v0.1.12 · Research preview

The platform to build
proactive agents.

A proactive agent acts when something happens - a document lands, a threshold trips, another agent disagrees - not when someone types. That is an event problem, not a graph problem. Cosmonapse gives you the substrate: design the system on a canvas in Genesis, run it on the open Core protocol and runtime, and watch it think in Prism. No supervisor loop, no control-flow graph to maintain - just components reacting to Signals on one bus.

Python 3.11+ · Apache 2.0 · Core, Genesis and Prism all ship in the one package

A synapse bus with a Receptor above it and a Neuron, an Engram and an Effector below it. Signals travel from the Receptor into the Neuron, from the Neuron to the Engram and to the Effector, and back again.terminalreceptorSynapseplannermemorysearch
One bus. A request from the edge, a memory read, a tool call and its result - each one a Signal that every participant could have seen.
// Why this exists

Three bets.

Cosmonapse is not a nicer wrapper around the same idea. It starts from three positions that most of the current tooling disagrees with: two about the systems people are building today, and one about the ones they cannot build yet.

// Bet 01

A graph is a diagram. It is not a program.

The problem

Graph and loop frameworks make you declare control flow before you know what the system does - nodes, edges, branches, and a supervisor that turns the crank. Then something arrives that you did not draw: a tool returns late, a human answers halfway through, a second model disagrees with the first. Every new case is a new edge, and the supervisor becomes the thing you cannot change.

What we do instead

Components emit Signals and react to Signals on one bus. Nobody holds the loop, so an agent can act the moment something happens rather than waiting for its turn in a graph. Concurrency, fan-out, retries, ordering and backpressure are properties of the transport - the same ones that have run distributed systems for twenty years.

// Bet 02

Wrappers do not have moats. Systems do.

The problem

A prompt, a vector store and a UI over somebody else's model is cheap to clone and hard to price. The teams stuck here have already solved which model to use; what they cannot do is get from a demo to something with a shape - observable, ownable, operable by a team, and defensible enough to charge for.

What we do instead

Build the system instead of the wrapper: domain memory that is yours, tools that touch real systems, interfaces people actually use, and an explicit policy for when a human steps in. Boutique AI, assembled from parts you own and running on an open protocol you can read.

// Bet 03

The domains that are coming have no prompt to wait for.

The problem

Physical intelligence, AI-native software and operating systems, communities of agents that outlive any one session - all of it is continuous and concurrent. A robot revises its picture of the world many times a second. An AI-native OS has to answer a file changing, a process dying, a peer coming online. A community of agents has members joining, disagreeing and dropping offline while the work carries on. None of it begins with a person typing, and none of it fits a request, a traversal and an answer. So every team working in these areas writes its own bus, its own scheduler and its own trace format before it gets near the actual domain problem.

What we do instead

For those systems an event substrate is not an optimisation, it is the only shape they have. Cosmonapse is that layer with agent semantics already in the envelope: a Signal knows whether it is a tool call, a memory write or a model output, so a sensor loop, an OS event stream and a community's membership churn are the same primitive rather than three bespoke integrations. Change the transport URL and what you prototyped on a laptop runs across a fleet. We are not building those products. The bet is that the substrate they need does not exist yet, and that whoever builds them should not have to build it twice.

A call stack beside an event bus. On the left a supervisor calls step A, which calls step B, which calls step C; a late reply arriving from outside has no frame to return into and is struck out. On the right, three participants and a late reply all sit on one synapse bus, and the late reply enters as an ordinary Signal.control flowsupervisorholds the loopstep Astep Bstep Clate replyno frame left to return intoevent streamlate replysynapseclassifierreviewerpolicy toolarrives as a Signal, like everything else
Left: a supervisor holding the loop, and a reply that arrives after its frame is gone. Right: the same reply as a Signal on the bus, indistinguishable from any other.
// The suite

Three products. One event stream.

Genesis, Core and Prism are separate products with separate jobs, and they share exactly one thing: the Signal envelope. That is what lets the designer, the runtime and the observability plane stay honest about the same system without any of them owning it.

CoreShips today

Cosmonapse Core

The protocol, SDK and runtime engine

The open Apache 2.0 protocol and the runtime that speaks it. One Signal envelope, one Synapse, replaceable Neurons. Memory (Engram), tools (Effector) and interfaces (Receptor) are first-class primitives, not framework add-ons - and the transport swaps from in-process to NATS or Kafka by changing a URL.

  • Signal envelope spec + the Python reference SDK
  • Neuron, Axon, Dendrite, Engram, Effector, Receptor primitives
  • Memory, NATS and Kafka transports behind one URL
pip install cosmonapseExplore
GenesisShips today

Cosmonapse Genesis

Design, build and test event-driven systems

The designer. Name a brain, scaffold it, and grow it on a canvas - one Synapse surrounded by the Neurons that think, Engrams that remember, Effectors that act and Receptors that listen. Genesis writes each module into your project as real source you own, edits it through the AST rather than regenerating it, then runs the brain and lets you talk to it without leaving the browser.

  • Canvas: place a primitive, Genesis writes the module and wires brain.py
  • Code: declarations as forms, handlers as code, edits applied through the AST
  • Test: run brain.py and drive its Receptors by terminal, HTTP or chat
cosmo genesisExplore
PrismShips today

Cosmonapse Prism

Observability for running agent systems

The observability plane. Prism is a read-only tap on the Synapse - it never claims a Signal - and turns one event stream into five views: the live neural graph, the per-run execution graph, the causal tree of a task, the raw Signal list, and metrics. Every number is derived from timestamps and lineage already in the envelope, so there is nothing to instrument and nothing extra for the SDK to emit.

  • Brain View: watch Signals fire between participants in real time
  • Signal Tree: causal parentage from TASK to FINAL, sub-tasks nested
  • Metrics: success rate, latency, memory hit rate, HITL waits, bidding
cosmo prismExplore
Next up · DeploymentServing a brain as a managed, quota-enforced, credential-scoped service is the next layer. Today you deploy Cosmonapse the way you deploy any Python or Node service. See the roadmap
A synapse bus labelled Core carries three participants below it. Above the bus, Genesis writes components onto it and Prism reads every Signal off it.Genesiswrites onto itPrismreads off itCore · one Signal envelopeclassifierpolicylookupnone of the three owns the system - they all read the same envelope
Genesis writes onto the bus, Prism reads off it, and Core is the bus. The Signal envelope is the only thing the three of them have to agree on.
// What is free, what gets sold

Where the open source line falls.

Core is Apache 2.0 and always will be. The essential builds of Genesis and Prism are free with it, in the same pip install, and they are enough to build and run a real system on your own machine. What gets sold later is Pro, Cloud and Enterprise - the deeper product surface and the operating burden around it, never the protocol.

Open source · Apache 2.0 · forever
CoreShips today

The protocol, the Python SDK and the runtime. Apache 2.0, and it stays that way - the substrate your system is built on is not something we intend to sell you back later.

  • Signal envelope spec and the reference SDK
  • Neuron, Axon, Dendrite, Engram, Effector, Receptor
  • In-process, NATS and Kafka transports behind one URL
Genesis + Prism, essential buildsShips today

Enough of the designer and the observability plane to build, run and debug a real system on your own machine, free, in the same pip install. No account, no key, no seat count.

  • Genesis: canvas, code editing through the AST, run and test a brain
  • Prism: live brain view, signal tree, the raw stream and core metrics
  • Local only, single developer, your source on your disk
Commercial · none of it shipping yet
ProPlanned

The deeper end of Genesis and Prism for people doing this full time - the surface that pays for itself in a working week rather than the surface you need to get started.

  • Advanced Genesis workflows and larger-system tooling
  • Retained history, comparison and analysis in Prism
  • Per-seat, self-serve
CloudPlanned

The hosted version of the same loop. Design and observe in a browser, deploy a brain as a service, and let a team work on one project at once.

  • Managed deploy with quotas and scoped credentials
  • Shared projects, presence and a persistent Signal history
  • Usage-based
EnterprisePlanned

What a regulated buyer needs before any of this is allowed near production, distributed as a licensed wheel off a private index.

  • RBAC, audit trails, multi-tenancy and HA
  • Offline-verifiable licensing, airgap friendly
  • Support, SLA and architecture review
A dashed vertical line divides the suite. On the open source side, Core sits under a free Genesis and Prism build. On the commercial side, a Pro tier sits beside them, with Cloud and Enterprise stacked above.OPEN SOURCE · APACHE 2.0 · FOREVERCOMMERCIAL · PLANNEDGenesis + Prism, essential buildsdesign it, run it, watch it - all localCosmonapse CoreSignal envelope · SDK · runtime · transportsone pip install, no account, no keynothing on the right is load-bearing for anything on the leftProthe deeper Genesis and Prism surfaceCloudhosted brains · deploy · collaborateEnterpriseRBAC · audit · tenancy · HA · supportsold against operating burden, never against the protocolthe protocol stays open so the systems you build on it stay yours
Core is Apache 2.0 and always will be. The essential Genesis and Prism builds are free with it. Pro, Cloud and Enterprise are the commercial layers, and none of them are shipping yet.

The rule we are holding ourselves to: nothing on the paid side is load-bearing for anything on the free side. A system you build today keeps running on an open protocol, an open runtime and a local toolchain whether or not you ever pay us. Early access is where the commercial tiers get shaped, so if you have an opinion about what belongs on which side, that is the room.

// How they fit

Design it. Run it. Watch it.

You are never handed off between tools. The canvas, the runtime and the traces are three views of the same running system, and the source on disk is the only artifact.

Design · Genesis

Design

Lay the system out on a canvas - Neurons that think, Engrams that remember, Effectors that act, Receptors that listen. Genesis writes real source into your project and edits it through the AST as the shape changes.

Run · Core

Run

Everything talks over one Signal envelope on one Synapse. Start in-process, move to NATS or Kafka by changing a URL. No orchestrator holds the loop, so nothing has to be rewritten to scale out.

Observe · Prism

Observe

Attach a read-only tap to the same bus and watch the system think: the live graph, the causal tree behind a single task, where a task's wall clock actually went. Tracing is free because the bus already saw it.

Three windows side by side. The Genesis canvas shows a Synapse with a Neuron, an Engram, an Effector and a Receptor placed around it. A terminal runs brain.py and prints REGISTER, TASK, TOOL_CALL and FINAL lines. Prism's Brain View shows the same four components around the same Synapse with Signals travelling between them. An arrow returns from Prism back to Genesis.DesignGenesisCanvasCodeTest+ Neuronneurons/planner.pyRunCorebrain.py$ python brain.pysynapse cosmo://127.0.0.1:7070REGISTER planner · memory · searchTASK → plannerTOOL_CALL → searchFINAL ok 1.4sObservePrismBrainTreeMetricswhat you watch sends you back to the canvas
The same four shapes in the first window and the third, with nothing between them but your own file running. Genesis writes the source, Core runs it, Prism watches it - and what you see sends you back to the canvas.
// Build on Cosmonapse

~20 lines. Three transports.

A working multi-agent system is about twenty lines - define an async function, wrap it in an Axon, attach it to a Dendrite, call dispatch_and_wait. No protocol boilerplate. And the scale path is a single URL swap: MemorySynapse NatsSynapse KafkaSynapse, with zero changes to your Neuron code.

main.py
import asyncio, os
from cosmonapse import Axon, Dendrite, connect_synapse

SYNAPSE_URL = "memory://"   # ← the only line that changes
axon = Axon.huggingface("greeter",
    endpoint="https://router.huggingface.co",
    model="meta-llama/Llama-3.1-8B-Instruct",
    api_key=os.environ["HF_TOKEN"],
    use_chat_api=True, capabilities=["chat"])

async def main():
    synapse = await connect_synapse(SYNAPSE_URL)
    worker = Dendrite(synapse=synapse, namespace="demo", role="worker")
    worker.attach_axon(axon)
    orch = Dendrite(synapse=synapse, namespace="demo")
    async with worker, orch:
        reply = await orch.dispatch_and_wait(
            neuron="greeter", input={"prompt": "Say hello to Cosmonapse."}, timeout_s=30.0)
        print(reply.payload["output"]["response"])

asyncio.run(main())
// What you build with it

Boutique AI systems, not another wrapper.

The shape is always the same: an interface people actually use, domain memory that is yours, tools that touch real systems, and a policy for when a human gets involved. The model is the cheapest part. Everything around it is the product.

Claims triage
A first-notice-of-loss intake that reads the document, pulls policy history, flags the fraud signals a generalist model would miss, and escalates to a human with the reasoning attached.
Receptor (API) → classifier Neuron → policy Engram → fraud Effector → adjuster escalation
Clinical intake
A patient-facing conversation that gathers history, checks it against a coded protocol, and hands a structured summary to a clinician - with every step in the record.
Receptor (chat) → intake Neuron → protocol Engram → summariser → clinician review
Research desk
Several models working the same question in parallel from different sources, disagreeing on the record, with the reconciliation step doing real work instead of a supervisor picking one.
fan-out TASK → n Neurons → competing AGENT_OUTPUTs → reconciler → FINAL
Contract review
Clause-by-clause analysis against a firm's own precedent library, where the value is the precedent and the escalation policy - not the model underneath it.
Receptor (upload) → splitter → clause Neurons → precedent Engram → risk report

None of these are graphs. Each one is a set of independent components reacting to Signals, where the interesting behaviour - escalation, disagreement, a late tool result, a human in the middle - is a Signal arriving rather than an edge you drew. Fourteen runnable examples show the patterns end to end.

An API Receptor above a synapse bus feeds a classifier Neuron, a policy history Engram and a fraud checks Effector below it. A dashed escalation path rises from the bus to a human adjuster.intakeAPI receptoradjusterhuman reviewESCALATIONSynapseclassifierpolicy historyfraud checksnothing here is a graph - the escalation is a Signal arriving, not an edge you drew
Claims triage, drawn out: an interface people use, memory that is yours, tools that touch real systems, and a policy for when a person gets involved. The model is the cheapest part of it.
// Early Access Program

Running an enterprise team or startup building something real on this? Come talk to us.

This round is scoped to enterprise teams and funded startups. We run architecture sessions with early adopters on the system they are actually building - free, framework-agnostic, and with the people writing the protocol. The first 10 members accepted also get credits, a standing discount on Cosmonapse Cloud, and the option to form a Letter of Intent with us.

Build a system, not a demo.

Core 0.1.12 is a research preview: the envelope is drafted, the Python SDK implements it, and Genesis and Prism run locally today off the same CLI. It is open source under Apache 2.0 - read the spec, disagree with it in public, and build on it either way.