Open-source human decision infrastructure

When your agent needs a decision, it calls you.

OpenConfer lets autonomous agents talk through a decision with you, return a confirmed structured outcome, and continue the original run.

View on GitHub Explore the decision loop

Self-hosted Model-agnostic Apache 2.0

agent / deploy-production

› Running release checks

✓ 184 tests passed

✓ Migration is reversible

DECISION REQUIREDShip this change to production?

Writes pause briefly during migration.

✓ Resumed with human decision

agent_waitingrun_7f31

OPENCONFER LIVE

YOU’RE LIVE 00:16

AGENT

I’m ready to deploy. Should I proceed?

YOU

Ship it behind the flag and monitor errors for ten minutes.

active → decision_confirmed

STRUCTURED OUTCOMECONFIRMED

{
  "decision": "approve",
  "conditions": [
    "feature_flag",
    "monitor_10m"
  ],
  "confirmed_by": "human"
}

result_delivered 200 OK

01 THE CATEGORY

The call is only
the interface.

OpenConfer is designed around a decision inside an existing autonomous workflow.

A voice-agent platform starts with a call and builds an assistant around it. OpenConfer starts with an agent that is already working—and gives it a standard way to reach a human without losing its thread.

Agent pauses→Human decides→Agent continues

02 YOUR STACK, YOUR RULES

Configure every layer
of the conversation.

OpenConfer stays open where it matters. Change the intelligence, the sound, the route, or the rules—without changing the decision protocol.

Any model

Choose the intelligence that facilitates the conversation—from open local models to hosted realtime systems.

Any voice

Bring the voice, speech pipeline, and language that make the interaction feel natural to you.

Any channel

Open an in-browser voice session, ring your phone, or add the transport your workflow needs.

Your policy

Control which agents may reach you, how urgently, and with exactly what context and permissions.

03 PROTOCOL + RUNTIME

One standard for
asking humans.

The protocol defines the request, context, states, and confirmed outcome. The runtime handles the live session, maintains continuity, and returns the answer to the exact run that asked.

Explore the open protocol

REQUEST

objective: "Ship release?"
urgency: high
channel: voice
continuation: run_7f31

OUTCOME

decision: approve
conditions:
  - feature_flag
confirmed: true
resume: run_7f31

created notified active confirmed acknowledged

04 MEET AGENTS WHERE THEY WORK

Built for the harnesses
you already use.

HermesAgent harness

↗

OpenClawAgent harness

↗

CodexCoding agent

↗

Claude CodeCoding agent

↗

MCPOpen protocol

↗

Your harnessCustom adapter

↗

05 HUMAN JUDGMENT, ON DEMAND

More autonomy.
Without guessing.

Ship or hold

An agent finishes a risky migration and calls for judgment before touching production.

Approval · 90 sec

Resolve an incident

Discuss the latest signals naturally, choose a response, and return a structured incident decision.

Incident · High urgency

Pick a direction

Talk through architecture tradeoffs without losing the context or momentum of the original run.

Decision · 3 options

Brief and continue

Let an agent summarize what changed, capture your guidance, and keep the workflow moving.

Briefing · Async

OPEN SOURCE · SELF-HOSTED

Give your agents
a way to reach you.

Clone the repository, run OpenConfer locally, and connect your first harness. Your sessions, providers, and policies stay under your control.

1 pnpm setup2 openconfer init3 openconfer serve

Quick start

$ git clone https://github.com/richardsondx/OpenConfer.git && cd OpenConfer && pnpm setup

View the repository

Node.js 20+ · macOS & Linux · Apache 2.0