Orchestrate every model.
Preserve every memory.

AICoven is a multi-model AI workspace with shared memory and bring-your-own-keys. Connect GPT, Claude, Gemini, and Mistral inside one encrypted vault — your keys, your memory, your budget.

Join the Beta

“Any sufficiently advanced technology is indistinguishable from magic.” — Arthur C. Clarke

Why AICoven

One vault instead of three subscriptions.

Stop renting fragmented copies of the same conversation. Bring your own keys, keep your memory, and route to whichever model is best for the task.

Before

· $720/year across redundant Pro plans
· Conversations fragmented across providers
· Memory resets between sessions
· Vendor lock-in for keys, history, exports

With AICoven

· Pay raw API costs (~$5/month for heavy usage)
· Quota-aware fallback across every provider
· Persistent, searchable memory fabric
· Bring-Your-Own-Keys; export anything, anytime

Roles, in plain JSON

role_definition.json

{
  "role": "Lead Architect",
  "primary_model": "claude-3-5-sonnet",
  "fallback": "gpt-4o",
  "context_window": "200k",
  "memory": "persistent",
  "encryption": "AES-256-GCM"
}

Each role gets its own preferred model, fallback, autonomy level, and encrypted memory scope.

Inside the circle

Encrypted API key vault inside the iOS app — Your keys are encrypted at rest. Only you hold the decryption key.

Memory graph showing persistent context — Pin critical context, prune irrelevant data, search across covens.

Token budgets per role and per session — Hard caps per turn, role, or session. No surprise bills.

Custom agent role configuration — Roles with preferred models, prompts, and tool access.

Multi-model orchestration on macOS — Switch between GPT, Claude, and Llama inline. Quota-aware fallback.

Real-time context window visualisation — See exactly which messages and memories are in context.

Encrypted chat history and audit trail — End-to-end encrypted history. Auditable tool calls.

Dynamic tool loading per turn — Agents load only the tools they need; lower token overhead.

What makes it feel like magic

Unified orchestration

Route prompts through any provider with quota-aware fallback.

Persistent shared memory

Recall work across threads, roles, and devices, with approval workflows.

BYOK security

Your keys never leave your vault. Encryption and audit trails by design.

Collaborative covens

Shared workspaces with memory, roles, budgets, and audit breadcrumbs.

Multiple agent roles

Each role gets its own model, autonomy level, max tokens, and prompt.

Encrypted by your keys

Chats and memories are encrypted with your key. No plaintext storage.

Open Source

AICoven Local — free and open source

Run AICoven entirely on your own machine. The local edition is fully open source — no cloud account, no telemetry, no limits. Bring your own API keys and orchestrate every model with complete privacy.

View on GitHub

100% local

All data stays on your machine. No cloud backend required.

Community driven

Contribute features, report bugs, shape the roadmap.

All features unlocked

Covens, memory, multi-model orchestration — every feature.

Orchestrate every model.Preserve every memory.