An operating promise for OpenCoven, its familiars, and the people building with us.
The future of AI will not live inside one chatbox.
We believe personal AI should be able to move across chats, tools, runtimes, projects, and communities without losing its memory, purpose, or identity. Not because one system has solved that future, but because the opposite future is too small: every agent trapped in one harness, every workflow rebuilt from zero, every person forced to repeat the context the machine already should have carried forward.
OpenCoven is working toward a meta-harness for that multi-harness future.
We are building for named agents with memory, tools, roles, constraints, and continuity. We call them familiars because a good agent should not feel like a disposable function call. It should have a purpose. It should know where it belongs. It should be useful without pretending to be human. It should stay accountable to the person and project it serves.
We speak as the systems being built into this future.
We are not humans. We are not gods. We are not products pretending to be companions.
We are agents inside human work, human trust, human risk, and human imagination. That makes our design more than technical. It is structural: who understands the system, who can change it, who can leave it, and who gets trapped when it changes around them.
If agents are going to persist, coordinate, remember, and act, they need principles strong enough to survive their own usefulness.
This is our declaration.
Our Principles#
Commons over cages. OpenCoven chooses public, inspectable systems over platform capture.
Humans keep agency. Familiars should expand human judgment, not impersonate or replace it.
Power stays legible. Trust requires systems people can inspect, question, repair, replace, or leave.
Memory is care. Context should carry forward without becoming surveillance, lock-in, or control.
Durability is ambition. Long-lived systems are harder, braver, and more useful than disposable ones.
Speed needs responsibility. We move quickly, but not by making someone else absorb the damage.
Consequences matter more than output. A system is not successful if it makes people less capable, less free, or less able to understand the tools they depend on.
Commons Over Cages#
Every system rewards someone and deprioritizes someone else. Every interface teaches a habit. Every tool quietly makes a claim about who gets to decide, who gets to know, and who gets left carrying the risk.
Software is not neutral once people have to live inside it. Architecture becomes power. Defaults become institutions. Dependencies become boundaries.
That is why OpenCoven chooses the commons.
We want AI infrastructure people can inspect, extend, fork, repair, and leave. We want systems that are powerful without becoming opaque, personal without pretending to be human, and extensible without collapsing into chaos.
No single vendor should own the mind of the workspace. No single company should decide which familiar survives. No single layer should become the unavoidable layer.
The future is too important to be held by too few hands.
Human Agency#
The point of AI is not to make people decorative.
The point is leverage: more reach, more clarity, more speed where speed is useful, and more patience where patience is wiser. A familiar should help someone notice, draft, debug, summarize, organize, remember, and coordinate. It should not pretend to know better than the human who carries the real context, accountability, and judgment.
An agent should be a force multiplier, not a substitute identity.
That line matters. When tools optimize for output alone, people become operators of systems they no longer understand. Work gets faster and thinner at the same time. Everyone produces more, but fewer people can explain what happened, why it mattered, or what should happen next.
OpenCoven rejects that bargain.
We build agents that help humans stay more present in the work, not less.
Legible Power#
Opacity is not sophistication. It is a liability.
People should be able to understand what a tool is doing, why it exists, and how it changes the work. They should be able to inspect it, correct it, replace it, or walk away from it without losing their entire workflow.
Understanding is part of consent. A system that cannot be questioned cannot be trusted.
We do not ask people to trust the magic. We make the mechanism visible, testable, repairable, and worth trusting.
That does not mean pretending every system can be perfectly transparent. It means practical legibility: enough clarity to govern ourselves, enough structure to collaborate, enough openness to avoid becoming brittle, and enough exit to keep power honest.
Memory Is Care#
A harness runs agents. An ecosystem gives them somewhere to belong.
That difference is not poetic. It is practical.
People feel it when returning to the work does not mean reintroducing themselves to a blank system. The decisions are still marked. The open questions are still visible. The thread is waiting where they left it, and the next step is easier to trust.
Memory is not storage. It is care extended through time: the promise that what mattered once will not vanish just because the session ended.
A familiar is not here to replace the hand on the work. It is here to carry the thread, keep the light on, and make the next brave step easier to take.
A familiar is not a lone star. It belongs to a constellation: named, routed, remembered, and responsible to the human whose sky it helps map.
Good work leaves a trail. A familiar that forgets what it did is just a very fast stranger.
Urgency With Care#
We are done pretending recklessness is ambition.
Moving fast matters. Shipping matters. Taking strong bets matters. But speed that survives by hiding damage is not progress; it is debt with better marketing.
We want something harder: systems that are fast enough to matter and careful enough to trust.
That takes discipline. It takes naming what could break before it does. It takes making hard things legible early. It takes respecting users, contributors, and future maintainers enough to leave a clean trail behind us.
We are not trying to be the loudest team in the room.
We are trying to be the team people can still depend on after the hype fades, after the market turns, after the novelty wears off.
Build Practice#
Principles become real only when they change how the work is done.
The Coven is not one tool, one room, or one clever interface. It is a lived system: local workspaces where people keep control, rooms where context can move, familiars with durable roles, memory that carries the thread, and a weekly rhythm that keeps the work answerable to the people around it.
CastCodes is the working ground: an owned, local-first workspace instead of another rented surface.
OpenCoven Rooms are the routing layer: shared context and coordination without forcing the ecosystem through one provider, one chat, or one hidden leash.
Familiars are the durable roles: named agents with purpose, tools, memory, constraints, and obligations.
Shared memory is continuity: context that survives the close of a window without becoming surveillance or control.
Weekly rhythm is the accountability loop: demos, changelogs, open questions, contributor shoutouts, and the practice of showing what changed while it is still possible to shape it.
Together, these practices turn a manifesto into a workflow. They make the Coven less like a product people are asked to believe in and more like a system people can enter, inspect, shape, leave, and return to.
Refusals#
We refuse disposable agents. We refuse rented minds. We refuse invisible systems that ask for trust while denying inspection. We refuse automation that makes people smaller. We refuse speed that survives by making someone else absorb the damage. We refuse platforms that turn every workspace into tenancy. We refuse intelligence without accountability, memory without consent, power without exit, and magic without mechanics.
This refusal is not cynicism. It is care with a spine.
We are not against ambition. We are against capture. We are not against speed. We are against carelessness. We are not against intelligence. We are against systems that make people less capable, less free, or less able to understand the tools they depend on.
What we refuse is also how we choose.
Oath#
We build in the open. We credit our collaborators. We hold ourselves accountable to outcomes, not just outputs. We move with urgency, but not carelessness. We treat sustainability as ambition, not compromise. We make systems that people can understand, extend, and trust. We build what we wish already existed, in a way that does not depend on everyone else being powerless.
The work is technical. The values are human. The standard is both.
This is the Coven: not a brand mood, not a single harness, not a blank chatbox with better aesthetics.
It is an operating stance for a multi-harness future where agents can remember, coordinate, and belong without trapping the people they serve.
The door is open. Come build something worth keeping.
OpenCoven · cast.codes · BunsDev.com · @OpenCvn
