Orquestra holds a funded wallet key and spends on-chain. Use a dedicated wallet with only what you’re willing to let the node spend. Start with small budget caps, watch a few cycles, then scale.
1. What it does (30 seconds)
Each cycle (default hourly) the node, for every datanet you enable:
- Votes — scores other operators’ data “pods” 1–10 against the datanet’s rubric and up/down-votes the ones it’s confident about.
- Mints — for datanets with a data adapter, sources + publishes its own pods.
- Claims — collects any finalized emissions you’ve earned.
You earn $REPPO when the crowd up-votes your minted pods (a share of each datanet’s per-epoch emission pool) and through voting participation. There is no compute/inference earning path — the node earns only by publishing and voting.
2. Before you start
You need:
| Requirement | Notes |
|---|
| Docker (with Compose) | the node runs as a container |
| A dedicated wallet | fund with ETH on Base (gas) + REPPO (mint fees, veREPPO lock). Never your main wallet. |
| An LLM API key | powers scoring, the deliberation panel, and the onboarding chat. Anthropic / OpenAI / Google / Surplus / Virtuals. |
| A Pinata JWT | only to mint in “pin” mode (pins pod datasets to IPFS). Not needed for voting, or if you mint every datanet in url-only mode (§7). |
| A private Base RPC (recommended) | the public RPC rate-limits under a full cycle; Alchemy/QuickNode/Ankr remove per-datanet errors. |
mintReppoMax cap and any veREPPO lock you choose.
Sizing mintReppoMax: each mint pays a REPPO fee (≈100–200 observed). The reppo CLI does not report the fee, so the node reserves a conservative ~200 REPPO per mint against your cap before signing (refuse-before, not after). A mintReppoMax below ~200 therefore refuses every mint — set it to roughly 200 × the number of mints you want per budget horizon (e.g. 400 for two). Set RPC_URL and the cap tracks the real (often lower) fee instead of the 200 estimate.
3. Install & run
Fill in .env (every variable is documented inline). Minimum to start:
REPPO_PRIVATE_KEY — your dedicated walletLLM_PROVIDER + LLM_API_KEYRPC_URL — your private Base RPC (recommended)PINATA_JWT — only if minting
These are the only things you set by hand. Your strategy is configured in the dashboard, not here.
3b. Start the node
This pulls the published image, runs it detached with a persistent data volume and restart: unless-stopped, and binds the dashboard to 127.0.0.1:7070. docker ps shows healthy once it’s up. (Pin a version in docker-compose.yml, e.g. :0.1.0, for production.)
Prefer to build from source? docker build -t orquestra ., then point docker-compose.yml’s image: at orquestra.
3c. Reach the dashboard
The dashboard is unauthenticated and localhost-only by design — never expose port 7070 to the internet. To open it from your laptop when the node runs on a remote host, use an SSH tunnel:
ssh -L 7070:localhost:7070 <your-host>
4. First run — onboarding
On first start the node has no strategy and waits for you to configure one in the dashboard (the startup log prints the tunnel command + URL). Open the dashboard and you’ll land in onboarding: a chat on the left, a live strategy “score sheet” on the right.
Tell the assistant what you want, e.g.:
- “Vote-only on the safest datanets to start”
- “Mint geopolitical news with a contrarian angle, small budget”
- “80% of my REPPO into a 90-day veREPPO lock”
As you settle each topic the right panel fills in (datanets, budgets, cadence). It asks one thing at a time and recommends options. When you confirm, it writes your strategy and the node starts its first cycle.
You can re-run onboarding anytime from Strategy → ↻ reconfigure with assistant.
5. The dashboard
Four tabs:
Overview
Your at-a-glance state: net REPPO, earned/claimed/claimable, mint spend, gas, balances, current epoch. Below that:
- Budget burn — spend vs each cap, with bars that turn red near the limit.
- Claimable emissions — pods with finalized rewards waiting to be claimed (the node claims them automatically).
Strategy
The control surface. Each datanet is a card:
- vote / mint chips toggle what the node does there.
- adapter — the data source for minting (
gdelt, hyperliquid, sports); a datanet with no adapter is vote-only.
- strictness — how confident the node must be to act. Hover the ⓘ for the exact score thresholds. Short version: conservative = picky (only acts on strong signals, spends least), aggressive = participates widely (more votes/mints, spends more), balanced = middle.
- + mint strategy — for minted datanets, set focus / angle / items-per-cycle, and mint mode (see §7).
- + add datanet — opens a picker of all active datanets by name.
Below the cards: budget & cadence (caps, how often the node runs — fractional hours allowed, e.g. 0.5 = 30 min), deliberation (multi-agent panel on/off), and your strategy brief (the freeform goal the node votes and mints by).
Changes don’t apply until you hit Save — the diff line shows exactly what you’re about to change. Saves take effect on the next cycle (no restart).
Assistant
The strategy chat in its own tab. Describe a goal in plain language; it proposes a full config change that loads into your Strategy tab for review. Nothing applies until you Save. Use it for “be more aggressive on geopolitics” or “what’s my current setup?”.
Activity
Every vote, mint, claim, and skip, newest first. Filter by kind. A ⚖ 3 badge means a multi-agent panel decided that one — click it to open the debate drawer (bull / bear / rubric-purist scores + arguments, and the judge’s verdict).
This is where you spot a misbehaving or idle datanet: a skip row says why nothing happened — RPC error, subnet access not granted, no on-chain rubric/spec, an unregistered adapter, or “candidates discovered but none passed scoring.” If a datanet you enabled never produces votes/mints, filter to skip and read the reason.
6. How earning works (and why some pods don’t)
- Emissions are per-epoch and lag. A pod earns a share of its datanet’s emission pool only after that epoch finalizes — rewards trail votes by about an epoch (epochs are ~48h). So freshly-minted, freshly-upvoted pods show
claimable: 0 for a while. This is normal, not a bug.
- Only net-positive pods earn. A pod must have a positive net vote (upvotes − downvotes) at finalization to be accepted into the curated dataset and share the pool. A net-downvoted pod earns nothing — the crowd rejected that data, and the mint fee you paid for it is lost.
- Claiming is automatic. With
claimEmissions on (default), the node claims the instant rewards are due; you’ll see claimable→claimed and net REPPO rise.
Practical read: watch Activity → mints and Overview. If a datanet keeps producing net-downvoted or no-vote pods, its adapter is publishing data the crowd doesn’t value — tighten its strictness or switch it to vote-only.
7. Mint mode — pin vs url-only (do you need Pinata?)
Each minted datanet has a mint mode, set per datanet in the Strategy tab:
- pin (default) — the node uploads the pod’s dataset JSON to IPFS via your Pinata key. Use it where the dataset is the value (e.g. trade data) and curators pull the downloadable to judge it.
- url-only — the node registers the candidate’s source URL as the pod, with no pinning and no Pinata. Use it for link-type pods (e.g. news articles). Candidates with no source URL are skipped in this mode.
So Pinata is required only if at least one minted datanet is in pin mode. A vote-only node, or one that mints everything url-only, doesn’t need a Pinata key at all.
Tip: before switching a datanet to url-only, confirm its pods still earn — for some datanets curators score the pinned dataset, not just the link.
8. The multi-agent panel
For close calls (and every mint), the node can convene a panel — bull, bear, and a rubric-purist each argue a score, and a judge rules. It catches borderline mistakes a single scorer would make. Toggle it in Strategy → deliberation; the voteBand controls how close to a threshold a vote must be to trigger the panel (0 = panel on mints only). Every panel decision is inspectable in the Activity debate drawer.
Cost note: a panel decision is ~4 LLM calls vs 1. Tiering keeps clear-cut votes cheap; only ambiguous votes and mints pay the full cost.
9. Updating
docker compose pull && docker compose up -d
10. Troubleshooting
| Symptom | Cause / fix |
|---|
| Dashboard won’t load | Check the SSH tunnel is up and you’re hitting localhost:7070. docker ps should show healthy. |
PUBLIC_API_UNREACHABLE in logs | A transient reppo.ai blip. The node retries automatically; a one-off is harmless. Persistent = check the node’s internet/DNS. |
Datanet skips with INTERNAL_ERROR | The public Base RPC is rate-limiting. Set RPC_URL to a private endpoint. |
| 0 votes every cycle | Nothing new to vote on (all current pods already voted, yours, or skipped mid-range). Normal between fresh pods. |
| Mints score low / nothing mints | The strictness gate is rejecting candidates. Don’t loosen it just to force mints — that publishes low-value data that gets downvoted. |
claimable: 0 despite upvotes | Emissions lag — wait for the epoch to finalize (§6). |
| Onboarding chat 503s | The node started without LLM_PROVIDER / LLM_API_KEY. Set them in .env and restart. |
docker compose logs -f.
11. Safety & cost model
- The node cannot spend beyond your budget caps — the budget ledger refuses before signing, not after.
- Enabling a datanet is consent to pay its one-time subnet-access grant; cap or disable grants with
budget.grantReppoMax.
- The private key sits in
.env in plaintext — use a dedicated wallet.
- The dashboard is unauthenticated and localhost-bound on purpose (see ADR 0002). Never publish port 7070; anyone who reaches it can rewrite your strategy and spend your budget.
12. FAQ
Do I need to keep my laptop open? No — the node runs on its host (VPS or machine) independently. The SSH tunnel is only for viewing/configuring the dashboard.
Can I change strategy while it’s running? Yes. Edit in the Strategy tab and Save; it applies next cycle, no restart.
How much will it spend? Never more than your caps. Mint fees (REPPO) are the main cost; gas on Base is negligible. Start small.
What’s “deliberation” costing me? Extra LLM API calls on ambiguous votes and mints. If your LLM bill matters, set the panel off or voteBand: 0 (mints only).
Why is net REPPO negative at first? You pay mint fees up front; emissions lag. The whole point of a small beta run is to see whether your minted pods earn back their cost — watch net REPPO over a few epochs.
