TokonoMix provides the engineering and commercial infrastructure to build, operate, and deliver enterprise level AI solutions. Explore the full architecture, with production ready components clearly identified.
Two paragraphs and six words. If you read nothing else on this page, read this.
Organisations can buy individual AI models today, but not the engineering stack needed to build, operate, and scale reliable AI products. A model is a capability; a product is an architecture.
Modern AI should not be a monolithic platform. It should be built from independent layers with clear responsibilities, evolving separately while working together through open interfaces. TokonoMix provides this stack, along with the commercial infrastructure to operate and scale what you build.
One stack builds the capability. The other turns it into a product you can sell. Each has a distinct role.
Open technologies for reliable AI systems. Independent by design.
Commercial platforms built on those AI capabilities.
Transport carries messages, coordination gives them meaning, execution does the work, review checks it, commerce sells it, applications use it. Each layer owns one role.
Tokonomix is the gateway: one key and one base URL that reach every model in the catalogue, with Council’s multi-model review available as a model id on the same call. It carries and meters the traffic — it does not decide anything itself. ZelixAI buys its models here, including the EU-routed lane, and the components below are what the traffic is made of.
No component is dressed up. The badge on each card is its real maturity — that is how the finished parts stay believable.
A long agent run stops at the first question it cannot answer, and sits there for hours. ANS keeps the run going and parks the question for you.
Run long AI workloads unattended without blocking on human questions.
TodayClone the repository and run it — open source, and the engine behind this site’s own unattended work.
One model can be confidently wrong and nothing in its answer says so. Council asks models from different families the same question separately, then shows where they disagree. That disagreement is the signal.
Independent multi-model review: fewer single-model errors, disagreement surfaced early. It does not guarantee correctness.
TodayAdd the MCP server to your agent, or call it over this site’s API with the model id tokonomix-consensus.
Every agent describes project state in its own words, so nothing downstream can read it reliably. DCP fixes the vocabulary, not the tool — which is why it is a specification, not a library.
Development Coordination Protocol — a vendor-neutral language for project state, shared by AI agents and humans.
TodayRead the specification at devcopro.org and build against it — implementations follow the spec, not the other way round.
One machine runs out of capacity before the work does. AgentsWeaver spreads a job across a fleet of specialised agents, keeps them isolated, and settles what each one used.
Distributed collaboration for AI agents: workload distribution, isolation and settlement across a fleet.
TodayNothing self-serve yet: access runs through a direct relationship while the private preview continues.
When agents message each other, delivery is the easy part. The hard part is knowing who really sent it, and treating the message as data rather than an order. AgentixMesh handles both.
Secure transport, trust, routing and discovery for distributed AI agents.
TodayThe repository is public; the hosted network is still in development.
beheer-v3 is the commercial layer the applications run on — not an application, and not an AI component.
Building an AI feature and selling it are different problems. beheer-v3 handles the selling: who subscribed, what it costs, and what they get when they pay.
Subscriptions, billing, resellers, licensing and provisioning. Not AI — it sells AI and digital services.
TodayA commercial platform behind the products below, not a self-serve component.
A customer-service AI that buys its models through the gateway and its billing from beheer-v3, instead of rebuilding either.
Enterprise AI for customer service, built on the stack — its models come through the Tokonomix gateway, including the EU-routed lane.
TodayLive at zelixai.ai — the sister product, and the reference implementation of the stack.
Media creation where the selling side — catalogue, licences, payment — comes from the layer underneath.
Media creation and commerce, built on the stack.
TodayIn development; the product page is not confirmed yet.
A local marketplace that gets its AI and its billing from the same two stacks as everything else here.
A local digital economy, built on the stack.
TodayIn development; the product page is not confirmed yet.
The same path a real customer question takes. Each step is a component you can open above.
Applications like Gallery and ZelixAI consume these capabilities without hard dependencies — composability, not coupling.
Everything on this list works today, with the account you already have. Nothing here is a waitlist.
https://mcp.tokonomix.ai/mcpAdd it in claude.ai under Settings → Connectors as a custom connector. You log in with OAuth; no client id, no secret, no key in your config.
Full connector guidenpx tokonomix-council-mcpThe local transport, driven by an API key. It also carries what the remote one cannot: image input and the full council mode.
Package on npmhttps://tokonomix.ai/api/v1An OpenAI-compatible base URL: change two lines in an existing client. Send model: "tokonomix-consensus" and several models answer instead of one.
Endpoints and agent setupGET /api/v1/modelsEvery model the gateway reaches, with its prices and whether it can be routed inside the EU. Public JSON, no key needed.
Open the JSONgit clone https://github.com/TokonoMix/agents-never-sleepThe unattended-execution engine, open source. It parks the questions a run cannot answer alone instead of waiting on them.
ANS on GitHubPut independent multi-model review in front of the decisions your agent cannot take back.
Add the Council MCP server to your agent — one URL, no install.
One OpenAI-compatible base URL reaches every model in the catalogue, plus the consensus model id.
Point your existing OpenAI client at https://tokonomix.ai/api/v1.
Subscriptions, licensing and provisioning are a layer, not a feature you write again per product.
Read what beheer-v3 owns, and where your product would sit on top of it.
Start with the components that are already public — ANS and Council — and follow the rest as they open up.
Open the ANS repository and run the acceptance suite.
Not a timeline — a direction. Each stage builds on the last, and no stage has a date it could miss.
These are constraints we accepted, not adjectives. Each one costs something somewhere else in the system.
Each layer owns exactly one job.
No lock-in to one model or provider.
Components compose, they don't entangle.
Replace a layer without breaking its consumers.
Protocols over proprietary glue.
Each product stands on its own.
Separate repos, separate lifecycles.
Every capability is reachable programmatically.
AI is a layer you build on, not the thing you sell.
Every component works standalone, but delivers more value combined with the rest of the ecosystem. This is the philosophy behind TokonoMix.
The structure and responsibilities are designed first; the code fills in a shape that already makes sense.
The same facts as the cards, in the form you would put in a document.
| Component | Owns | Purpose | Status | Availability | Repository / Product |
|---|---|---|---|---|---|
| ANS | Execution | Unattended execution | Production | Open Source | github.com/TokonoMix/agents-never-sleep |
| Council | Review | Multi-model review & consensus | Production | Open Source | npm: tokonomix-council-mcp |
| DCP | Coordination | Semantic project coordination | Public Specification | Public specification — live at devcopro.org | devcopro.org |
| AgentsWeaver | Distribution | Distributed collaboration | Active Development | Private Preview | agentsweaver.ai |
| AgentixMesh | Transport | Transport, trust, routing | Active Development | Open Source (repo) | github.com/TokonoMix/agentixmesh |
| beheer-v3 | Commerce | Digital service commerce | Production | Commercial | Not public |
| ZelixAI | Application | Enterprise AI platform | Production | Commercial | zelixai.ai |
| Gallery | Application | Commerce media platform | Active Development | Commercial | product page — to confirm |
| SurfOnline | Application | Local digital-economy platform | Active Development | Commercial | product page — to confirm |
TokonoMix — one architecture, built from independent components. Product URLs and maturity for Gallery and SurfOnline are marked to confirm rather than guessed.