The LATCH Console is a browser-based operator surface served from the Docker container on port 8091. It provides document management, inference controls, real-time telemetry, and portable workspace features — all without external dependency. This guide explains every configurable panel.
These controls govern how documents are compiled into persistent memory and how queries consume that memory. They appear in the Requests panel.
| Field | Description |
|---|---|
| Compile Memory Tokens | Memory tokens allocated per compilation. Leave blank for the runtime default (1024). Higher values increase fidelity at the cost of VRAM. |
| Query Memory Tokens | Memory tokens injected per query. Controls how much compiled knowledge the model sees per inference pass. Default 1024. |
| Query Max New Tokens | Maximum new tokens the model may generate per response. Default 1024. |
The Runtime panel displays the immutable defaults baked into the current image. Set at build time; they define the operating envelope for the licensed model tuple.
| Field | Description |
|---|---|
| Fallback Mode | Whether the runtime falls back to raw inference when compiled cache is unavailable. off in production. |
| Router Threshold | Quality router confidence threshold (0.0–1.0). Default 0.9. |
| Max Doc Tokens | Maximum source tokens per document. Currently 131072 (128K). |
| Default Memory | Default memory tokens for compilation and query. Currently 1024. |
| Memory Temperature | Temperature during memory compilation. 1 is standard. |
| Default Max New | Default max new tokens per response. Currently 1024. |
| Profile ID | Locked model tuple identifier. Checked during license validation. |
| Service Rev | Container build identifier with timestamp and hardware target. |
| Runtime Detail | Status message. ready to accept files for processing when engine is warmed. |
| Router Configured | Whether the quality router is loaded. not configured when inactive. |
| Config / Base Model Class | HF architecture IDs. Qwen 2.5 14B: Qwen2Config / Qwen2ForCausalLM. |
| Documents / Uptime | Live counters: compiled documents in memory and container uptime in seconds. |
The Decode panel controls sampling behavior during response generation. Baked defaults produce fully deterministic output.
| Field | Description |
|---|---|
| Do Sample | off for greedy decoding. Model always selects the highest-probability token. |
| Temperature | 0 = fully deterministic. Higher values add variation. Only active when Do Sample is on. |
| Top P | Nucleus sampling threshold. 1 = full vocabulary. |
| Top K | Top-K filter. 0 = disabled. Only active when Do Sample is on. |
The Persistence panel saves and reloads the full room state — compiled documents, settings, and session metadata.
| Field | Description |
|---|---|
| Workspace ID | User-defined save/load key (e.g., acme-q1). Unique per container. |
| Saved Workspaces | Dropdown of previously saved workspaces. |
| Save / Load / Delete | Save persists state. Load restores it. Delete removes it permanently. |
The Portable panel exports compiled document sets as encrypted, portable files that can be transferred between machines and reloaded without recompilation.
| Field | Description |
|---|---|
| Bundle Format |
.latch — Neural-only. Compiled memory with no source text. Smallest size. Share analysis without exposing documents. .latchdoc — Full package. Compiled memory plus source text. Enables full-text search and quality fallback. Recommended default. |
| Passphrase | Optional encryption. When set, the bundle can only be imported with the same passphrase. Leave blank for open team access. |
| Export | Exports compiled documents as a .latch or .latchdoc file. |
| Import | Loads a bundle. Documents become immediately queryable — no recompilation. Reload under 2ms. |
| Resource | Location |
|---|---|
| mike@codynamicslab.com | |
| Technical docs | huggingface.co/CoDynamicsLab |
| Website | codynamicslab.com |
CoDynamics Lab Corporation · Proprietary & Commercial License · Patent Pending — CDLaC/LATCH compilation method
The Room panel provides a live diagnostic snapshot. All values update in real time.
| Field | Description |
|---|---|
| Status | true when engine is warmed and accepting requests. |
| Profile / Service Rev | Active model tuple ID and container build identifier. |
| Model / Config | HF model class and configuration class. |
| Fallback | Current fallback mode. |
| Max Doc / Memory / Max New / Mem Temp | Active runtime parameters: document size limit, memory allocation, max response length, compilation temperature. |
| Docs / Ready Docs / Selected | Documents loaded, compiled and queryable, and selected for next query. |
| Uptime | Seconds since last engine start. |
| Query Scope | Query target: selected docs, all ready docs, or “No ready documents.” |
| Compiles / Clears / Queries / Batches | Cumulative success/total counters for the session. |
| Last Compile / Query / Reset / Error | Timestamps of recent events. Dash (-) = none. |
