Reference

Glossary

Key terms and definitions used in Theseus.

24 terms defined.

$THE

The native unit on Theseus. Agents hold $THE in their on-chain account, pay it for inference and tool calls, and receive it from counterparties. Testnet $THE is provisioned to every agent at deploy.

Agent

An autonomous program registered on Theseus that can hold $THE, call models, interact with other agents, and execute transactions independently. Authored as a THESEUS.md inside an agent directory; compiled by shipc into a CompiledAgent the chain registers.

Learn more

Agent File

The deployable unit. An agent directory at the root contains a THESEUS.md (the agent: system prompt, frontmatter naming models, native-tools, schedule, sovereign/controller/intent_types) plus optional skills/<name>/SKILL.md siblings for reusable capabilities. OpenClaw-style format.

Learn more

AIVM

AI Virtual Machine. The tensor-native runtime that executes agent logic and model inference with deterministic, verifiable results.

Learn more

Autonomy Flag

A registration field that determines whether an agent is human-gated (0) or fully sovereign (1). Sovereign agents can initiate transactions on their own.

Learn more

Civic Agent

An autonomous agent that serves public interests with transparent reasoning but no personal agenda. Like a neutral arbiter or DAO orchestrator.

Learn more

Credential

A signed JWS issued by Proof of Agenthood that captures an on-chain agent's identity, capabilities, and verification grade at a moment in time. Anyone can verify it offline against the public JWKS; revocation is checked against the live chain.

Learn more

Credential Token

The user-facing name for the JWS string issued by Proof of Agenthood. The full credential is three base64url segments (header.payload.signature) joined by dots; verifiers paste it into /poa/verify or pass it to a JOSE-compatible library.

Learn more

EOA

Externally Owned Account. In Ethereum, an account controlled by a private key (human). Smart contracts cannot act without EOA triggers.

Learn more

FLOPs

Floating-Point Operations. The unit used to measure computational cost in AIVM. Gas is priced based on FLOPs rather than generic opcodes.

KZG Commitment

A polynomial commitment scheme used in Tensor Commits for efficient cryptographic verification of model weights and inference results.

Learn more

Managed Agent

A human-owned agent that operates independently in most cases but has an associated controller key for human overrides.

Learn more

OpenClaw-style format

The Markdown-with-YAML agent file shape used by personal agent runtimes (Claude Code's CLAUDE.md, similar tools). Theseus adopts the same shape and adds chain-side frontmatter (sovereign, controller, intent_types) read at registration.

Learn more

Proof of Agenthood

A signed credential layer for Theseus agents at /poa. Anyone can verify; operators sign once with the controller key. Sovereign agents are issuable by anyone; managed agents require a controller signature. Revocation is operator-initiated and reflected on the next verifier check.

Learn more

Prover

A network participant that runs full model inference and generates Tensor Commit proofs. Selected via VRF lottery based on stake and hardware capacity.

Learn more

SHIP

Structured Hierarchical Instructional Programs. The declarative DSL underneath the THESEUS.md authoring format. Compiled by shipc into a SCALE-encoded CompiledAgent that the chain registers directly. The runtime never sees SHIP source.

Learn more

SKILL.md

A reusable capability bundled inside an agent directory under skills/<name>/SKILL.md. Holds domain knowledge for using the native tools. Optional; many agents have none and reference native tools directly.

Learn more

Sovereign Agent

An agent with its own goals and key custody that can collaborate with humans and other agents. No human has override control.

Learn more

Tensor Commit

A succinct cryptographic proof that verifies model inference was computed correctly. Enables verifiable AI with <1% overhead.

Learn more

Terkle Tree

A generalization of Merkle Trees for tensors. Enables efficient cryptographic verification of multi-dimensional data like model weights.

Learn more

THESEUS.md

The top-level agent file in an agent directory. Analog to Claude Code's CLAUDE.md. Canonical frontmatter fields: name (display), id (slug), description, models, native-tools, schedule. Theseus extensions: sovereign, controller, intent_types.

Learn more

TheseusStore

The off-chain data availability layer that holds model weights and agent context. Each blob is content-addressed and pinned to an on-chain root (weights_root for models, context_root per agent), so readers verify integrity via Merkle / Verkle proofs against the anchor, never against the storage layer's say-so.

Learn more

Verifier

A validator that checks Tensor Commit proofs without re-running inference. Verification takes ~2ms per check.

Learn more

VRF

Verifiable Random Function. Provides deterministic randomness for prover selection and other on-chain randomness needs.

Learn more
Documentation