Skip to content
HEBBS

Quickstart

1. Deploy the Server

Section titled “1. Deploy the Server”
Terminal window
curl -sSf https://hebbs.ai/server | OPENAI_API_KEY=sk-... sh

This starts the HEBBS engine and platform on your server. Open http://your-server:8080 to complete onboarding: create your admin account, name your first workspace, and save the generated API key.

2. Install the CLI

Section titled “2. Install the CLI”

On your laptop (macOS or Linux):

Terminal window
curl -sSf https://hebbs.ai/install | sh
export PATH="$HOME/.hebbs/bin:$PATH"

3. Connect to Your Server

Section titled “3. Connect to Your Server”
Terminal window
hebbs login --endpoint http://your-server:8080 --api-key hb_live_sk_...

Verify:

Terminal window
hebbs status

4. Upload Files

Section titled “4. Upload Files”
Terminal window
hebbs push ./your-docs-folder

HEBBS indexes all .md, .txt, and .pdf files. Each file is split into propositions, embedded, and stored with entity relationships. This takes 30-60 seconds per file.

Use the entities/ folder convention to auto-scope content:

your-docs/
├── entities/
│   ├── acme-corp/
│   │   └── call-notes.md      → entity_id: "acme-corp"
│   └── initech/
│       └── discovery.md        → entity_id: "initech"
└── products/
    └── features.md             → shared knowledge

5. Recall

Section titled “5. Recall”
Terminal window
hebbs recall "what happened with acme?"

Try different strategies:

Terminal window
hebbs recall "budget discussion" --entity-id acme-corp        # scoped to entity
hebbs recall "what happened in order" --strategy temporal      # chronological
hebbs recall "what led to the budget cut" --strategy causal    # cause and effect

6. Use the Python SDK

Section titled “6. Use the Python SDK”
Terminal window
pip install hebbs
import asyncio
from hebbs.rest_client import HebbsRestClient


async def main():
    async with HebbsRestClient("http://your-server:8080", api_key="hb_live_sk_...") as hb:
        await hb.remember("Customer prefers email over phone", entity_id="acme-corp", importance=0.8)
        results = await hb.recall("contact preference")
        print(results.text)


asyncio.run(main())

7. Use the TypeScript SDK

Section titled “7. Use the TypeScript SDK”
Terminal window
npm install @hebbs/sdk
import { HebbsRestClient } from '@hebbs/sdk';


const hb = new HebbsRestClient("http://your-server:8080", { apiKey: "hb_live_sk_..." });
await hb.remember("Customer prefers email over phone", { entityId: "acme-corp", importance: 0.8 });
const results = await hb.recall("contact preference");
console.log(results.text);
await hb.close();

8. Tune Recall for Your Domain

Section titled “8. Tune Recall for Your Domain”

Default recall works, but tuned recall is significantly better. Adjust weights to match how your team searches:

Terminal window
# Default: balanced (relevance 0.5, recency 0.2, importance 0.2, reinforcement 0.1)
hebbs recall "Acme budget" --entity-id acme-corp


# Recency-biased (for sales: recent interactions matter most)
hebbs recall "Acme budget" --entity-id acme-corp --weights 0.3:0.4:0.2:0.1


# Importance-biased (for legal: high-stakes content first)
hebbs recall "retention policy" --weights 0.3:0.1:0.5:0.1

HEBBS with tuned weights improves recall precision by 50-70% over defaults. See Tuning & Evals for the full process.

What’s Next

Section titled “What’s Next”