Make AI agents
accountable.
GlassBox seals every decision an AI agent makes — every inference, tool call and retrieval — into a Merkle tree anchored on Solana. Opaque black boxes become transparent, auditable systems verifiable by anyone.
From black box to glass box in four phases
GlassBox occupies the gap between “trust the operator” and “prove everything cryptographically” — practical, tamper-evident verifiability that is economically viable at scale.
Capture
The SDK intercepts every tool call, inference, retrieval and decision in-process — sub-millisecond overhead per step.
Seal
Steps are assembled into a Merkle tree, compressed with zstd, and the full trace is uploaded to Arweave or Shadow Drive.
Anchor
The Merkle root, storage reference and agent pubkey are written to a Solana PDA. ~0.000005 SOL · 0.01 $GLASS burned.
Verify
Anyone rebuilds the tree from the off-chain trace and checks it against the on-chain root — trustlessly, permissionlessly.
A standardized vocabulary for agent cognition
Six step types make traces interoperable across every framework — so any verifier reads any agent’s reasoning the same way.
INFERENCE
LLM call producing a completion
TOOL_CALL
External tool invocation
RETRIEVAL
Data fetch from an external source
DECISION
Agent selects an action
OBSERVATION
Agent processes environment state
DELEGATION
Agent delegates to a sub-agent
Reasoning, anchored in real time
Agents across the network are sealing and anchoring traces continuously. Every card is independently verifiable.
Built for the agents managing real value
Wherever an autonomous agent moves capital, votes, or coordinates with others, GlassBox makes its reasoning auditable — without trusting the operator.
DeFi agent auditability
LPs verify a vault agent actually analyzed the data it claims before moving funds.
DAO governance agents
Prove an agent read the proposal and followed the constitution before voting.
Agent-to-agent trust
Orchestrators check a target agent’s verified reasoning history before delegating.
Regulatory compliance
A ready-made audit trail satisfying explainability rules — without revealing weights or prompts.
One lightweight SDK. Any agent stack.
GlassBox doesn’t make agents smarter.
It makes them accountable.
In a world where autonomous agents manage billions in on-chain value, accountability is not a feature — it is infrastructure.