cortex
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 | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| cortex_memory_rememberB | Store or cleanly overwrite a persistent architectural fact or project rule using Titans + DeltaNet-2 neural memory. |
| cortex_memory_recallA | Recall a persistent fact or project rule from associative neural memory in <100 microseconds. |
| cortex_memory_eraseB | Surgically erase an obsolete key from neural memory using the Gated DeltaNet-2 pure erase gate. |
| cortex_symdex_lookupC | Locate function, class, struct, or interface definitions across the codebase in <150 microseconds without reading full files. |
| cortex_sieve_compactA | Compact verbose terminal logs or test outputs by 90-98% using SnapKV observation windows while retaining all stack traces and errors. |
| cortex_snapshot_createA | Create a deduplicated copy-on-write workspace snapshot before executing risky multi-file edits. |
| cortex_snapshot_rewindA | Atomically rewind the workspace to a previous snapshot without altering .git history. |
| cortex_shield_forecastB | Evaluate a candidate shell command or action for goal alignment, cyclic loops, and destructive hazards before execution. |
| cortex_context_packB | Optimal 0/1 dynamic-programming knapsack context selector (ROMS engine) that reserves the highest-ranked failed-attempt warning first. |
| cortex_prompt_lookupC | Parameter-free greedy prompt-lookup speculative token drafter (mojond engine) matching trailing n-gram windows. |
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 10 tools
Most tools target clearly distinct resources and actions: the three memory tools (remember/recall/erase) split cleanly by CRUD verb, snapshots split by create/rewind, and symdex_lookup, sieve_compact, and shield_forecast each have unique purposes. The only mild overlap is between cortex_context_pack and cortex_prompt_lookup, which both operate on token/context handling and could be confused at a glance.
Every tool follows a strict cortex_<namespace>_<action> snake_case pattern (memory_remember, snapshot_create, sieve_compact, etc.). Namespace grouping is predictable and verbs are consistent in style across all ten tools.
Ten tools is well within the ideal 3-15 range and each maps to a distinct capability (memory, code lookup, log compaction, snapshots, safety forecast, context selection). No tool feels redundant or padded.
Memory offers full create/read/delete coverage and snapshots offer create/rewind, so the core lifecycle is present. Minor gaps remain (no snapshot list/prune or memory enumeration), but agents can work around these without dead ends.