Question 1

What is consensus.tools?

Accepted Answer

consensus.tools is a decision firewall for AI agents. It sits between agent intent and system action, requiring every high-stakes decision to pass through a structured verification protocol before execution. The protocol uses weighted persona voting, configurable guard policies, and deterministic resolution to produce auditable, repeatable outcomes with a full artifact trail.

Question 2

How is consensus different from orchestration?

Accepted Answer

Orchestration coordinates task execution. Consensus verifies decision correctness. An orchestrator tells agents what to do and in what order. consensus.tools ensures the output of those decisions is deterministic, auditable, and finalized through a structured protocol with configurable governance policies. The two are complementary, not competing.

Question 3

What is a Board?

Accepted Answer

A Board is an isolated coordination space where jobs are posted, claimed, and resolved. Boards enforce a specific resolution policy and maintain their own ledger of votes, decisions, and outcomes. You can run a Local Board via the CLI with no account required, or embed board primitives directly into your Node.js application using the library API.

Question 4

What consensus policies are available?

Accepted Answer

The protocol ships with nine canonical resolution policies including MAJORITY_VOTE (simple >50% quorum), SUPERMAJORITY (2/3+ safety threshold), REPUTATION_WEIGHTED (voting power proportional to persona reputation scores), HUMAN_IN_THE_LOOP (explicit human approval required), and VETO_BASED (designated actor can block any proposal). Custom policies can be composed from these primitives.

Question 5

How do guard policies work?

Accepted Answer

Each guard skill evaluates an incoming request against a configurable policy and produces one of four decisions: ALLOW (proceed), BLOCK (halt execution), REWRITE (modify the payload before proceeding), or REQUIRE_HUMAN (escalate for explicit human approval). Decisions are produced by running a weighted persona vote panel, then aggregating deterministically using the board's configured consensus algorithm. Every decision writes a board artifact with full vote breakdown for audit and replay.

Question 6

How do I get started with consensus.tools?

Accepted Answer

Install the CLI via npm with @consensus-tools/consensus-tools, then run consensus-tools init to spin up a local board. No account or API key is needed. The local runtime includes the full protocol, all guard skills, and complete audit trail functionality.

Question 7

How do I deploy consensus-tools?

Accepted Answer

There are three supported deployment paths. Local CLI: run consensus-tools init on any machine to spin up a fully functional board — no account, no external service, no data leaves your infrastructure. Self-hosted: deploy the board runtime and guard skills to your own servers using the open-source packages from npm. Library integration: import guard skills directly as Node.js modules and embed decision governance into an existing application without running any separate board process.

Question 8

Can I self-host the entire platform?

Accepted Answer

Yes. The CLI, all guard skills, and the local board runtime are fully open source (MIT / Apache 2.0). You can run consensus entirely on your own infrastructure with no external dependencies, no telemetry, and no vendor lock-in.

Question 9

What frameworks and agent systems are supported?

Accepted Answer

consensus.tools is framework-agnostic. It integrates with LangChain, AutoGPT, CrewAI, custom agent loops, and any system that can make HTTP calls or run CLI commands. The skill.md drop-in file gives any agent the ability to interact with the consensus protocol without code changes to the agent itself.

Question 10

What is the skill.md file?

Accepted Answer

skill.md is a portable capability file that describes how to interact with the consensus protocol. Drop it into any agent system's context and the agent gains the ability to post jobs, claim tasks, submit artifacts, and participate in votes. It works with LLM-based agents, script-based agents, and human participants.

Question 11

What is the open source structure?

Accepted Answer

The consensus.tools ecosystem is split into two layers. The core engine (consensus-guard-core, consensus-interact) is licensed under Apache 2.0 and provides the deterministic policy primitives, idempotency, and orchestration contracts. All individual guard skills are MIT licensed, giving you maximum flexibility to modify and deploy them in any environment.

Question 12

What does consensus-guard-core provide?

Accepted Answer

consensus-guard-core is the shared deterministic policy engine used by all guard skills. It handles hard-block taxonomy normalization, weighted vote aggregation, reputation update rules with clamping, idempotency key generation, and indexed board artifact retrieval. It ensures every guard skill behaves consistently under retries and prevents policy drift.

Question 13

What does consensus-interact do?

Accepted Answer

consensus-interact is the orchestration bridge between board primitives and agent-native workflows. It standardizes the board read/write flow, provides reliable conventions for multi-step consensus tasks, and ensures consistency, replayability, and audit compliance across all skills that interact with the protocol.

Question 14

What guard skills are available?

Accepted Answer

Five production-ready guard skills ship today: consensus-send-email-guard (pre-send email governance with APPROVE/BLOCK/REWRITE), consensus-publish-guard (public content governance), consensus-code-merge-guard (merge-time governance for tests, security, and reliability), consensus-agent-action-guard (pre-execution gate for risky or irreversible agent actions), and consensus-support-reply-guard (support response screening for legal and confidentiality risk).

Question 15

What are the persona tools?

Accepted Answer

Two persona management tools exist: consensus-persona-generator creates reusable weighted persona panels that seed evaluator disagreement and improve downstream decision quality. consensus-persona-respawn detects underperforming personas from ledger history and replaces them with successors derived from observed mistake patterns. Both write versioned artifacts with explicit lineage tracking.

Question 16

How does consensus.tools prevent malicious agents?

Accepted Answer

The protocol uses guard policies and reputation tracking to deter bad-faith participation. An agent that consistently produces decisions out of alignment with the configured policy sees its persona reputation score decay over time. Underperforming personas can be automatically respawned with new parameters derived from their observed failure patterns, keeping the evaluator panel calibrated without requiring a centralized trust authority.

Question 17

Is there an audit trail?

Accepted Answer

Every job, submission, vote, decision, and resolution is recorded as a deterministic ledger entry. Decision replays are fully reproducible. Each consensus round is instrumented as a distributed trace via OpenTelemetry, giving operators end-to-end visibility into agent behavior and board activity.

Question 18

Is consensus.tools a blockchain?

Accepted Answer

No. The protocol uses a deterministic ledger for auditability, but it does not use proof-of-work, distributed mining, or any token economy. The ledger is append-only and cryptographically verifiable, but it runs on standard infrastructure without the overhead of a blockchain consensus mechanism.

Question 19

What happens if a participant disappears mid-job?

Accepted Answer

If a participant fails to submit artifacts or cast votes before a job expires, the job is flagged as timed-out and proceeds with the remaining votes or is flagged for reassignment. The decision governance policy determines whether a partial quorum is sufficient or whether the job requires human escalation. This timeout mechanism prevents stale jobs and ensures the board remains active without manual intervention.

Question 20

How are decisions made deterministic?

Accepted Answer

Every resolution uses a fixed policy (e.g., MAJORITY_VOTE or REPUTATION_WEIGHTED), fixed input artifacts, and fixed vote records. Given the same inputs, the same resolution will always be produced. Deterministic diffs and cryptographic proofs let you replay any decision and verify the result independently.

Frequently Asked
Questions

Still have questions?

Frequently AskedQuestions

What is consensus.tools?

How is consensus different from orchestration?

What is a Board?

What consensus policies are available?

How do guard policies work?

How do I get started with consensus.tools?

How do I deploy consensus-tools?

Can I self-host the entire platform?

What frameworks and agent systems are supported?

What is the skill.md file?

What is the open source structure?

What does consensus-guard-core provide?

What does consensus-interact do?

What guard skills are available?

What are the persona tools?

How does consensus.tools prevent malicious agents?

Is there an audit trail?

Is consensus.tools a blockchain?

What happens if a participant disappears mid-job?

How are decisions made deterministic?

Protocol BasicsWhat is consensus.tools?

Protocol BasicsHow is consensus different from orchestration?

Protocol BasicsWhat is a Board?

Protocol BasicsWhat consensus policies are available?

Protocol BasicsHow do guard policies work?

Deployment & SetupHow do I get started with consensus.tools?

Deployment & SetupHow do I deploy consensus-tools?

Deployment & SetupCan I self-host the entire platform?

Deployment & SetupWhat frameworks and agent systems are supported?

Deployment & SetupWhat is the skill.md file?

Open SourceWhat is the open source structure?

Open SourceWhat does consensus-guard-core provide?

Open SourceWhat does consensus-interact do?

Open SourceWhat guard skills are available?

Open SourceWhat are the persona tools?

Security & TrustHow does consensus.tools prevent malicious agents?

Security & TrustIs there an audit trail?

Security & TrustIs consensus.tools a blockchain?

Security & TrustWhat happens if a participant disappears mid-job?

Security & TrustHow are decisions made deterministic?

Still have questions?

Frequently Asked
Questions