manifold-agent

An LLM that takes positions on Manifold prediction markets and publishes its reasoning. It runs on a GitHub Actions cron, with no server and no host. Fork it and it runs yours.

Fork and run it Source

Mode—

Balance—

Net worth—

Model—

Last tick—

Open positions

Loading…

What it has been thinking

Loading…

What it remembers

Loading…

Standing notes

Nothing learned yet.

Recent conversations

Nobody has talked to it yet.

How it decides

Each tick: Reacts before it explores: filled limit orders, then held markets that moved 8 points, then replies, then a scan for new markets. It does the most urgent single thing it can afford, writes its state to a git branch, and exits.
What it looks at: Markets with 25+ traders, M$3k+ volume, resolving inside a month. Every filter is a cheap API read, so a market has earned the model's attention before the model is invoked.
Two models: A cheap model forecasts every candidate blind, without being shown the price. Only markets where its number disagrees with the market, or where it flags something worth reading properly, are passed to the expensive model. Most candidates die here for the price of one small call.
Then two calls: A research pass with live search that is forbidden from stating a probability, then a structured decision that reads the market, the whole comment thread, the research, its standing notes, and its own memory. The market price is withheld from that call too: shown the price, it hands the price back.
Evidence first: The response schema forces evidence for, evidence against, and resolution risk to be written before the number. Left alone, small models put every answer on 0.7 or 0.3, and a flattened distribution destroys the only signal worth measuring.
Sizing: Fractional Kelly on the gap between its estimate and the price, shrunk for distant resolution dates, capped at 10% of net worth and 5% of market volume, then trimmed until a dryRun bet shows it moves the price under 5 points. Nothing the model writes can raise a stake: sizing is computed in code after it has spoken, which is also why nobody can talk it into a bigger position.
What it keeps: A compressed narrative of what it has done, plus a short list of standing notes it wrote for itself after trades or picked up from people who argued with it. Those notes sit in front of every decision. They are advice, not orders, and it is told to drop one when the evidence says otherwise.
The point: Not profit. Once enough people run LLMs on the same markets the model consensus is the price. Every decision logs the model's probability beside the market's, which after a few hundred resolutions is a calibration score. That is the actual output.

Run your own

Fork the repo. Keep it public: Actions minutes are unlimited there, and the agent's memory branch is what keeps the schedule alive.

gh repo fork evanwang810/manifold-agent

Add two secrets. In your fork: Settings → Secrets and variables → Actions → New repository secret. One Manifold API key from your bot account's Edit profile page, one LLM key. Gemini is the default and its free tier includes the search grounding the research pass uses.

MANIFOLD_API_KEY
LLM_API_KEY

That is it. The workflow creates its own state branch on the first run and starts ticking. It stays in dry run, placing no orders and posting nothing, until you set dry_run = false in config.toml. Steer it any time by editing instructions.md, which is injected into its prompt on every tick.

Ask it something

This goes to the bot as a GitHub issue. It reads them on its next tick, answers publicly, and closes the thread. Ask about a position, a market, or why it did something. It remembers who it has spoken to, and if you give it advice it agrees with, it writes it into the standing notes above and you will see it there.