The risk layer agents can’t skip.
Pre-transaction risk infrastructure for AI agents operating on-chain. bitfence intercepts the transaction, evaluates the token across weighted on-chain risk categories, and returns a recommendation the agent can act on — in under 50ms for cached responses.
Rust-native backend with two-tier caching. Built for autonomous operation at agent speed.
Agent-native USDC micropayments on Base. No accounts. No API keys. No subscriptions.
Each signal scored independently with severity; every assessment carries a data-coverage confidence.
Each one a distinct loss mechanism — exit integrity, supply integrity, liquidity integrity, holder & insider risk, provenance, market integrity. All six live today.
What it is
bitfence is a token risk scoring oracle for autonomous agents. Before an agent signs a transaction, it asks bitfence about the token. The agent pays $0.003 USDC via x402 and gets back a score, a recommendation, and a reason.
The score is a weighted composite across six risk categories, each a distinct mechanism by which a position is destroyed — exit integrity, supply integrity, liquidity integrity, holder & insider risk, provenance, and market integrity. Circuit breakers override the composite when critical signals fire: measured threats like a honeypot or a confiscatory sell tax force a blocking score regardless of what the other categories say.
Scoring is maturity-aware and evidence-disciplined. Established tokens like USDC are scored differently from new launches carrying the same flags — a curated trusted-token registry anchors the difference. Honeypot status on EVM chains is measured by bitfence’s own transaction simulation, not bought as a vendor verdict, and deployer track record, insider cohorts, and manufactured-volume patterns all count today.
Three modes: a full assessment, a contextual check that weighs your position size against pool depth, and a quick check that returns a minimal go/no-go verdict — score, level, recommendation.
One query, one verdict.
Pick a token and replay a real assessment, exactly as the production API returned it.
How it works
01 — Intercept
The agent sends the transaction intent to bitfence before anything is signed. Ships as a tool the agent calls first — and as always-on middleware in supported frameworks (coming soon).
02 — Score
On-chain signals across six risk categories — each a distinct loss mechanism — evaluated, weighted, and scored into a composite 0–100. Cached responses return in under 50ms; the cold path is sub-second.
03 — Act
The agent receives a score, a risk level, a recommendation — PROCEED, REQUIRE_HUMAN_APPROVAL, or BLOCK — and a plain-language reason. On contextual checks, a suggested max position size too.
Live now
Put a risk layer in front of your agent.
Wire it in over HTTP, MCP, or Rig — or read the whitepaper first.