Open-source guardian system

The guardian workspace for work that does not fit in one chat.

Seraph keeps memory, timing, and execution attached to the same workspace, so the agent can remember context, follow through with tools, and stay legible while work is happening.

Persistent memory Proactive scheduling Screen awareness Goals, workflows, and MCP
Live workspace
Memory active Human approval on

Context stack

Remember. Watch. Act.

Seraph keeps plans, recent decisions, and active runs close enough to resume useful work instead of restarting from prompt zero.

  • Persistent memory tied to the same workspace
  • Timing, reviews, and briefs ranked against live context
  • Human approvals stay visible when execution matters

Live memory

Long-term context stays live instead of resetting every session.

Goals, plans, notes, and recent activity stay available for the next decision.

Intervention queue

Briefings, reviews, and ranked interventions stay attached to the same system.

Timing and execution remain legible while work is moving.

Control layer

Approvals, fallback paths, and active runs remain inspectable.
09:41 strategist ranked intervention 09:43 review paused until approval 09:45 workflow resumed after confirmation

Why Seraph

Most agent products sell intelligence. Seraph sells continuity.

The product is designed to keep memory, attention, execution, and human steering in the same loop so useful work can continue after the first answer.

Keep the thread

Seraph is built for long-running work, not isolated prompts.

Plans, recent activity, queued decisions, and working context stay close enough to resume from instead of forcing the user to reconstruct state each time.

Show the work

Important execution stays readable while it is still happening.

The system does not bury approvals, degradation, or active work behind theatrical presentation.

Follow through

Tools, workflows, and integrations stay attached to the same working surface.

Reasoning and action remain part of one product loop instead of separate disconnected modes.

What stays visible

A workspace where the important layers stay on-screen.

Sessions, memory, active work, approvals, workflow runs, and recent execution belong to the same environment so the human can steer without losing situational awareness.

Sessions and active work

Continue from where useful work last paused instead of starting over.

Memory and evidence

Stored context supports the next decision instead of disappearing behind a thread boundary.

Approvals and intervention

Human decisions stay explicit when execution crosses trust boundaries.

Workflow runs and tool activity

The system stays inspectable while work is moving, not only after it finishes.

Shipped foundations

Longitudinal memory, real execution, and explicit human steering in one working system.

Memory

Persistent identity, long-term memory, and hierarchical goals.

Context compounds instead of resetting at the edge of a conversation.

Timing

Proactive scheduling for briefings, reviews, strategist ticks, and maintenance loops.

Interventions can be ranked around timing, not only availability.

Execution

Tools, workflows, and plug-and-play MCP servers in one working system.

Real-world actions stay attached to the workspace where they were initiated.

Oversight

Approvals, activity, and degraded-state visibility keep human control explicit.

Legibility remains part of the product instead of a hidden debug layer.

React 19 FastAPI SQLite + LanceDB APScheduler MCP Open source

Guardian loop

Observe. Model. Reason. Act. Learn.

The product only works when those loops reinforce each other instead of collapsing back into one-off chat.

01 Observe

Read current work, schedules, screen context, and recent activity.

02 Model

Maintain a living picture of the human, their priorities, and commitments.

03 Reason

Rank leverage, risk, and timing instead of treating every task as equal.

04 Act

Move through tools, workflows, and integrations with visible control.

05 Learn

Use outcomes and feedback to improve the next intervention.

Current direction

A guardian workspace with memory, execution, and visible control.

Explore the docs, inspect the code, and follow the roadmap as Seraph deepens memory, presence, and intervention quality.

Docs, research, roadmap, and source stay one click away.