Copied to clipboard
🪣 Amazon Bedrock · Nova Pro vision · promo credits

Amazon Bedrock for Vision — Free Credits ($100–200)

An advanced, optional step: claim new-AWS promotional credits, enable Amazon Nova Pro on Bedrock, and wire it into Hermes as a vision model for vision_analyze and auxiliary image tasks. Reka's free monthly credits get you running first; Bedrock's much larger one-time pool (~$100–200) then takes over the heavy/high-end vision while it lasts — the setup wizard's default Reka / InferHub path is cheaper and simpler and stays on as your everyday fallback.

1Create AWS account 2Check credit balance 3Enable Nova access 4Create IAM key 5Wire into Hermes

📌 What this is, honestly

Bedrock is Amazon's managed foundation-model platform. It has no standalone free tier — but brand-new AWS accounts start with promotional credits you can spend on Bedrock (including Nova models). This page helps you claim them while they last.

🎁

Promo credits, not a free tier

Bedrock bills per token and has no perpetual free tier. Instead, new AWS accounts typically get $100 in signup credits plus up to ~$100 more for using services — roughly $100–200 total, spendable on Bedrock including Nova. They're valid about 6 months.

💳

A card is required

You need a credit/debit card for AWS account identity verification. AWS won't charge it during the credit period for these workloads, but the card must be valid to create the account.

👁️

Nova Pro = vision

Amazon Nova Pro is a capable multimodal vision model — well-suited for Hermes' vision_analyze and auxiliary vision tasks. On Bedrock its model id is us.amazon.nova-pro-v1:0.

🔐

Auth = AWS IAM

Bedrock uses AWS IAM credentials — AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, and a region — not a simple API key. You create a programmatic key with Bedrock permissions in IAM.

🆓
Credit this page. The promo-credit picture is based on freellmcredits.com and pricepertoken.com. Exact offers change and vary by region — always confirm your actual balance in the AWS console before relying on it.

📘 Hermes has a first-class Bedrock guide

You don't have to wing the wiring. Hermes' official docs include a dedicated Bedrock setup, covering the native Converse API, IAM authentication, and model access. Nova Pro is available there.

Official guide
https://hermes-agent.nousresearch.com/docs/guides/aws-bedrock
✅
Use the guide as your source of truth for exact config.yaml syntax, Converse API details, and IAM wiring. The steps below are the "what & why" — the guide is the exact "how."

🪜 How to claim & enable

Five steps from a brand-new AWS account to a working Nova Pro vision model in Hermes. Expect a bit of AWS console navigation — it's more friction than the Reka/InferHub path, but the credits are worth it while they last.

1Sign up for AWS aws.amazon.com

  1. Go to aws.amazon.com → Create an AWS Account and follow the identity + contact steps.
  2. Enter a valid credit/debit card for identity verification (required — it may place then release a small hold; it is not a charge for these workloads).
  3. Complete phone & email verification to finish creating the account.

2Confirm your credit balance

  1. In the AWS console, open Billing → Cost Management and look under Credits.
  2. You should see your promotional credit balance (commonly ~$100, sometimes with service-usage bonuses toward $100–200).
  3. Note the expiry (commonly ~6 months). This is a one-time promo, not a perpetual allowance.

3Enable model access on Bedrock

  1. Open the Amazon Bedrock console in your region.
  2. Under Bedrock configurations → Model access, click Modify / Manage model access.
  3. Request access to Amazon Nova (Nova Pro included) and confirm. Access grants are typically quick.

4Create an IAM programmatic key

  1. Open IAM → Users and create a user (programmatic access).
  2. Attach a policy granting Bedrock access (e.g. AmazonBedrockFullAccess or a scoped Bedrock invoke policy).
  3. Create an access key and note the Access Key ID and Secret Access Key — these are your AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY credentials.

5Wire Nova Pro into Hermes

  1. Follow the official Bedrock guide for exact config (region, credentials, model wiring).
  2. Set Nova Pro as Hermes' vision model in config.yaml under auxiliary.vision, pointing at model id us.amazon.nova-pro-v1:0.
  3. Test with hermes doctor and a quick vision_analyze on a sample image.

⚙️ Where it goes in config

Nova Pro plugs in as Hermes' auxiliary vision model. Because exact syntax and credential placement can vary with your Hermes version, treat the snippet below as the shape and confirm the precise fields in the official Bedrock guide.

config.yaml — auxiliary vision (confirm exact syntax in the guide)
auxiliary: vision: provider: bedrock # region + AWS IAM creds come from env / guide model: us.amazon.nova-pro-v1:0 # Amazon Nova Pro on Bedrock
Credentials (in your Hermes secrets / env — never commit)
AWS_ACCESS_KEY_ID=AKIA... AWS_SECRET_ACCESS_KEY=... AWS_REGION=us-east-1
⚠️
Don't trust the snippet verbatim for your exact Hermes version. The authoritative, current syntax lives at hermes-agent.nousresearch.com/docs/guides/aws-bedrock — follow it for the precise key names, env-var vs config placement, and Converse API settings.

🧲 Same credits: switch embeddings to Cohere for precision

Your Bedrock free credits don't just unlock vision — they can upgrade semantic search too. Hermes looks things up in its notes/memory using embeddings, and the embedding model you choose sets how precisely those lookups match. On a free Oracle VPS you typically default to a small self-hosted embedder; with Bedrock credits you can switch to Cohere Embed v4, which is a big precision step up.

1Why Cohere Embed v4

🎯

Precision lookups

Cohere Embed v4 (multimodal — text & images) ranks near the top of embedding benchmarks. For searching your own notes, questions, or lecture files by meaning, it returns the right note far more reliably than a basic local embedder.

💲

Cheap + covered by credits

About $0.12 per million input tokens, $0.00 output. Embedded pages are small, so your Bedrock credits stretch a long way here — a perfect use of the free pool.

☁️

On Bedrock (no extra signup)

Cohere Embed v4 is a first-class Amazon Bedrock model — same AWS account, same IAM credentials, same credits you enabled for Nova. Model id: cohere.embed-v4:0.

2The switch in Hermes

  1. Enable Cohere Embed v4 in your Bedrock console (Model access → Amazon Bedrock → add cohere.embed-v4), alongside Nova.
  2. Re-run whatever indexes your notes/memory so chunks are re-embedded with Cohere (semantic search rebuild).
  3. Confirm the embedding provider/model in your Hermes config points at Bedrock Cohere — follow the official Bedrock guide for the exact key name on your version.
  4. Query by meaning and notice the better matches on specific-fact lookups.
🔎
Two-stage tip (best of both): for really precise fact lookups, keep a broad local recall first, then re-rank the top candidates with Cohere Rerank (also on Bedrock, model cohere.rerank-v3.5:0). Embedding finds *plausible* notes; reranking picks the *best* one. Same credit pool.
⚠️
Don't trust exact config keys from this page. The precise embedding/rerank wiring and whether your Hermes version exposes it as a config key or via your memory/search tool varies — follow the official Bedrock guide and your memory/search provider's docs.

🧭 Honest caveats before you commit

Bedrock is great for free vision — but only while the credits last. Know what you're signing up for.

💸
It bills per-token after credits run out. Once the promo credits are spent or expire, Bedrock charges per token at Nova Pro rates. Set a billing budget alert in AWS (Billing → Budgets) so you're never surprised.
⚖️
Nova Pro costs more than your default. Per-token, Nova Pro is pricier than DeepSeek V4 Flash or Reka edge. Treat it as a free-credit "nice vision" option, not your everyday default model — keep Reka/InferHub as the workhorse.
🚧
AWS signup friction is higher than Reka or InferHub — card verification, account setup, IAM, console navigation. And the credits are a one-time promo, not perpetual. If Bedrock ever stops making sense, simply switch your vision model back to the default.
Start AWS signup → Read the Bedrock guide ← Back to setup wizard