FAQ

Port in use

EYAS_PORT=3200 ./bin/eyas start or free the process.

The UI is not on port 3000

Default listen port is 3100, chosen so it does not collide with Grafana or Create React App on :3000. Open http://localhost:3100. Override with EYAS_PORT or server.port in YAML. Docker maps "${EYAS_PORT:-3100}:3100".

No UI

bun run build:web (auto on start unless EYAS_SKIP_WEB_BUILD=1).

/docs 404

bun run docs:build or restart without EYAS_SKIP_DOCS_BUILD. Package: packages/docs. Do not run generate-full-docs.mjs / bun run full-docs — it overwrites prose.

Provider auth error

Re-enter key under Providers/Secrets. For Claude Code, ensure claude is signed in in the same environment. Grok and Kimi are signed in for EYAS on their provider panel, not on the host.

Conversations keep reading my ~/.claude / ~/.grok memory

They no longer can. Claude Code always runs isolated — no host CLAUDE.md, settings, hooks, skills, MCP servers or auto-memory, and no transcripts under ~/.claude/projects; the old Load host Claude config switch is gone. Grok CLI and Kimi Code CLI run in their own EYAS home and never see ~/.grok, ~/.kimi or ~/.claude. The security gate refuses reads as well as writes of other tools’ memory for every model, and memory-store MCP servers are blocked. To bring that knowledge into EYAS, import it once with Settings → System → Data portability → Import data. See Providers — Claude Code isolation and Memory.

Grok or Kimi stopped answering after the upgrade

Grok CLI and Kimi Code CLI now run in EYAS’s own home, so the CLI’s login on the computer is not used. Sign in once for EYAS: Providers → Grok CLI / Kimi Code CLI → Sign in for EYAS (device code; Grok also takes an xAI API key). Until then the card shows Sign-in required and turns fail with … is not signed in for EYAS. See Providers — Sign in Grok and Kimi for EYAS.

A turn failed with “could not confirm that it runs isolated”

EYAS found something on the host that would break the CLI’s isolation — for example an extra MCP server, hook, plugin or skill, a system-wide Grok config, or a managed Claude Code policy that forces another permission mode. Remove the cause and send the message again; the turn is never handed to another model. See Providers — Isolation check and Claude Code isolation.

An MCP server shows “Blocked: memory store”

It keeps a second memory outside EYAS (Memory, Qdrant, Obsidian, MCPVault, …) or points at a protected folder, so EYAS never starts it. Edit it to point elsewhere, or delete it, and bring that memory in with the data import. See MCP.

Claude Code is installed but the provider is not available

EYAS needs the Claude Code binary to be signed in, not only on PATH: a claude.ai login, ANTHROPIC_API_KEY, or a Bedrock/Vertex setup. Check eyas doctor — the Claude Code runtime line shows which binary EYAS runs and whether it is signed in. A service’s PATH may lack claude; set EYAS_CLAUDE_CODE_BIN to its absolute path. After signing in, turn the provider off and on under Providers, or restart. See Providers — Claude Code runtime.

Durable notes are being written and I want that off

Set memory.capture.enabled: false in local.yaml (key path memory.capture.enabled, default true). Capture skipped because it is off writes no memory_capture_runs row. See Memory and Configuration.

Where is data?

$EYAS_HOME or cwd: data/sqlite, data/vault, data/agents, backups, logs. EYAS_DATA_DIR moves the whole data directory; the vault moves with it (<data dir>/vault). Conversation workspaces of a source install run from a git clone live in your user’s application-data directory — see Configuration.

I set EYAS_DATA_DIR and my memory notes are missing

Earlier versions kept the vault in <EYAS home>/data/vault even when EYAS_DATA_DIR pointed elsewhere. The first start after upgrading copies those notes once into <data dir>/vault, but only while the new vault holds no note. Run eyas doctor: its Vault line says whether a copy is pending, or whether the old folder is no longer used because both hold notes — then copy any note you still need by hand. See Configuration — Data directory and vault.

The model gets the wrong local time

Set i18n.timezone (an IANA name such as Europe/Berlin) in local.yaml and restart. Unset, EYAS uses the server’s zone — TZ, else the OS; containers are usually UTC. See Configuration.

FAQ