Audrey
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AUDREY_HOST | No | REST sidecar bind address. Set to 0.0.0.0 only with AUDREY_API_KEY. | 127.0.0.1 |
| AUDREY_PORT | No | REST sidecar port. | 7437 |
| AUDREY_AGENT | No | Logical agent identity stamped on writes. | local-agent |
| AUDREY_DEBUG | No | Set to 1 to print MCP info logs (server started, warmup completed). Errors always log. | 0 |
| AUDREY_DEVICE | No | Local embedding device (gpu or cpu). Falls back to CPU if GPU init fails. | gpu |
| AUDREY_API_KEY | No | Bearer token required for non-loopback REST traffic. | |
| AUDREY_PROFILE | No | Set to 1 to emit per-stage timings via MCP _meta.diagnostics. | 0 |
| AUDREY_DATA_DIR | No | SQLite memory store path. Use one per tenant or agent identity for isolation. | ~/.audrey/data |
| AUDREY_LLM_PROVIDER | No | anthropic, openai, or mock. | auto |
| AUDREY_ONNX_VERBOSE | No | Set to 1 to restore ONNX runtime EP-assignment warnings (suppressed by default). | 0 |
| AUDREY_ALLOW_NO_AUTH | No | Set to 1 to allow non-loopback bind without an API key. Don't. | 0 |
| AUDREY_PROMOTE_ROOTS | No | Colon/semicolon-separated extra roots for audrey promote --yes writes. | |
| AUDREY_DISABLE_WARMUP | No | Set to 1 to skip background embedding warmup at MCP boot. | 0 |
| AUDREY_PRAGMA_DEFAULTS | No | Set to 0 to revert SQLite PRAGMA tuning to better-sqlite3 defaults. | 1 |
| AUDREY_EMBEDDING_PROVIDER | No | local, gemini, openai, or mock. Cloud providers require explicit opt-in. | local |
| AUDREY_ENABLE_ADMIN_TOOLS | No | Set to 1 to enable export, import, and forget routes/tools. Disabled by default. | 0 |
| AUDREY_CONTEXT_BUDGET_CHARS | No | Default Memory Capsule character budget. | 4000 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| prompts | {
"listChanged": true
} |
| resources | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| audrey-session-briefing | Start a session with an agent-scoped Audrey greeting and relevant memory packet. |
| audrey-memory-recall | Recall Audrey memories for a concrete question or action. |
| audrey-memory-reflection | Reflect at the end of a meaningful session and encode durable lessons. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| audrey-status | Machine-readable Audrey memory health, store counts, and runtime metadata. |
| audrey-recent | Recent agent-scoped memories for session bootstrapping. |
| audrey-principles | Agent-scoped consolidated principles and identity memories. |
TDQS
Scored across 23 tools
All tools share the memory_ prefix and many names describe overlapping cognitive operations (e.g., memory_consolidate, memory_reflect, memory_dream, memory_encode, memory_recall, memory_introspect) with no descriptions to clarify boundaries. An agent would struggle to choose between several of these tools.
Every tool follows the exact snake_case memory_<verb> pattern with a clear prefix and action verb. The naming convention is perfectly consistent across all 23 tools.
23 tools is on the heavy side for a single server, even for a memory-focused domain. The count is manageable but begins to feel bloated, especially with many near-synonymous operations.
Without descriptions, the intended domain is only inferable from names, which suggest a memory lifecycle. Obvious operations like search/list or a clear memory store are absent, while many speculative tools (memory_dream, memory_greeting, memory_capsule) exist, creating significant gaps in a coherent surface.