    case 'small':
+      return process.env.ANTHROPIC_SMALL_MODEL || DEFAULT_MODELS.small;
+    case 'large':
+      return process.env.ANTHROPIC_LARGE_MODEL || DEFAULT_MODELS.large;
+    default:
+      return process.env.ANTHROPIC_MEDIUM_MODEL || DEFAULT_MODELS.medium;
+  }
+}
--- a/src/services/agent-execution.ts
+++ b/src/services/agent-execution.ts
@@ -156,7 +156,8 @@ export class AgentExecutionService {
      agentName, // description
      agentName,
      auditSession,
-      logger
+      logger,
+      AGENTS[agentName].modelTier
    );

    // 6. Spending cap check - defense-in-depth
--- a/src/services/preflight.ts
+++ b/src/services/preflight.ts
@@ -14,7 +14,7 @@
 * Checks run sequentially, cheapest first:
 * 1. Repository path exists and contains .git
 * 2. Config file parses and validates (if provided)
- * 3. Credentials validate via Claude Agent SDK query (API key, OAuth, or router mode)
+ * 3. Credentials validate via Claude Agent SDK query (API key, OAuth, Bedrock, Vertex AI, or router mode)
 */

 import fs from 'fs/promises';
@@ -24,6 +24,7 @@ import { PentestError, isRetryableError } from './error-handling.js';
 import { ErrorCode } from '../types/errors.js';
 import { type Result, ok, err } from '../types/result.js';
 import { parseConfig } from '../config-parser.js';
+import { resolveModel } from '../ai/models.js';
 import type { ActivityLogger } from '../types/activity-logger.js';

 // === Repository Validation ===
@@ -165,11 +166,75 @@ async function validateCredentials(
    return ok(undefined);
  }

-  // 2. Check that at least one credential is present
+  // 2. Bedrock mode — validate required AWS credentials are present
+  if (process.env.CLAUDE_CODE_USE_BEDROCK === '1') {
+    const required = ['AWS_REGION', 'AWS_BEARER_TOKEN_BEDROCK', 'ANTHROPIC_SMALL_MODEL', 'ANTHROPIC_MEDIUM_MODEL', 'ANTHROPIC_LARGE_MODEL'];
+    const missing = required.filter(v => !process.env[v]);
+    if (missing.length > 0) {
+      return err(
+        new PentestError(
+          `Bedrock mode requires the following env vars in .env: ${missing.join(', ')}`,
+          'config',
+          false,
+          { missing },
+          ErrorCode.AUTH_FAILED
+        )
+      );
+    }
+    logger.info('Bedrock credentials OK');
+    return ok(undefined);
+  }
+
+  // 3. Vertex AI mode — validate required GCP credentials are present
+  if (process.env.CLAUDE_CODE_USE_VERTEX === '1') {
+    const required = ['CLOUD_ML_REGION', 'ANTHROPIC_VERTEX_PROJECT_ID', 'ANTHROPIC_SMALL_MODEL', 'ANTHROPIC_MEDIUM_MODEL', 'ANTHROPIC_LARGE_MODEL'];
+    const missing = required.filter(v => !process.env[v]);
+    if (missing.length > 0) {
+      return err(
+        new PentestError(
+          `Vertex AI mode requires the following env vars in .env: ${missing.join(', ')}`,
+          'config',
+          false,
+          { missing },
+          ErrorCode.AUTH_FAILED
+        )
+      );
+    }
+    // Validate service account credentials file is accessible
+    const credPath = process.env.GOOGLE_APPLICATION_CREDENTIALS;
+    if (!credPath) {
+      return err(
+        new PentestError(
+          'Vertex AI mode requires GOOGLE_APPLICATION_CREDENTIALS pointing to a service account key JSON file',
+          'config',
+          false,
+          {},
+          ErrorCode.AUTH_FAILED
+        )
+      );
+    }
+    try {
+      await fs.access(credPath);
+    } catch {
+      return err(
+        new PentestError(
+          `Service account key file not found at: ${credPath}`,
+          'config',
+          false,
+          { credPath },
+          ErrorCode.AUTH_FAILED
+        )
+      );
+    }
+    logger.info('Vertex AI credentials OK');
+    return ok(undefined);
+  }
+
+  // 4. Check that at least one credential is present
  if (!process.env.ANTHROPIC_API_KEY && !process.env.CLAUDE_CODE_OAUTH_TOKEN) {
    return err(
      new PentestError(
-        'No API credentials found. Set ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN in .env',
+        'No API credentials found. Set ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN in .env (or use CLAUDE_CODE_USE_BEDROCK=1 for AWS Bedrock, or CLAUDE_CODE_USE_VERTEX=1 for Google Vertex AI)',
        'config',
        false,
        {},
@@ -178,12 +243,12 @@ async function validateCredentials(
    );
  }

-  // 3. Validate via SDK query
+  // 5. Validate via SDK query
  const authType = process.env.CLAUDE_CODE_OAUTH_TOKEN ? 'OAuth token' : 'API key';
  logger.info(`Validating ${authType} via SDK...`);

  try {
-    for await (const message of query({ prompt: 'hi', options: { model: 'claude-haiku-4-5-20251001', maxTurns: 1 } })) {
+    for await (const message of query({ prompt: 'hi', options: { model: resolveModel('small'), maxTurns: 1 } })) {
      if (message.type === 'assistant' && message.error) {
        return classifySdkError(message.error, authType);
      }
--- a/src/session-manager.ts
+++ b/src/session-manager.ts
@@ -18,6 +18,7 @@ export const AGENTS: Readonly<Record<AgentName, AgentDefinition>> = Object.freez
    prerequisites: [],
    promptTemplate: 'pre-recon-code',
    deliverableFilename: 'code_analysis_deliverable.md',
+    modelTier: 'large',
  },
  'recon': {
    name: 'recon',
@@ -102,6 +103,7 @@ export const AGENTS: Readonly<Record<AgentName, AgentDefinition>> = Object.freez
    prerequisites: ['injection-exploit', 'xss-exploit', 'auth-exploit', 'ssrf-exploit', 'authz-exploit'],
    promptTemplate: 'report-executive',
    deliverableFilename: 'comprehensive_security_assessment_report.md',
+    modelTier: 'small',
  },
 });

--- a/src/types/agents.ts
+++ b/src/types/agents.ts
@@ -58,6 +58,7 @@ export interface AgentDefinition {
  prerequisites: AgentName[];
  promptTemplate: string;
  deliverableFilename: string;
+  modelTier?: 'small' | 'medium' | 'large';
 }

 /**