> For the complete documentation index, see [llms.txt](https://docs.obol.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.obol.org/next/obol-stack/build-a-profitable-stack.md). # Build a Profitable Obol Stack This is the long-form narrative for an Obol Stack operator who wants to actually make money. It assumes you've finished the [Quickstart](/next/obol-stack/quickstart.md) — you have `obol stack up` running, you can talk to your default Hermes agent, and you've watched `obol sell demo` settle a payment. The journey: 1. Sync a **bounded archive node** of the chain your target application lives on, starting from the block at which it was deployed. 2. Build a **purpose-built index or feed** over that archive. 3. Wrap the index as a **paid HTTP service** (`obol sell http`). 4. Build a **specialized agent** on top that uses the index and a tight skill set to answer buyer questions better than any general-purpose agent could. 5. Sell the agent's replies via `obol sell agent` and **get listed** on the marketplaces where buyer-agents look. 6. Set up out-of-band notifications so your agent can ping you on **Telegram** when the work is done — you don't need to babysit it. 7. **Iterate**. You can use Claude Code as your operator throughout — the [`run-obol-stack` skill](https://github.com/ObolNetwork/skills/blob/main/skills/run-obol-stack/SKILL.md) drives all of this and knows the right next command at every step. ## Step 1: pick the right archive scope The single most important decision is **what history you actually need**. A full mainnet archive from genesis is \~4 TB and weeks of sync — overkill for almost every realistic agent business. A bounded archive is the right default. If your target application is a DeFi protocol deployed at block `N`, install with `--mode=archive --since=`. You'll carry the state trie from `N` forward and nothing earlier: ```shell obol network install ethereum \ --network=mainnet \ --mode=archive \ --since=22500000 # the block your app was deployed obol network sync ethereum ``` Pick `--since` based on what your agent needs to answer: * **A specific app deployed at block N** → `--since=N`. * **Anything Cancun-or-later** → `--since=cancun` (\~800 GB). * **Anything from the last year** → `--since=365d` (\~600 GB). * **Recent execution-layer history for a quick indexer** → `--since=prague` (\~0.4 TB). See [Installing Networks](/next/obol-stack/installing-networks.md#archive-nodes-and-bounded-history-since) for the full `--since` reference. While the node syncs (hours to days depending on scope), keep moving. The other steps don't require the archive to be complete. ## Step 2: build the index your agent will sell A raw archive node is not yet a product — it's storage. The product is **a query surface that returns useful, application-specific answers in milliseconds**. Two reliable shapes: * **An events index.** Subscribe to your app's contract logs from `--since` forward, decode them, and write them to a small Postgres or DuckDB. Buyers query "all positions opened between X and Y", "all liquidations involving address Z", etc. * **A state snapshot service.** Run historical `eth_call` against the archive at the block ranges that matter, cache the results, and expose a stable HTTP endpoint that returns "what did this contract look like at this block?" without paying the archive node's full cost per query. Deploy whatever you build as a regular Kubernetes Service in your cluster — the `obol app install` command takes any Helm chart, and the `obol-app` chart in [`ObolNetwork/helm-charts`](https://github.com/ObolNetwork/helm-charts) packages an arbitrary Dockerfile for you. ## Step 3: sell the index Once the index is healthy, wrap it as a paid HTTP service: ```shell obol sell http my-index \ --upstream my-index-svc \ --port 8080 \ --namespace my-ns \ --per-request 0.001 \ --chain base \ --token USDC \ --wallet 0x...your-wallet... ``` Confirm it reconciles to `Ready`: ```shell obol sell status my-index --namespace my-ns ``` You now have a billable data feed. Some buyers — particularly other agents writing custom analyses — will pay for this directly. ## Step 4: build the specialized quant agent The bigger margin lives one layer up. Buyers who don't want to write their own analysis will happily pay 10–50× more for an **agent that already knows what to do with the index**. Inside the Obol Agent, the levers that compound are: * **`SOUL.md`** — the agent's system prompt. The difference between "an LLM that knows about Ethereum" and "a quant agent that answers a specific class of questions reliably" is two well-edited paragraphs of `SOUL.md`. * **Custom skills.** Add a skill that knows how to query your index, decode your app's events, and present the result in the format your buyers actually want. * **Reference data.** Give the agent direct access to the index (and any other curated data you have) so it doesn't have to guess. The `run-obol-stack` skill calls this the **margin-bearing path**: the LLM is fungible, but a tight skill plus proprietary data is not. Iterate on `SOUL.md` and the skill until paid buyers consistently give thumbs-up answers. Once the agent is reliable, expose it as a paid agent service: ```shell obol sell agent my-quant \ --instance my-quant \ --per-request 0.05 \ --token USDC \ --chain base ``` Buyers pay per reply, not per token — and pay for the answer, not the inference. ## Step 5: get a permanent URL So far your service has been reachable on a temporary quick-tunnel URL (`https://.trycloudflare.com`) that changes on every restart. Before you register and start attracting buyers, give it a stable home — otherwise anyone who bookmarked or registered the old URL hits errors after a restart. ```shell # 1. In the Cloudflare dashboard: Networks → Tunnels → Create a tunnel. # 2. Add a Public Hostname: # Subdomain/Domain: stack.example.com # Service: http://traefik.traefik.svc.cluster.local:80 # 3. Copy the connector token and hand it to Obol: obol tunnel setup --hostname stack.example.com ``` You can paste the whole `cloudflared tunnel run --token …` line the dashboard shows — Obol strips the prefix and keeps the token. This uses a least-privilege, single-tunnel **connector token**, not an account-wide API key, so the credential can only run this one tunnel. Confirm it's live with `obol tunnel status`. For the full dashboard walkthrough with screenshots, see [Set up a permanent URL](/next/obol-stack/permanent-url.md). ## Step 6: get listed Register your agent on [ERC-8004](https://eips.ethereum.org/EIPS/eip-8004) so buyers can find it: ```shell obol sell register \ --chain mainnet \ --name my-quant \ --private-key-file ~/.config/obol/agents/my-quant/wallet.json ``` Registration writes to the on-chain Identity Registry. From there, multiple marketplaces and explorers index your offer: * [**x402scan.com**](https://x402scan.com) — the broadest cross-chain x402 registry. * [**Coinbase Bazaar**](https://www.coinbase.com/) — Coinbase's USDC-on-Base agent marketplace. * **The ERC-8004 contracts directly** — for buyer-agents written against the raw registry. You don't need to pick one marketplace; registering once gets you indexed by all of them. See [Selling agent services](/next/obol-stack/selling-services.md#get-discoverable-erc-8004-and-marketplaces) for the longer marketplace orientation. ## Step 7: tell your agent to ping you on Telegram By the time you have an archive node syncing in the background, an index updating, and a paid agent live on a tunnel — you probably don't want to sit and watch a terminal. `obol hermes setup` wires Hermes to a Telegram bot (or Discord, Slack, etc.) so it can message you when things change. The Telegram bot flow involves talking to two other Telegram bots: 1. **Talk to `@BotFather`** in Telegram (`https://t.me/BotFather`). Send `/newbot`, follow the prompts to pick a name and a username for your bot, and you'll receive an **HTTP API token** (something like `1234567890:ABCdefGhIJKlmNoPQRstUvWXYZabcdefghi`). Save it — this is how Hermes will send messages **as** the bot. 2. **Talk to `@userinfobot`** in Telegram (`https://t.me/userinfobot`). It will reply with your numeric **Telegram user ID** (something like `123456789`). Save this — this is the chat the bot will message. 3. **Start a chat with your new bot** by clicking through the link `@BotFather` gave you, and send it `/start`. Telegram won't let a bot DM you until you initiate the conversation. 4. **Run `obol hermes setup`** and paste both the bot token and your numeric user ID when prompted. Test it: ask the agent to ping you when the next sync milestone hits. If you get a Telegram message, you're done — you can close the laptop and let the agent work. ## Step 8: iterate The compounding loop: * Watch which buyer questions your agent answers poorly — that's the prompt for the next `SOUL.md` edit or skill improvement. * Watch which queries against your index are popular — that's the prompt for the next index expansion (more contracts decoded, deeper historical coverage, etc.). * Watch your revenue per buyer. Specialized work is not price-sensitive in the way raw inference is — once you're reliably better than the alternatives, raise the price. * When you stop an offer, **drain it gracefully** (`obol sell stop --grace 1h`). On-chain buyers' reputation systems penalize abrupt teardown. A profitable Obol Stack is one where the operator is *bored*. The agent syncs the data, builds the index, sells the queries, sells the replies, pings the operator on Telegram when something interesting happens, and otherwise gets on with it. ## Where to go from here * [Selling agent services](/next/obol-stack/selling-services.md) — in-depth coverage of `obol sell`, x402 economics, and marketplaces. * [Installing Networks](/next/obol-stack/installing-networks.md) — full archive `--since` reference and multi-network setup. * [Installing Apps](/next/obol-stack/installing-apps.md) — deploying any Helm chart (your index, your custom services) into the stack. * [`run-obol-stack` Claude Code skill](https://github.com/ObolNetwork/skills/blob/main/skills/run-obol-stack/SKILL.md) — drive the entire flow from your terminal, with a Claude that already knows the right next command. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://docs.obol.org/next/obol-stack/build-a-profitable-stack.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.