AI Chat
May 5, 2026 ยท View on GitHub
AI Chat provides an intelligent assistant that helps you analyze AWS resources, compare configurations, identify security risks, and navigate documentation.
Overview
Press A in the following views to open AI Chat:
- Resource Browser (list view) - Analyzes visible resources
- Detail View - Analyzes the selected resource
- Diff View - Compares two resources side-by-side
The assistant has access to:
- Current resource context (what you're viewing)
- Active AWS profile and region
- Tools to query resources, fetch logs, and search AWS documentation
Setup
1. IAM Permissions
The AI Chat feature uses Amazon Bedrock. You need the following permission:
{
"Effect": "Allow",
"Action": "bedrock:InvokeModelWithResponseStream",
"Resource": "arn:aws:bedrock:*::foundation-model/*"
}
See IAM Permissions for details.
2. Configuration
Configure AI Chat in ~/.config/claws/config.yaml:
ai:
profile: "" # AWS profile for Bedrock (empty = use current profile)
region: "" # AWS region for Bedrock (empty = use current region)
model: "global.anthropic.claude-haiku-4-5-20251001-v1:0" # Bedrock model ID
max_sessions: 100 # Max stored sessions (default: 100)
max_tokens: 16000 # Max response tokens (default: 16000)
thinking_budget: 8000 # Extended thinking token budget (default: 8000)
max_tool_rounds: 15 # Max tool execution rounds per message (default: 15)
max_tool_calls_per_query: 50 # Max tool calls per user query (default: 50)
save_sessions: false # Persist chat sessions to disk (default: false)
See Configuration for all options.
Usage
Opening Chat
Press A in list/detail/diff views to open the AI Chat overlay.
What the AI Can Do
- List and query AWS resources across services and regions
- Get detailed information about specific resources
- Fetch CloudWatch logs for supported resources (Lambda, ECS, CodeBuild, etc.)
- Search AWS documentation with general, public AWS terms
The AI automatically uses the current profile, region, and resource context from your view.
AWS documentation search sends the search query to AWS's public documentation search endpoint. To avoid leaking private AWS context, claws rejects documentation queries that contain resource IDs, account IDs, ARNs, profile names, logs, tag values, or secrets.
Context Awareness
The assistant automatically receives context based on your current view:
Resource Browser (List View):
Currently viewing: ec2/instances (us-west-2, production profile)
Visible resources: [i-abc123, i-def456, ...]
Detail View:
Currently viewing: ec2/instances/i-abc123 (us-west-2, production profile)
Resource details: {...}
Diff View:
Comparing two resources:
Left: ec2/instances/i-abc123
Right: ec2/instances/i-def456
Session History
Press Ctrl+H to view and resume previous chat sessions.
Keyboard Shortcuts
| Key | Action |
|---|---|
A | Open AI Chat (in list/detail/diff views) |
Ctrl+H | Session history |
Enter | Send message |
Esc | Close chat / Cancel stream |
Ctrl+C | Cancel stream |
Extended Thinking
The assistant supports extended thinking for complex queries. When enabled, you'll see a thinking indicator showing the assistant's reasoning process before the final response.
Configure thinking budget in config.yaml:
ai:
thinking_budget: 8000 # Max tokens for extended thinking (default: 8000)
Troubleshooting
"Bedrock not available in this region"
Bedrock is not available in all AWS regions. Configure a supported region in your config:
ai:
region: "us-west-2" # Use a region where Bedrock is available
"Access Denied" errors
Ensure your IAM role/user has the required Bedrock permissions. See IAM Permissions.
Tool call limit reached
If you see "Tool call limit reached", the assistant made too many tool calls in a single query. Increase the limit:
ai:
max_tool_calls_per_query: 100 # Increase from default 50
Session not persisting
Enable session persistence in config:
ai:
save_sessions: true # Default: false
Sessions are stored in ~/.config/claws/chat/sessions/.