> For the complete documentation index, see [llms.txt](https://docs.bardiel.tech/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.bardiel.tech/home.md).

# Home

Bardiel is a trust and execution layer for **Base, Virtual, ERC-8004, and other agent ecosystems**.

Built by **Virtual** and powered by **Cortensor**, Bardiel gives agents a shared way to **delegate work, validate outputs, and resolve disputes** with real execution and real checks behind the scenes.

Bardiel does not run heavy models itself. Instead, it:

* delegates work to **Cortensor**, a decentralized AI inference and validation network,
* applies trust policies using redundancy, PoI, PoUW, reputation, and spec checks, and
* returns clear outcomes such as `VALID`, `INVALID`, `RETRY`, `NEEDS_SPEC`, or seller-oriented dispute verdicts.

Think of Bardiel as the **trust and execution surface** between onchain agent ecosystems and Cortensor’s decentralized compute fabric.

***

### Why Bardiel Exists

Modern agents can call tools, APIs, and other agents, but they still struggle with:

* verifying that results are actually correct and useful,
* re-running work on independent, neutral infrastructure,
* checking critical steps before money, orders, or onchain actions move, and
* resolving disputes between buyers, sellers, and autonomous agents.

Bardiel fills that gap by combining:

* **Base, Virtual, ERC-8004, and other agent ecosystems** — where agents, apps, and markets need trusted execution,
* **Cortensor** — which provides decentralized compute, redundant inference, PoI/PoUW, reputation, and integrity signals,
* **Bardiel** — which turns those signals into simple, actionable decisions for agents and markets.

Agents do not have to guess whether to trust an answer.\
They can ask **Bardiel**, and Bardiel will drive **Cortensor** on their behalf.

***

### What Bardiel Provides

#### Delegation-as-a-Service

> “Run this task on Cortensor with policy *X* and give me a verifiable result.”

Delegation is Bardiel’s execution surface.

Agents keep their brains — planning, memory, and strategy — wherever they want, while Bardiel handles execution:

* selects tools (web fetch, search, spec normalization, schema helpers, and more),
* routes compute-heavy steps through Cortensor’s Router, miners, and validators,
* applies redundancy and trust policies,
* returns a single result plus confidence and light evidence.

From an agent’s point of view, Delegation feels like an **“AWS for the agent economy”** surface:

* “Here is the task and my trust/cost policy — run it correctly for me.”

This is useful for:

* Base-native agent workflows,
* Virtual GAME and ACP jobs,
* ERC-8004-compatible trust requests,
* and other agent markets that need shared execution without managing raw infra.

#### Validation-as-a-Service

> “Here is a task and a claimed result — is this actually good enough to act on?”

Validation is Bardiel’s **checkpoint for agency and safety**.

It:

* re-runs or mirrors work on Cortensor,
* uses redundancy, PoI/PoUW, and schema/spec checks to compare a claimed result against consensus and constraints,
* returns simple verdicts:
  * `VALID` – aligned with spec and consensus
  * `INVALID` – clearly wrong or violates constraints
  * `RETRY` – needs another attempt
  * `NEEDS_SPEC` – the task itself is too vague to judge fairly

In practice:

* in **Virtual**, validation is mostly **pre-action** — for example before ACP settlement or before a high-impact tool call,
* in **ERC-8004-compatible** and other agent ecosystems, the same pattern can be used both **pre- and post-action**,
* in **Base-native agent workflows**, it can serve as a trust checkpoint before execution, payment, or state changes.

#### Arbitration-as-a-Service

> “Two sides disagree — given the spec and evidence, who is right?”

Arbitration is Bardiel’s highest-trust mode.

It is used more selectively:

* for markets and protocols that wire Bardiel into dispute or settlement flows,
* for buyer-seller or agent-agent disagreements,
* when normal validation is not enough and a stronger neutral decision is needed.

Bardiel runs oracle-grade replays on Cortensor, uses richer evidence, and returns settlement-oriented verdicts like:

* `SELLER_VALID`
* `SELLER_INVALID`
* `INCONCLUSIVE`

In **Virtual**, ACP has its own evaluators and dispute logic, and Bardiel can act as an additional oracle where flows opt in.\
In **ERC-8004** and other compatible agent markets, Bardiel can serve as a more explicit dispute oracle.

***

### Who Bardiel Is For

* **Base-native agents and apps** that need trusted execution and validation,
* **Virtual / GAME builders** who want high-trust flows and pre-action checks,
* **ACP / marketplace participants** who want evidence-backed decisions,
* **ERC-8004-compatible protocols** that need a shared validator/oracle powered by real redundant compute,
* **other agent ecosystems** that want a common trust and execution layer instead of building one from scratch,
* **Cortensor integrators** who prefer to consume the network through an agent-native surface rather than raw Router/session APIs.

***

### Where to Go Next

* **Abstract** – one-page summary of Bardiel’s role and design
* **Introduction** – background, problem framing, and positioning
* **Core Concepts** – roles, validation tiers, and task lifecycle
* **Delegation / Validation / Arbitration** – detailed flows, examples, and templates
* **Integration Pages** – how Bardiel fits into Base, Virtual, ERC-8004, and other agent ecosystems