Vireon Dynamics · Studio roadmap

One harness today. A stack worth following.

Liminal AI is the product you can install now. Everything else on this page is in development or research — we name it so you know where we're headed, not to imply shipping dates.

Shipping now

Alpha · v0.0.11

Liminal AI

The open-source agent harness you run on your own machine.

Model-agnostic ReAct loop with 140+ tools, persistent memory, persona system, terminal + web UI, and hosted one-command install. MIT licensed — you bring the API key.

  • Files, shell, git, web, browser, vault, documents, sub-agents
  • Two-tier routing + prompt caching on supported providers
  • Obsidian vault + hybrid memory recall
  • Public alpha on GitHub — moving toward v0.1.0 preview

The Vireon pipeline

Same studio rules for every line: local-first where it matters, model-agnostic, inspectable loops, MIT or open-core when we ship code.

In development

Liminal for Teams

Shared agent infrastructure for engineering teams — not another black-box copilot.

Built on the same harness as Liminal: shared persona profiles, team-scoped memory and vault patterns, centralized provider config, and audit-friendly session logs. Waitlist open; we ship when it is ready to trial, not before.

  • Shared personas and provider defaults across the team
  • Team vault sync (Obsidian-compatible patterns)
  • Tool-call audit trail from session JSONL
  • Private channel + priority fixes for early design partners
Early scaffold

Liminal Mobile

Your harness in your pocket — approve tools, send messages, stay in the loop.

A native companion for the Liminal web bridge: chat, streaming, approvals, and attachments against the same local harness you run on your desktop. Android work has started in the monorepo; iOS and store release follow a stable v0.1.0 harness.

  • Same SSE + REST contract as the web UI
  • Approval and ask-user flows on the go
  • Reconnect/resume on flaky networks
  • Pairs with self-hosted or LAN-reachable harness — not a cloud runtime
Research

Vireon Bench

CI-grade regression testing for agent harnesses — the eval suite, productized.

Liminal already ships 20+ scenario packs (`packages/eval`). Vireon Bench would package that for teams: run harness regressions in CI, compare runs over time, and gate releases on agent behavior — not just unit tests.

  • Scenario packs for memory, tools, web, approvals, long horizon
  • JSON run sink + pass/fail gates for pipelines
  • Works against your fork or pinned Liminal release
  • Open-core aligned with the MIT harness
Research

Harness SDK

Embed the Liminal loop in your product — IDE plugin, internal bot, or vertical agent.

For teams building their own surfaces: programmatic access to the same dispatcher, tools, memory bus, and settings model — without re-implementing retries, compression, and safety gates.

  • Headless harness API for custom UIs
  • Pluggable tool families and lazy loading
  • Same env / prefs precedence as Liminal
  • Design-partner program — tell us your integration shape

Liminal release milestones

Alpha → beta → RC → v0.1.0 public preview is the path for the harness itself (install GA, tagged releases). Other products above generally land after or alongside that foundation.

Shape what we build next

Tell us which product line matters to you — Team, Mobile, Bench, or SDK — and we'll prioritize design partners.

Install a local AI coding agent today

Install Liminal AI in one command. Free software, your API key, your machine — edit code, run tests, and ship faster without a SaaS middleman.