Skip to content

aws-setup.md

Latest commit

 

History

History
127 lines (94 loc) · 3.03 KB

aws-setup.md

File metadata and controls

127 lines (94 loc) · 3.03 KB

AWS Setup

code-index uses AWS Bedrock for two things:

  1. LLM summaries — generating natural language descriptions of your code
  2. Vector embeddings — converting summaries into vectors for similarity search

Required Bedrock models

You need access to these models in your AWS account:

Purpose Model Model ID
Function summaries Claude Haiku 4.5 us.anthropic.claude-haiku-4-5-20251001-v1:0
File/package summaries Claude Sonnet 4.6 us.anthropic.claude-sonnet-4-6
Embeddings Cohere Embed v4 cohere.embed-v4:0

The Anthropic models are available by default on Bedrock. Cohere Embed v4 is an AWS Marketplace model — a user with marketplace permissions needs to invoke it once to enable it account-wide.

Credential configuration

code-index uses the standard AWS SDK credential chain. Any of these methods work:

AWS SSO (recommended for developers)

# Configure your profile (one-time)
aws configure sso --profile my-profile

# Login before using code-index
aws sso login --profile my-profile

# Set the profile
export AWS_PROFILE=my-profile

Environment variables

export AWS_ACCESS_KEY_ID=...
export AWS_SECRET_ACCESS_KEY=...
export AWS_SESSION_TOKEN=...    # if using temporary credentials
export AWS_REGION=us-east-1

IAM role (for CI)

In GitHub Actions, use OIDC-based role assumption:

- uses: aws-actions/configure-aws-credentials@v5
  with:
    role-to-assume: arn:aws:iam::123456789012:role/code-index-gha
    aws-region: us-east-1

See CI Setup for the full workflow.

IAM permissions

The IAM role or user needs these permissions:

For LLM summaries and embeddings

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": "bedrock:InvokeModel",
      "Resource": [
        "arn:aws:bedrock:*::foundation-model/anthropic.*",
        "arn:aws:bedrock:*::foundation-model/cohere.*"
      ]
    }
  ]
}

For S3 vector distribution (optional)

If you're distributing the vector database via S3:

{
  "Effect": "Allow",
  "Action": [
    "s3:GetObject",
    "s3:PutObject",
    "s3:ListBucket"
  ],
  "Resource": [
    "arn:aws:s3:::my-code-index",
    "arn:aws:s3:::my-code-index/*"
  ]
}

Profile auto-detection

The scripts/pull-code-index-vectors.sh script can automatically find a working AWS profile. Configure it in .code-index.json:

{
  "aws": {
    "account": "123456789012",
    "profiles": ["dev", "staging", "production"]
  }
}

The script checks the current profile first. If it doesn't match the configured account, it tries each profile in the list until one works.

Cost estimate

At typical usage (10K items indexed):

Operation Model Cost
Function summaries Haiku ~$0.50
File summaries Sonnet ~$1.00
Package summaries Sonnet ~$0.50
Embeddings Cohere Embed v4 ~$0.16
Total full rebuild ~$2

Incremental runs (typical daily update) cost a few cents.