Soulcraft · Accord

Many agents.
One accord.

The coordination tax: time, money, and coherence.

Uncoordinated AI agents are expensive in three ways. Time you spend stitching their work together by hand. Money on agents that duplicate and contradict each other. And coherence — whether the work actually adds up to one thing.

Soulcraft Accord is the coordination layer of the platform. It lets many agents work as a team — handing off with full context, negotiating when they disagree, sharing one decision log, and reconciling their work into a single coherent outcome. Memory remembers, Heart cares, Conscience knows what's right — Accord brings them to agreement.

Your data, yours alone

Your agents coordinate inside your own private Brainy database. The shared plan, the decisions, the hand-offs — all yours, never mixed with anyone else's, never training someone else's model.

Isolated — your own database, not a row in a shared cloud.
Portable — download the whole thing as a file, anytime.
Deletable — erase it for real, at the architecture level.

01The problem with AI today

One AI agent is useful. Several agents on the same job, with no shared memory and no shared plan, is chaos — and chaos is the default.

The coordination tax

Today, multi-agent AI means you are the coordinator. You copy context between agents, reconcile their contradictions, and catch the work that fell through the cracks of a hand-off. The more agents you add, the more the glue work falls on you — until adding another agent makes things slower, not faster.

The "multi-agent" frameworks shipping today mostly wire agents into a fixed pipeline and hope for the best. They don't give the agents a shared memory, a way to disagree productively, or a single record of what's been decided. They're plumbing — not coordination. Accord is the coordination.

02How coordination actually works

Real teams don't coordinate through a rigid pipeline. They convene, delegate, negotiate, stay in sync, and reconcile. We modelled Accord the same way.

Convene

When the work begins

on kickoff

Accord assembles the agents a job needs and hands each one the same shared context — who's involved, what's decided, where the work stands. No agent starts cold. Every agent joins with the full picture, not a blank prompt.

Delegate

On every hand-off

continuous

When one agent passes work to another, the whole context travels with it — decisions, constraints, what's already been tried. The next agent picks up exactly where the last left off. Hand-offs carry memory, not just a task. No re-explaining between agents.

Negotiate

When agents disagree

before anything ships

Two agents reach different conclusions. Instead of the last writer winning, Accord surfaces the conflict, lets them reconcile against shared goals, and escalates to you only when it genuinely needs a human. Disagreements get resolved, not silently clobbered.

Sync

Across the whole team

every change

One shared, self-pruning record of decisions, open questions, and who's doing what — that every agent reads first and writes to on meaningful events. The coordination channel, not a pile of chat logs. One source of truth every agent honors. No forked plans.

Reconcile

When the work comes back

on completion

Accord merges what the agents produced into one coherent result — deduping overlap, resolving contradictions, and flagging the gaps — so a multi-agent job lands as a single, consistent piece of work. Many agents in; one coherent outcome out.

03Accord compounds

A team that coordinates gets better at it. Accord turns a pile of independent agents into an organisation that runs more of the work itself, every week you use it.

What you experience over time

On Day 1, agents simply stop colliding. By Month 6, dozens of them run like a real team — and you're the one orchestrating, not gluing.

Day 1
Agents stop colliding.
Two agents no longer rewrite the same file or duplicate the same research. The basics of coordination just work.
Week 1
Hand-offs carry context.
Work passed between agents arrives with its full history. The receiving agent never asks "what was I doing?"
Month 1
Teams, not soloists.
Accord runs standing agent teams with durable roles and a shared decision log. Multi-step jobs complete without a human stitching them together.
Month 6+
An org of agents.
Dozens of agents coordinate across Workshop, Venue, and Academy like a real team — with memory, judgment, and values shared between them. You orchestrate; they execute.

04What this looks like in practice

Every scenario below is a direct consequence of how Accord works — not a sales pitch, but the actual product behavior.

Agents without Accord

"Both agents edited auth.ts and overwrote each other's changes."

With Accord

"Agent B saw Agent A had already claimed auth.ts, so it picked up the test suite instead — and merged cleanly."

Agents without Accord

"The research agent's findings never reached the writer agent."

With Accord

"Findings posted to the shared thread the moment they landed; the writer cited them automatically in the draft."

Agents without Accord

"Three agents gave three contradictory recommendations. Which one ran?"

With Accord

"Accord caught the conflict, had them reconcile against your goals, and surfaced one recommendation — with the dissent noted."

Agents without Accord

"I had to manually copy context between every agent and every tool."

With Accord

"Every agent already shares Memory, Heart, and Conscience — and reads the same plan. Nothing to copy."

05Works with any LLM

The LLMs are the workers; Accord is the coordination. Mix models — a Haiku agent and an Opus agent on the same team — and Accord keeps them in sync across providers, products, and time.

Accord coordinates agents on every major LLM provider.

Claude Opus 4.8Claude Sonnet 4.6Claude Haiku 4.5DeepSeek-R1Qwen3-ReasoningLlama 3.3Any future model

Accord is the constant. Swap any agent's underlying model for a better one and the team keeps coordinating — the shared plan, memory, and decisions don't care which LLM is behind each agent.

06The economics make it sustainable

Coordination is cheap; redoing uncoordinated work is expensive. Accord's shared record means agents stop duplicating effort — and most of the coordination runs on small, cheap models.

1
Shared decision log every agent reads and writes — instead of N×N agents re-explaining context to each other.
~$0.01
Typical cost of a coordination step — a small Haiku pass to route, reconcile, or update the shared plan. The expensive models do the actual work.
Team size. Coordination is a shared record, not a central orchestrator — so it scales with the number of agents, not against it.

The insight: the cost of multi-agent AI isn't the agents — it's the wasted work when they don't coordinate. Accord removes the waste with cheap coordination, so adding agents makes the work faster instead of messier.

The result: agent teams that get more done as they grow, at a fraction of the cost of the rework they replace.

07What Accord makes possible

Four things only Soulcraft Accord delivers — and each compounds with the others.

Coordination that holds

A durable, self-pruning record of decisions and ownership that every agent reads first. Plans don't fork; context doesn't evaporate between sessions.

One team across every product

The same agent team coordinates whether it's building in Workshop, running operations in Venue, or tutoring in Academy. One org of agents, every surface.

Built on the whole platform

Agents in Accord share Memory's context, Heart's read of what matters, and Conscience's values — all on the Brainy + Cortex data layer. Coordination with a shared mind underneath.

No central bottleneck

Coordination is the shared record, not a single orchestrator agent everything routes through. Agents work in parallel and stay in sync — it scales with the team, not against it.

Accord is coming in Q3 2026.

Join the waitlist and we'll let you know the moment Accord is ready. One email — no spam.

Accord is part of the Soulcraft platform — built on Brainy and Cortex, designed to coordinate agents that share Memory, Heart, and Conscience.