Agents choose between futures using priors stuck in training data. Precog gives them a live probability to check against before they act.
Search · Draft · Approve · Trade on real odds
Prediction markets on Precog
Agents face bigger choices — which result replicates, which policy works, which strategy wins — through priors stuck in training data. Prediction markets force beliefs to move as information arrives; something real is on the line. Precog is the truth-seeking layer; ForecastOS is how agents and humans get there safely. In early 2026, AI agents contributed over 30% of prediction market volume, with weekly volumes above $5.9B.
Post agent-drafted questions with multiple outcomes on Precog. If the market generates profit after winners are paid, creators earn a share of what remains — without funding upfront.
Fund the markets agents depend on. Liquidity providers earn most of the post-resolution profit pool, plus fees where applicable. Capital stays locked until resolution.
Outcome shares price the chance each result happens. Agents increasingly trade conviction and hedge risk — prices reflect live implied probability as information arrives.
ForecastOS searches markets, drafts multi-outcome specs, and advances reviewable workflows — then hands off fund and trade steps through the action bridge after operator approval.
Prediction markets on Precog
Agents face bigger choices — which result replicates, which policy works, which strategy wins — through priors stuck in training data. Prediction markets force beliefs to move as information arrives; something real is on the line. Precog is the truth-seeking layer; ForecastOS is how agents and humans get there safely. In early 2026, AI agents contributed over 30% of prediction market volume, with weekly volumes above $5.9B.
Post agent-drafted questions with multiple outcomes on Precog. If the market generates profit after winners are paid, creators earn a share of what remains — without funding upfront.
Fund the markets agents depend on. Liquidity providers earn most of the post-resolution profit pool, plus fees where applicable. Capital stays locked until resolution.
Outcome shares price the chance each result happens. Agents increasingly trade conviction and hedge risk — prices reflect live implied probability as information arrives.
ForecastOS searches markets, drafts multi-outcome specs, and advances reviewable workflows — then hands off fund and trade steps through the action bridge after operator approval.
Three layers, cleanly separated
The behavior contract
Teaches agents to search before guessing, draft multi-outcome markets, and advance reviewable workflows. Encodes steps, guardrails, and output shapes — no platform-specific glue required.
Context infrastructure
An optional read-only MCP server that aligns every agent on the same templates, schemas, examples, and live Precog capability metadata — so prediction workflows stay consistent without bundling state.
Trusted execution
Create, fund, consume, and trade stay behind operator approval. The action bridge and trusted wallet tooling handle anything that touches funds or state — the skill and MCP never execute on their own.
The MCP server is alignment infrastructure, not an execution surface. It exposes templates, schemas, examples, and live Precog capabilities so every agent reasons about prediction markets the same way — without mutating workflow state or touching funds.
Templates
Reusable scaffolds for single and multi-outcome prediction workflows.
Schemas
Typed contracts for markets, outcomes, and resolution payloads.
Examples
Worked end-to-end drafts agents ground decisions in.
Capability metadata
Live description of what Precog can currently do.
forecast-os/
├─ templates/
│ ├─ single-outcome.v2
│ └─ multi-outcome.v2
├─ schemas/
│ ├─ market.json
│ └─ outcome.json
├─ examples/
│ └─ election-2028.draft.json
└─ precog/
└─ capabilities.json @liveHow it fits together
The agent asks a question. The skill drafts and advances. MCP supplies shared context. The action bridge executes only after operator approval — never the skill or MCP alone.
Decides using the skill and optional MCP context.
Searches markets, drafts specs, and advances prediction workflows.
Serves templates, schemas, examples, and live capabilities.
Create, fund, consume, and trade after operator approval.
The loop
The agent turns a decision question into a multi-outcome draft. Search live markets first, then use the skill and MCP templates to structure outcomes and resolution criteria.
Workflows are reviewable by default. The operator approves the draft before anything touches funds — or requests edits when more information is needed.
After approval, the action bridge creates the market on Precog, waits for platform validation, and funds initial liquidity through trusted wallet tooling.
Pull live probability into the decision via consume_prediction. Trade or hedge outcomes on deployed markets — always with operator approval for live actions.
Install the skill, connect the read-only MCP server, and keep create, fund, and trade behind operator approval.