Oi (oi v0.2.0)
Copy MarkdownOi means Orchid integration.
Core concepts
Workspace — pure data struct holding graph, cluster, interventions. No process, no GenServer. Compiled and dispatched via pure functions.
Compile + Dispatch — two-phase pipeline:
Oi.compile/1— topology → static bundles (reusable)Oi.execute/2— bind interventions → plan → execute via pluggable executor
Executor — pluggable task execution strategy. Built-in:
Sync(serial),TaskSup(Task.Supervisor),Pool(NimblePool)(in future).Session — optional process tree per tenant, wrapping
OrchidSymbiont.Runtimefor multi-tenant isolation.
Quick start
Define a step
defmodule MyApp.Steps.Upcase do
use Oi.Step, name: :upcase
manifest(
inputs: [:text],
outputs: [result: :string]
)
routine text, _opts do
text |> String.upcase() |> ok()
end
endBuild and run
graph =
Graph.new()
|> Graph.add_node(%Node{
id: :up,
container: MyApp.Steps.Upcase,
inputs: [:text],
outputs: [:result]
})
{:ok, compiled} = Oi.compile(graph)
{:ok, result} =
Oi.execute(compiled,
inputs: %{"up|text" => "hello"}
)
{:ok, res} = Oi.Result.reify(result, "up|result")
# => "HELLO"Multi-tenant with Session
Oi.Runtime.Session.start("tenant-1")
Oi.Runtime.Session.start("tenant-2")
ws = Oi.Workspace.new("tenant-1", graph)
{:ok, compiled} = Oi.compile(ws)
Oi.execute(compiled,
executor: Oi.Executor.TaskSup,
executor_opts: [sup: Oi.Runtime.Session.tasks_tuple("tenant-1")]
)Symbiont step
defmodule MyApp.Steps.Predict do
use Oi.Step, name: :predict, symbiont?: true
manifest(
inputs: [:features],
outputs: [prediction: :string],
models: [:model]
)
routine features, models, _opts do
{:ok, result} = OrchidSymbiont.call(models.model, {:predict, features})
ok(result)
end
end
Summary
Functions
Compile graph into static bundles + plan.
Execute compiled plan with inputs and interventions.
Types
@type name() :: String.t()
Functions
@spec compile(Oi.Topology.Graph.t(), Oi.Topology.Cluster.t()) :: {:ok, Oi.Compiled.t()} | {:error, :cycle_detected}
Compile graph into static bundles + plan.
Pure topology — no interventions. Reusable across different intervention sets and inputs.
@spec execute( Oi.Compiled.t(), keyword() ) :: {:ok, Oi.Result.t()} | {:error, term()}
Execute compiled plan with inputs and interventions.
Options
:inputs— map of external io_key => payload, seeded into drafting memory:interventions— map of{:port, node, port} => {type, payload}:executor—Oi.Executor.Sync(default),Oi.Executor.TaskSup, orOi.Executor.Pool:executor_opts— passed to the executor (e.g.[sup: MyTaskSup]):plugins— OrchidPlugin pipeline:orchid_baggage— merged into Orchid run baggage
Examples
# Compile once
{:ok, compiled} = Oi.compile(graph, cluster)
# Execute with inputs + interventions
{:ok, result} = Oi.execute(compiled,
inputs: %{"step1|in" => "foo"},
interventions: %{{:port, :step1, :in} => {:override, "Bar"}}
)
# Same compiled plan, different inputs
{:ok, result_b} = Oi.execute(compiled, inputs: %{"step1|in" => "baz"})
# With Task.Supervisor
{:ok, result} = Oi.execute(compiled,
executor: Oi.Executor.TaskSup,
executor_opts: [sup: Oi.Runtime.Session.tasks_tuple("svs-1")]
)