Products · Helios · Agent runtime for the JVM
Helios. Agents that run reliably in production.
A provider-agnostic Model interface, a streamable and steerable session SDK, sandboxed code execution, structured output, durability primitives, and end-to-end tracing. Plain Java you can read and step through.
- Java 25
- Maven Central · ai.singlr
- MIT
- 95% coverage gate
- ~3,200 unit tests
- Used in production
Quickstart
Running in minutes. Self-serve.
> Add to your pom.xml
<dependency>
<groupId>ai.singlr</groupId>
<artifactId>helios-session</artifactId>
<version>${helios.version}</version>
</dependency>
<dependency>
<groupId>ai.singlr</groupId>
<artifactId>helios-anthropic</artifactId> <!-- or -gemini, -openai -->
<version>${helios.version}</version>
</dependency> > Run an agent session
try (var model = new AnthropicProvider().create(
AnthropicModelId.CLAUDE_OPUS_5.id(),
ModelConfig.of(System.getenv("ANTHROPIC_API_KEY")));
var session = AgentSession.create(
SessionPresets.workspace(model, Path.of("/path/to/repo")).build())) {
var terminal = session.runBlocking(UserMessage.text(
"Summarise the public API of the session module."));
System.out.println(terminal.result());
} How it works
Simple. Explicit. No magic.
One Model interface.
Anthropic, OpenAI, and Gemini implement the same interface. Swap providers without touching the rest of your code.
helios-anthropic · helios-openai · helios-geminiLong-lived, steerable agents.
AgentSession runs the agent loop on a virtual thread: send, stream 15 event types, interrupt mid-turn, gate actions with declarative permissions.
helios-session · helios-runtimeCode execution, contained.
A sandboxed JShell subprocess with a private RPC channel and descendant reaping, so generated code can't forge output or orphan processes.
helios-replSecrets stay out of the model.
CommandGrant hands the agent one CLI under tight controls. Registered secrets are redacted from everything the model sees.
helios-coreTyped, validated, cited.
Schemas from Java records, self-correction on mismatch, and provenance-tagged output where every field carries sources and confidence.
helios-coreRuns you can audit and resume.
Run stores, tool-call journals, and end-to-end tracing, with PostgreSQL-backed persistence for regulated audit trails.
helios-core · helios-persistenceChoosing a tool
When Helios fits. And when another tool fits better.
You want explicit, dependency-light primitives on the JVM: sandboxed code execution, secret redaction, provenance-tagged output, and durable, traceable runs.
You want the broadest catalog of ready-made integrations, vector stores, and chains on the JVM.
Your stack is Spring Boot and you want AI inside the Spring programming model.
You want planning-driven agent flows modeled as goals and actions.
In the field
Helios runs the agents on every SAIL engagement, including clinical and financial systems where every output needs a source.
How we work