The dashboard

Sigil ships a local web dashboard served by the daemon. It’s where setup happens, and afterwards where you inspect memory, watch capture happen live, and change configuration without touching a file.

sigil daemon open    # open it
sigil daemon url     # or print the URL with its auth token

It runs on 127.0.0.1:7777 by default, bound to loopback and gated by a token stored at ~/.sigil/gui.token. Nothing is exposed to your network unless you change http.host yourself.

Layout

Eight screens in three groups down the left rail:

Memory

• Home — status, counts, recall health, recent activity
• Knowledge Base — every fact and document, searchable and editable
• Graph — the entity graph, rendered
• Activity — a live trace of every capture and recall

Connections

• Agents — which coding agents are wired, and what each has written
• Devices — paired machines and pending pairing codes

System

• Engine — worker pool, queue depth, database
• Settings — every configurable value, plus the danger zone

Press ⌘K anywhere to jump between screens — or to run a live memory search without leaving the page you’re on.

Home

The at-a-glance answer to “is my memory working?”

• Status — daemon, database, LLM provider, embedder, each with a real health check behind it rather than a green light that means “the config file exists”
• Counts — stored memories, entities (people · topics · docs), graph edges
• Recall health — whether retrieval is actually returning things, which is a different question from whether writes are landing
• Recent activity — the last few captures, with a link into the full trace

If something is wrong, Home says so in a banner with the action attached to it — not a red dot you have to go interpret.

Knowledge Base

Every fact and document in one list, filterable by namespace, category, and pod. Click a fact to open the detail panel: full content, category, confidence, importance, the pod it belongs to, which agent and device wrote it, and its provenance chain back to a source document.

This is where you correct memory. Edit a fact, change its visibility scope, or delete it. Corrections here take effect on the next read, on every device.

Graph

The entity graph — typed nodes (person, project, service, concept) connected by extracted relations. Useful for two things: seeing what Sigil thinks the shape of your work is, and spotting a mis-extracted entity that’s splitting one real thing into three nodes.

Activity

A live stream of everything the engine does: each hook fire, each search, each classification decision, each LLM call with its model and latency. When a fact you expected wasn’t captured, this is the screen that tells you whether the classifier skipped it, the extractor produced nothing, or AUDM decided you already knew it.

Filterable by agent, so you can see exactly what Claude Code wrote versus what came in over MCP.

Agents

Every coding agent Sigil detected, its wiring status, and per-agent activity. Re-connect a client from here after a Node version switch or a reinstall — the same thing sigil connect does from the terminal.

Devices

Paired machines, their roles (reader / writer / admin), and pending pairing codes. Create a code here or with sigil pair create; revoke one the same way. See Multi-device.

Engine

The machine room. Worker pool state, queue depth (tasks waiting, tasks in flight), ingestion job status, and database connection detail. If ingest feels slow, the answer is visible here — usually queue depth against ingest.workerConcurrency.

Also where the managed-session engine reports itself when enabled.

Settings

Every value in config.json, grouped by category and split into two tiers:

• General — everyday choices: your name, default namespace, fact extraction, synthesis, update checks
• Advanced — retrieval thresholds, ranking weights, engine internals

Advanced settings fail quietly

A wrong value in the Advanced tier degrades recall rather than throwing an error. Raise memory.skipThreshold too far and near-duplicates pile up; drop memory.injectionFloor too low and irrelevant facts get injected into every prompt. Each control documents its default — Configuration has the reasoning behind the tuned values.

Settings that need a daemon restart are marked as such and restart it for you when you apply them.

The Danger zone at the bottom holds the destructive operations — resetting the database, disconnecting every agent. Both have terminal equivalents (sigil reset, sigil uninstall).

Changing providers later

Settings → Change provider re-runs the same tested step from setup: pick a provider, enter a key, live-test it, apply, restart. Switching your embedding provider is the one change with a data consequence — if the existing vectors came from a different model, Sigil reports exactly how many rows are affected and makes you choose before touching anything.

If the dashboard won’t open

sigil daemon status    # is it running?
sigil daemon logs      # what did it say?
sigil daemon restart

More in Troubleshooting.