Introduction
Introduction
Holaryn Agent is a proprietary, local-first AI agent from Synergentic, Inc. It runs entirely on
your machine — its own agent loop, tools, memory, scheduler, and web dashboard — with no account
and no external services required.
What it is
Holaryn Agent is a general-purpose, always-on assistant for both coding and personal or business work. Everything it needs lives on your computer:
- Its own agent loop — the agent calls a model, executes tools, and repeats until the task is done, with an approval gate in front of consequential actions.
- Tools — shell and filesystem tools, image generation, and any external tools you connect over the Model Context Protocol (MCP).
- Memory — a local hybrid memory (SQLite with full-text and vector search) that persists across sessions. See Memory.
- Scheduler — unattended jobs on cron, interval, or one-shot schedules that survive restarts. See Scheduler and automation.
- Web UI — a local, accessible browser dashboard served by the agent itself. See Web interface.
- Skills, plugins, subagents, teams, and peers — installable capabilities and multi-agent coordination, all running locally.
Accessibility is a founding design goal, not a retrofit: the terminal output, web UI, and desktop app are all built to work well with screen readers.
You bring your own model. Holaryn Agent talks to Anthropic, OpenAI, Gemini, or any OpenAI-compatible endpoint (including local servers such as Ollama or LM Studio). There is no built-in fallback provider — until you configure a connection, the agent tells you exactly what to set up instead of silently assuming a vendor. See Connections and models.
Relationship to Holaryn Space
Holaryn Agent is one of two products:
- Holaryn Agent (this one) — proprietary and standalone. It never requires an account or a server.
- Holaryn Space — Synergentic's optional hosted service for managing fleets of agents across machines, with shared memory, multi-channel delivery, and a hosted console. Its internal backend is called Holaryn Platform.
The dependency only points one way: Holaryn Space builds on the agent, never the reverse. If you
connect the agent to Holaryn Space, it happens over the versioned HACP control protocol ("connected
mode"); disconnect it and the agent keeps working with its local defaults. You can check whether
connected mode is configured with holaryn hacp status — a fresh install reports standalone.
How it works, at a glance
Three ideas explain most of the agent's behavior:
One event stream
Everything the agent does — model turns, tool calls, approvals, task progress — is emitted as one typed stream of events. The terminal renderer, the web UI, audit logs, and replay are all just consumers of that same stream, so what you see in the terminal and what you see in the browser are always the same underlying record.
Autonomy and approvals
Every tool the agent can use carries a consequence tag: is the action reversible, and what category is it (filesystem, network, and so on)? Before a consequential action, the agent resolves your autonomy posture to one of four levels: act silently, act and notify you, ask you first, or refuse to run unattended. The default is conservative — reversible actions proceed, irreversible ones ask — and an unanswered "ask" suspends the run rather than timing out into approval. Details in Autonomy and approvals.
Interfaces with local defaults
Providers, memory, channels, approvals, scheduling, voice, and execution are all pluggable interfaces, each shipping with a local implementation that works out of the box: SQLite memory, a terminal approval prompt, an APScheduler backend, local process execution. Richer implementations (for example, Holaryn Space-backed ones) can plug into the same seams at runtime, but nothing depends on them.
Where to go next
- Installation — desktop app or authorized private-source installation.
- Quick start — from zero to your first task.
- Running the agent — attended sessions, the persistent host, and the OS service.
- CLI reference — every
holarynsubcommand.