live Open-source agent runtime · built on the BEAM

The durable runtime for swarm intelligence

Z8 runs fleets of AI agents that make real decisions — and never lose one. Every run is an event stream: it survives crashes, replays to the exact step without re-calling the model, and scales to millions of agents in parallel. The production runtime the agent frameworks forgot to ship.

deps {:z8, "~> 0.1"}
Crash-proof by construction Every run replayable, exactly Runs on your infrastructure
z8 agent swarm · operations live
Agentsdecide Eventsrecord Handlersreact Actionsact Sagascoordinate Sensorsingest Skillspackage Swarmsscale
Built on the BEAM — the runtime behind the apps that can't afford to go down:
WhatsApp Discord Ericsson Pinterest Heroku
Millions
of agents in parallel, one runtime
0
runs lost when a process dies
100%
of every run replayable, exactly
1
framework: agents, jobs, sagas, AI, dashboard
The core idea

The record is the runtime.

A Z8 agent doesn't hold its state in memory and hope. Every decision it makes is appended to an event stream — and that stream is the agent. Rebuild it after a crash, replay it for an audit, fork it for a test: the past is data, not a re-run. And replaying an AI agent never calls the model again — the reasoning already happened; the events remember it.

  • Crash mid-run? The agent rebuilds from its log and continues — nothing lost, nothing half-done.
  • Replays are exact and free. Recovery and audit never pay for inference twice.
  • Pure logic. decide / evolve are pure functions — agents you can unit-test with no mocks and no network.
One AI run · as events
RunStarted InferenceRequested ToolRequested ToolSucceeded InferenceRequested RunCompleted
fold → %Run{status: :done, answer: "…"}

Rebuilt from the log in microseconds · 0 model calls on replay.

The platform

A framework where production isn't an afterthought.

Most agent tools nail the demo and leave durability, concurrency and operations to you. Z8 ships them in the box — because that's the difference between a notebook and a system you can run a business on.

Durable by construction

State lives in an append-only event log, not in RAM. Processes die, machines reboot, deploys roll — and every agent comes back exactly where it was. No lost runs, no half-finished work.

A real swarm

Built on the BEAM — the VM that ran ~2M live connections per server at WhatsApp and 11M+ concurrent users at Discord. Millions of lightweight, isolated agents run in true parallel under supervision; one crashing is contained in milliseconds, and the fleet never notices.

Reasoning, priced to the task

A full compendium of strategies — from one-shot answers to tree- and graph-of-thought deliberation — with an adaptive router that spends deep reasoning only where the problem earns it. Tools, policies and budgets included.

Batteries included

Background jobs, durable scheduling, sagas with automatic compensation, event-driven sensors, dead-letter queues, backpressure — the whole production periphery, in one framework with one mental model. Not ten libraries you wire together.

See everything, live

A real-time control room streams every agent, run, event and trace straight from the event store — no polling, no extra infrastructure. Built for operators and risk teams, not just engineers.

Provable & compliant

Because the event log is the source of truth, a tamper-evident audit trail isn't a feature you bolt on — it's the substrate. Replay any past decision exactly, prove it to a regulator, retain it for years.

Build

An agent is a function. A swarm is a supervision tree.

The surface is small and pure: decide what to do, evolve your state from the events, and let the runtime handle persistence, concurrency, retries and recovery. Here's a real durable agent, an AI agent in five lines, and the wiring that turns them into a running application.

# A durable agent: the consistency boundary.
# Pure decide/evolve over an event stream — no I/O, no clock.
defmodule Bank.Account do
  use Z8.Agent, snapshot: [every: 100]

  defstruct account_id: nil, balance: 0, status: :new

  # decide: current state + command -> events (+ optional reply)
  def decide(%__MODULE__{status: :new}, %OpenAccount{} = cmd, _ctx),
    do: %Z8.Write.Decision{
      events: [%AccountOpened{account_id: cmd.account_id}],
      reply: :opened
    }

  def decide(%__MODULE__{balance: b}, %Withdraw{amount: a}, _ctx) when a > b,
    do: {:error, :insufficient_funds}

  # evolve: fold an event back into state. Runs on every replay, forever.
  def evolve(acc, %AccountOpened{} = e),
    do: %__MODULE__{acc | account_id: e.account_id, status: :open}
end
Why developers stay
  • Pure core. decide / evolve are deterministic functions — replayable and testable with no process and no network.
  • One agent, every shape. The same aggregate is a business entity, an AI run, a background job, or a state machine. One thing to learn.
  • It compiles to OTP. Routers, supervisors, persistent subscriptions, snapshots — the BEAM's 25 years of reliability engineering, generated for you.
  • Open and inspectable. An in-memory event store ships as the reference adapter; no database required to run the full stack locally.
$ mix deps.get && iex -S mix
Reasoning

Every way to think — and the judgment to pick one.

Routine questions shouldn't cost the same as high-stakes calls. Z8 ships a full reasoning compendium and an adaptive router that sizes each task automatically — so quality stays high where it matters and spend stays low where it doesn't. Every step is recorded as events.

Direct routine

One-pass answers for high-volume, low-risk questions — fast and cheap, by the thousands.

Chain-of-thought step by step

Works a problem through one step at a time, with a parsed, on-record conclusion.

ReAct reason + act

Interleaves reasoning with tool calls — look up records, hit your systems, then decide.

Tree-of-thought weigh branches

Generates, evaluates and expands several lines of thought, keeping the strongest.

Graph-of-thought connect

Builds a DAG of ideas — generate, connect, aggregate — for problems that branch and merge.

Algorithm-of-thought search

Explicit algorithmic search at temperature zero, for answers that must be exact.

Recursive (TRM) self-improve

Reasons, supervises its own answer, and revises — with halting that knows when to stop.

Adaptive the router

Classifies each task and routes to the cheapest strategy that will get it right — the one piece that turns the seven above into a system.

Tools are governed, not trusted. Agents act through validated actions with JSON-schema arguments, narrowing-only policies (deny wins), deadline budgets and failure-preserving compensation — so an agent can only ever do what you've allowed, and a provider hiccup retries instead of failing the run.
Every run, on the record

Open the box. Read the reasoning.

A Z8 run isn't a black box that returns a verdict — it's a readable record: what triggered it, the facts it weighed, the reasoning, the outcome, and a tamper-evident seal. The same record debugs an incident on Tuesday and answers an auditor a year later.

What every run carries
  • Who & what. The subject, the request, and which agent handled it.
  • The inputs it weighed. Every fact and signal the decision was based on.
  • The reasoning. A plain-language rationale — and the strategy that produced it.
  • The outcome. Approved, declined, escalated — with terms and a confidence score.
  • The exact moment. A precise, monotonic, trustworthy timestamp.
  • An append-only seal. Each record linked to the one before, so any later change is caught.

Hand it to a customer, an auditor, or your own on-call engineer — and back it with proof it hasn't been touched since the moment it was made.

Run · loan-4217 Approved
SubjectMaria S. · #9f2a (masked)
RequestPersonal loan · $24,000
Decided byLending Assistant · ReAct
When2026-06-23 14:02:11 UTC
Inputs considered
Credit score742
Debt-to-income28%
IncomeVerified
Adverse flagsNone
Reasoning

"Applicant meets every criterion for the Standard tier. Debt-to-income is within policy and income is verified, with no adverse flags. Approving at the standard rate."

Outcome
ResultApproved · $24,000 · 6.4% APR
Confidence0.94
Sealed & tamper-evident · e7f3…88 · replayable on demand
The control room

See your whole swarm, live.

A real-time control room for every agent you run — what's deciding, what's waiting, what needs a human. Driven straight from the event store: no dashboards to refresh, no polling, no second system to operate. It updates the moment anything changes.

Runs today
14,902+12%
Agents live
312+7
Avg response
1.2s
Flagged for review
7
Live runs 5
ReferenceAgentStatusTime
loan-7f3aLending AssistantReasoning1.2s
claim-9c11Fraud ReviewCleared0.8s
cust-2bd0OnboardingAwaiting docs
pay-5e8fPaymentsSettled0.4s
loan-aa20Lending AssistantEscalated2.7s
Throughput · last hour
low 0.9kavg 1.4kpeak 2.1k
Reactors
Projections caught up
Notifications +14
Transfer saga 3 in flight
Scheduler & DLQ
Scheduled jobs28 queued
Next run00:42
Dead-letterempty
Why Z8

The agent frameworks made prototypes easy. Z8 makes them survive production.

CrewAI, LangGraph and AutoGen are great at the first demo. Durability, true concurrency and operations are left as an exercise for the reader — usually discovered the first time an agent crashes mid-task in production. Z8 starts where they stop.

How Z8 compares with CrewAI, LangGraph and AutoGen across runtime, durability, operations and governance.
z8Z8 all built in CrewAI LangGraph AutoGen
Runtime & scale
Runtime & concurrency BEAM · true parallel Python · GIL Python · GIL Python · GIL
Fault isolation — one agent ≠ the fleet
Durability & recovery
Survives a crash mid-run ~
Replay without re-calling the model
Operations
Jobs, scheduling, retries, dead-letter
Sagas with automatic compensation ~
Governance & trust
Audit trail by construction ~
Live observability included ~ ~
What you still build yourself Nothing — it's the runtime Durability · ops · audit Persistence · ops · glue Most of production
built in partial — bring your own or paid add-on not provided
And the durable-execution engines? Temporal, Inngest and Restate give you crash-safe workflows — but you still bring your own agent, your own reasoning loop, your own model orchestration, and you glue an LLM onto a workflow step. Z8 is both: a durable runtime and an agent framework, native to each other, where the agent's reasoning is the durable record.
Why now

Agents are moving from demos to decisions.

Enterprises are putting agents on the critical path — money, eligibility, customer outcomes. The moment an agent's decision matters, "it usually works" stops being good enough, and durability, isolation and proof become the whole game. That's the market Z8 is built for.

$52.6B by 2030
AI-agents market, growing ~46% a year from $7.8B in 2025 as agents move onto the critical path.
MarketsandMarkets, 2025
52%
of executives say their organization has already deployed AI agents — and most hit the same wall: reliability in production.
Google Cloud, 2025
>40%
of agentic-AI projects will be scrapped by 2027 — for cost, reliability, or unclear value. That's the gap Z8 closes.
Gartner, 2025
The wedge

A moat the Python frameworks can't copy by shipping a feature.

01

Architecture, not add-ons

Durability and audit fall out of event sourcing for free. Retrofitting them onto a stateless Python loop is a rewrite, not a release.

02

The BEAM is the substrate

Millions of supervised, isolated agents in true parallel — 25 years of telecom-grade reliability you can't bolt onto a GIL.

03

Replay never pays for inference

Recovery, audit and testing re-fold events, not re-run models — the cost and determinism story no stateless framework can match.

04

Lands where the value is

Open-source adoption bottom-up; regulated, high-stakes operations — finance, insurance, healthcare — top-down. One runtime, both motions.

05

The layer the BEAM was missing

Elixir gave the world concurrency and supervision — but not event-sourced, cross-deploy durability. That's the one piece its community rents from workflow engines. Z8 makes it native.

06

The reliability layer is the prize

Capital is pouring into agent infrastructure — orchestration, durability, observability, memory. Z8 is all four in a single runtime, not four vendors stitched together.

Compliance & audit

When the decision is regulated, the record already exists.

Loan approvals, fund transfers, eligibility — regulators want those decisions logged immutably, traced to their inputs, retained for years, and independently auditable. In Z8 that isn't a module to buy: every agent decision is already a block in an append-only event chain — linked to its cause and ordered for life.

The event chain — every agent decision an immutable, linked block · no personal data on-chain
genesis
0000…0000
block 4216sealed
agentonboarding-agent
decisionAccount opened
prev ←71e9…04
hash9f2a…c1
block 4217sealed
agentpayments-agent
decisionWithdrawal cleared
prev ←9f2a…c1
hashc44b…20
block 4218latest
agentlending-assistant
decisionLoan approved
prev ←c44b…20
hashe7f3…88
Verify the full history intact & complete Alter a single past record tampering detected, instantly
How the event chain works
01

Append-only blocks

Every agent decision is an immutable event — appended, never edited or deleted. The ledger only grows, and that history is the single source of truth the agent is rebuilt from.

02

Totally ordered

Each block carries a monotonic UUIDv7 — a time-ordered id with a per-millisecond counter — so the whole chain has one provable, replay-safe order, with no coordinator and no clock to trust.

03

Causally linked

Every block points to its cause (causation_id) and its originating request (correlation_id) — so you can walk any decision back to its origin, or forward to everything it set in motion.

04

Walk & verify

causes, effects and trace rebuild the full provenance on demand. Drop or alter a block and its links, order and hash stop reconciling — the break can't hide.

A private event chain, not a public blockchain. You get the properties that matter for audit — append-only, totally ordered, causally linked, independently verifiable — with none of the tokens, miners, or settlement latency. It runs inside your own infrastructure, at the speed of an event store.

Tamper-evident by linkage

Every block is linked to its cause and stamped with a content hash. Drop or alter a past decision and the chain stops reconciling — caught the moment it's verified, and an auditor can check the whole history independently, without touching your live systems.

Replay any past decision

Reconstruct exactly what an agent did, and why, at any point in the past — every input it relied on still on file. Reproducibility is the default, not a forensic project.

Privacy by design

Personal data lives off the permanent record. Honour "right to be forgotten" requests in full — and the audit trail still verifies afterward.

Retention & legal hold

Keep records exactly as long as the rules require on write-once storage, then retire them automatically — and freeze everything the moment an investigation begins.

How Z8 maps to the regulatory landscape

RegimeWhat it asks forHow Z8 supports it
EU AI ActReg. (EU) 2024/1689 · Arts. 12, 15, 18–19 · high-risk rules from Aug 2026 Automatic logs over a high-risk system's lifetime, protection against unauthorised changes, multi-year retention. Every decision logged automatically into an append-only, immutable record, kept as long as you require.
DORAReg. (EU) 2022/2554 · Art. 9 Preserve the authenticity and integrity of data; documented cryptographic controls. Records form an append-only, immutable history; pairs with documented cryptographic controls and a write-once backup where required.
SEC 17a-4Rel. 34-96034 · Option A A re-creatable audit trail of any modification — who, what, when — on immutable media. An append-only history on write-once storage; every record linked to the one before it.
FINRARules 4510, 3110 Books-and-records, supervision, and a record of what stood behind AI-driven recommendations. Every AI step recorded — the inputs, the reasoning, the model used, and the outcome.
GDPRReg. (EU) 2016/679 · Art. 17 Right to erasure — in direct tension with a permanent, immutable log. Personal data lives off the record and is truly erased on request; the audit trail still verifies.
EIOPAOpinion EIOPA-BoS-25-360 Reproducibility and traceability of how the AI reached its decisions. Any past decision can be replayed exactly, with every input it relied on on file.
The technology, not the verdict. These regulations are principles-based — none mandates a specific technology, and no software is "compliance" on its own. What Z8 gives you is the hard part done right: a tamper-evident, retainable, independently verifiable record your legal and risk teams can build on. Read the mapping as "gives you what the requirement needs" — each claim checked against the primary sources.
Get started

Put your agents on a runtime that can take it.

For developers

Start building

Add one dependency, boot the full stack locally with no database, and write your first durable agent in minutes. Open source, Apache-2.0, batteries included.

For the enterprise

Book a demo

Tell us about your use case — lending, claims, onboarding, payments — and we'll show you Z8 deciding, recovering, and proving it, on a workload like yours, on your infrastructure.