oh-my-laya
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
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": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| laya_tell_meA | Run a local typed decision with Laya-MLX. Use for bounded classification, ordered scoring, and yes/no probability. Question types are choice, score, and noul. Treat results as advisory; never use them as authorization for destructive or consequential actions. Unless advisor is provided, questions is required: an object keyed by question id. Each question requires type and instructions (use a string). choice requires criteria: a nonempty list of unique string labels or an object mapping labels to descriptions. score requires a nonempty criteria list ordered from lowest to highest. noul optionally accepts a criteria object with false/true descriptions. Use criteria, not options. Example: {"risk": {"type": "choice", "instructions": "Classify risk", "criteria": {"low": "Reversible documentation edit", "high": "Data loss"}}}. For model advice, omit questions and pass advisor with models and current_model. Each host-verified model has id, reasoning_efforts (least to most), and an optional user-approved tier: fast, balanced, strong. No model is switched. For Alpha Squad, also pass role and verified current_reasoning_effort. Non-auto execution roles also require the confirmed session execution_choice pair; omit it for auto and reviewer routing. Returned delegation parameters are recommendations for the host's spawn call, never execution authorization. The current session is not changed. |
| laya_advisor_preferencesA | Read preferences, or save ONLY after the user's Confirm and continue. always: ask on every assessment; conditional: ask only for high complexity, high risk or uncertainty; auto: accept within the user-approved ceiling. For auto, provide ceiling={model, reasoning_effort} and the host-verified, locally filtered models catalog. Auto uses only that model and at most that effort. Missing ceilings require setup; no automatic acceptance is allowed. These authorize recommendations, not actual switching or any other action. Omit policy to read. Unknown/unconfigured preferences default to asking. Optional squad={enabled: bool, reviewer: {model, reasoning_effort} or null} enables Alpha Squad's bounded spawn recommendations after explicit setup. Null reviewer uses the orchestrator; difficult reviews always use it. Omitted squad preserves its prior setting. This never authorizes task actions. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 2 tools
laya_tell_me runs a typed decision while laya_advisor_preferences reads/saves configuration policy, so their core purposes are distinct. The shared 'advisor' vocabulary and overlapping policy/recommendation language create minor potential confusion, but an agent can tell them apart.
Both names use a consistent laya_ snake_case prefix, which is good. However the second element is inconsistent in style: laya_tell_me is an imperative verb phrase while laya_advisor_preferences is a noun phrase, breaking a predictable pattern.
Two tools is thin for a server that apparently covers decision execution, model catalog advice, policy configuration, and squad setup. Much functionality appears overloaded into these two, which is borderline rather than comfortably scoped.
The domain (advisory decisions plus preference management) seems approximated by running decisions and configuring policy. There is no separate operation to inspect models, ceilings, or squad state independently, so several capabilities are only reachable indirectly.