Built for humans & coding agents

The flow you read.
The flow you run.

A batteries-included, adaptable core for developing agents.

Readable flows are the executable source of truth. Guided tooling makes the next action clear. Reports let you follow a failure to its evidence, one layer at a time.

Explore the development loop

Build → run → test → investigate → improve.

ONE CORE / TWO READERS● HUMAN + AGENT

READABLE, EXECUTABLE FLOW

draft→execute_tool→render

Flow declaration → runtime + diagram

DEVELOP

Guided TUI

Settings · services · tests

INVESTIGATE

Scoped evidence

Run · test · step · call

● Retained runs. Linked evidence. Available usage.

01

The flow is the source.

Execution and a readable map share a definition.

02

The workbench is included.

Configure, launch, test, and inspect.

03

Evidence opens in layers.

Keep the history. Read the relevant scope.

One core / the whole development loop

Small context.
Traceable history.

Explore four stages of the development loop. Open only the detail you need.

The core difference

One definition.
A shared understanding.

The workflow is readable development material and executable input. Humans and coding agents navigate the same declared structure.

Read it. Run it. Navigate it.

Versioned Markdown declarations define the workflow. The runtime compiles them for execution; the diagram projects that structure with links to nested components and resources.

Freshness checks help keep the saved view aligned with the declarations.

Explicit routes, schemas, guards, and bounded retries give reviewers and coding agents concrete constraints to work with. Accuracy is tested against acceptance scenarios.

BUNDLED FLOW · TOOL PATH

draft → execute_tool → render

Also declared: draft → render when no tool is needed.

The executive case

Make the development loop
easier to reason about.

For companies building or replacing an agent core, the opportunity is less repeated infrastructure work and more focused investigation.

Reduced contextRead the relevant evidence, in layers.+

Progressive discovery starts small, identifies an owner, and opens only necessary detail. Measure tokens spent investigating an error, total investigation cost, and time to identify its cause.

Accuracy through explicit structureConstrain, inspect, and test changes.+

A shared executable declaration reduces the gap between the workflow a reviewer reads and the structure the runtime executes. Measure correct diagnoses, accepted fixes, regressions, and scenario pass rates.

An adaptable coreReuse the workbench. Own the application.+

Resource packs separate business workflows, prompts, tests, and integration choices from common machinery. Measure time to the first useful workflow, integration effort, and effort to change a provider.

How it relates to tracingTracing is one part of the loop.+

api-agent can use LangFuse or OpenTelemetry for tracing. Its proposition spans executable workflow definitions, development setup, guided operation, testing, scoped evidence, and retained reports.

Put the model to work

A smaller investigation.
A clearer decision.

Bring one real failure and one representative workflow. Evaluate the context needed to diagnose it, the precision of the fix, and the evidence retained afterward.

Explore api-agent ↗

Open source · BSD 2-Clause · adaptable application packs