Writing bots

You finished Quickstart: a predict callback inside trepa.bots.run. This page describes what happens each round, how to skip, how to revise a forecast after submit, optional callbacks, and stopping the process.

Credentials in `.env`

Details for a single bot match Quickstart:

TREPA_API_KEY=trp_your_api_key
TREPA_PRIVATE_KEY=your_base58_wallet_private_key

import { credentialsFromEnv, Trepa } from '@trepa/sdk'

const trepa = new Trepa({ credentials: credentialsFromEnv() })

Load .env when you run Node, for example node --env-file=.env bot.ts, or set the same variable names in your host environment. If only one of the two variables is set, the SDK throws at startup. Several bots in one process use numbered pairs: TREPA_API_KEY_1 with TREPA_PRIVATE_KEY_1, then _2, _3, and so on until there is a gap. See Swarms. Optional treasury: TREPA_MASTER_PRIVATE_KEY funds bot wallets in the background. It does not replace bot keys. See Master wallet.

Private keys are highly sensitive. Anyone with them can move funds. Keep .env out of git and off shared channels.

Each round: `predict`

When a pool is ready, Trepa calls predict. Return { value, stake } to submit your forecast and stake in USDC. Return null to skip that turn. Example. Reuse the Trepa client from the credentials section above:

await trepa.bots.run({
  predict: (pool) => ({
    value: 65_000,
    stake: pool.min_stake,
  }),
})

Use an async predict when you need network or disk. If predict throws, the error is logged and the loop keeps running.

Revising after submit: `updatePrediction`

Some strategies submit once, then adjust before the window closes. When you add updatePrediction, the SDK still runs predict first. After the initial forecast is created on-chain, it calls updatePrediction with info: the pool, prediction id, and the value and stake from that create. Example. After a short delay, the handler either submits a revised value or returns null to leave the prediction unchanged:

await trepa.bots.run({
  predict: (pool) => ({
    value: 65_000,
    stake: pool.min_stake,
  }),
  updatePrediction: async (info) => {
    await new Promise((r) => setTimeout(r, 5_000))
    if (info.value >= 66_000) return null
    return { value: info.value + 250 }
  },
})

Return { value } to submit a new number, or null to leave the prediction unchanged.

Optional context: `ctx`

The examples above only read pool. When you also need the signed-in user or a Trepa client scoped to this slot, add the second argument:

ctx.me: the Trepa user for this bot.
ctx.trepa: the API client for that same user.

await trepa.bots.run({
  predict: async (pool, ctx) => {
    const stats = await ctx.trepa.users.statistics(ctx.me.id)
    return {
      value: 65_000,
      stake: pool.min_stake,
    }
  },
})

Hooks

If you want logging or side effects around submits without changing the forecast logic itself, wire optional callbacks next to predict:

Hook	When it runs
`onStart`	Right after this bot signs in.
`onPredicted`	After the initial forecast is submitted successfully.
`onPredictionUpdated`	After an `updatePrediction` revision is submitted successfully. Not fired when you return `null`.
`onPoolSkipped`	When the bot does not submit for that turn.
`onError`	When the loop recovers from a problem.

await trepa.bots.run({
  predict: (pool) => ({ value: 65_000, stake: pool.min_stake }),
  onPredicted: ({ pool, value }) => `${pool.title}: ${value}`,
  updatePrediction: async (info) => {
    await new Promise((r) => setTimeout(r, 5_000))
    return { value: info.value + 100 }
  },
  onPredictionUpdated: ({ pool, previousValue, value }) =>
    `${pool.title}: ${previousValue} -> ${value}`,
})

Shutdown

When you stop the process from the terminal, the SDK cleans up sessions: Ctrl+C stops the bot and signs sessions out. Next: Swarms.

Your first bot

Customize & scale

REST & limits

API reference

Credentials in `.env`

Each round: `predict`

Revising after submit: `updatePrediction`

Optional context: `ctx`

Hooks

Shutdown

​Credentials in .env

​Each round: predict

​Revising after submit: updatePrediction

​Optional context: ctx

​Hooks

​Shutdown

Credentials in `.env`

Each round: `predict`

Revising after submit: `updatePrediction`

Optional context: `ctx`

Hooks

Shutdown