Skip to content
SAIL

Open source

Forward-deployed tooling.
Maintained in the open.

The lab ships what it uses. Helios runs our agents. Sail provisions our environments. MIT-licensed from the first commit. No telemetry, no paywall, no tier.

Helios · JVM agentic runtime

Helios.

Agentic runtime for the JVM. Provider-agnostic models, sandboxed code execution, three-block memory with offline consolidation, schema-validated output with self-correction. Simple, explicit, no magic.

helios · helios-repl
Depth 0 · running
Cycle 01
λ 01 · Providers

Provider-agnostic models.

Anthropic, OpenAI, Gemini — same Model interface. Swap providers without touching the rest of your code. Resource lifecycle handled by the runtime.

helios-anthropic · helios-openai · helios-gemini
02 · Sandboxed REPL

Code execution loop.

RLM pattern. The model writes code, the sandbox runs it, errors feed back. ReplSession is AutoCloseable — it owns its sandbox subprocess and cleans up on close.

helios-repl
03 · Memory

Three-block memory.

Letta-inspired three-surface memory: identity, user_profile, working_memory. Agents self-edit during conversations. Offline consolidation via Dreaming.

helios-core · helios-persistence
04 · Tool security

Secrets stay out of the model.

CommandGrant lets the model invoke a single CLI binary under tight controls. Secrets registered via withEnv are auto-redacted from model output via byte-level Aho-Corasick.

helios-core
05 · Structured output

Schema-validated. Self-correcting.

When the response doesn't conform to the schema, the agent receives a corrective USER turn carrying the diff and re-iterates. The model self-corrects on the next attempt.

helios-core
06 · Fault tolerance

Built-in. Zero-dep.

Composable retry, circuit breaker, and timeout. Parallel tool execution on virtual threads with per-tool fault isolation — one timeout doesn't abort the others.

helios-core
Java 25 · Maven Central · MIT Read the README
sail
Running
$
Cycle 01

Sail · environment orchestrator

Sail.

A single native binary that provisions bare-metal servers and isolates AI-assisted dev environments. One project per Incus container — own filesystem, network stack, rootless Podman runtime. Spec-driven autonomous loop with guardrails and snapshot rollback.

Interactive mode Daytime · Zed

SSH-connect from your editor. Agent runs in the Agent Panel. You're in the loop — brainstorming, exploring, writing specs.

Autonomous mode Overnight · headless

Write specs during the day. Walk away. `sail spec dispatch` picks the next ready spec and runs the agent under guardrails.

Fig · 02Spec lifecycleRev 1.0 — sail
01
pending
ready
02live
in_progress
agent working
03
review
auto-hooks
04
done
merged
Drafted in-labSAIL · SPEC · 001

> Install in one line

$ curl -fsSL https://raw.githubusercontent.com/singlr-ai/sing/main/install.sh | bash
Native binary · Zero deps · GraalVM · <1ms startup · MIT Read the README

Why we publish

Built in the field.
Shipped to the commons.

I. Origin

Every tool here started as something we built for ourselves. Helios came out of needing a clean, primitives-first agent runtime on the JVM. Sail came out of running multiple AI-assisted dev environments on a single bare-metal server, isolating projects so one runaway agent can't affect another.

II. License

Once they harden in our own work, they go MIT. Take them. Fork them. Run them in your stack. The license is the contract.

III. No tiers

We don't gate features. We don't gate the source. There is no enterprise tier behind a sales call. What we build for the lab's own dev loop is what flows back into these repos.