_lt_

I’d like to use the mod if possible.

I found you off the other discussion and honestly this looks very promising. I loved the other mod and to see someone passionate taking the mod to the next level is great imo. I would love to see where this goes with some time and love from a new developer. I am here for the ride and wherever that takes us.

I have had so much fun with the updates that I will release it privately to those who have reached out and registered interest. I’m still working pretty hard on it so you’ll need to wait a bit.

1. AI Optimisation First I’ve completely reworked and optimised the AI before even attempting to add squad functionality. The original Tarkov AI was poorly optimised, resulting in low PMC counts and no team coordination. This meant a well-coordinated player squad using proper fire team tactics - covering lanes, clearing rooms - would easily wipe them out, making the game too easy.

2. Increased AI Budget and Squad Tactics To fix this, I optimised the AI system to effectively double the number of AI units a map can handle. I then created PMC squads that spawn in groups of 2-5 players (with solo spawns being rare). These AI squads now use upgraded tactics and proper coordination, so when players bring in their own squad, they’ll face a real challenge against other organised units.

Updates:

Current Foundations

• Baseline SAIN AI: Full SAIN source (MIT) builds cleanly on netstandard2.1 with BepInEx/Harmony integration. Benefit: no decomp bugs, stable updates.
• BigBrain Layers: Vanilla SAIN combat/squad layers compiled and ready for extension. Benefit: proven tactical logic as the base.
• Config & Presets: SAIN preset system intact; ready for new toggles and cluster/LOD controls. Benefit: tweak live without restarts.
• Cluster Scaffolding: Global ClusterManager/PerceptionCluster/TargetSnapshot wired into the plugin loop with per-frame/per-second budgets. Benefit: squads share intel and stay within a hard perf envelope.
• Path Budgets + Shared Route Cache (new): Path/cover queries are budgeted and now cached per-cluster (TTL/size-limited) before hitting NavMesh. Benefit: fewer navmesh builds, smoother movement, less “stuck until aggro.”
• Room-Sweep Layer (PMC-only): BigBrain sweep layer live for PMCs (scavs excluded per design). Primary sensor runs budgeted LOS probes to mark sectors; non-primaries consume shared intel (no LOS) and move via scored cover with path budgets. Benefit: believable entry/clear behavior without extra CPU spikes.
• LOD Throttling + Budget Telemetry: Vision publish rate scales by distance/engagement; budget sampler logs avg/peak raycasts/cover/path every ~30s (DebugMode). Benefit: predictable perf, easy profiling.
• Live Editor Toggles (F6): Home tab toggle for Room Sweep and Global Debug Mode without touching files. Benefit: test/iterate mid-raid.

In Progress / Planned (to align with DESIGN_INTENT)

• Cluster Perception: Primary-sensor rotation, shared TargetSnapshots, and raycast/path budgets to cut redundant LOS work.
• Multi-Target Tracking: Per-threat snapshots with per-enemy timers; non-primary bots consume cluster intel.
• Execution Wiring: Map SAIN decisions through BigBrain to movement/cover/fire actions with budget gates.
• LOD & Budgets: Frame/second budgets for heavy ops (LOS, cover/path); distance/engagement-based update throttling.
• LLM Gates: Feature flags to swap decision providers later without changing Tarkov-facing code.
• Coordinated Maneuvers: Cluster per-threat roles (primary sensor, flank, suppress, breach) so squads act in concert.
• Adaptive Aggression: Budget/confidence-aware aggression-push when intel is fresh and budgets allow, hold/search when stale.
• Shared Cover Caches: Per-cluster cover caches so bots reuse found cover and leapfrog efficiently.
• Hearing-Led Reactions: Cluster-published audio cues (shots/doors/explosions) wake bots and orient them before visual contact.
• Intent-Aware LOD/Stance: LOD/stance tied to current intent-high intent = faster ticks/aggressive stances; low intent = slower ticks/cover bias.

Future Ideas

• Fireteam Scaling: Split large squads into fireteams with shared perception and high-level intents.
• Advanced Cover System: Shared cover caches per cluster; coordinated cover-to-cover movement.
• Safety & Robustness: Defensive null checks, graceful fallbacks, and low-noise diagnostics under debug flags.

I just read the discussion on Discord and I can’t believe it, I’m perplexed.

On discord I’m alec_wan.kenobi if you wanna chat there

I’m trying to figure out how to start a “conversation” with ya, but this website’s a bit unintuitive to me too. for joining the SPT discord, I think you’d wanna start here: https://sp-tarkov.com/#community. There’s probably some “welcome channel” emoji reacts you gotta do to set your region, notifications, and acknowledge the rules - I expect this unlocks the other channels.

I’m assuming haha I joined the discord years and years ago so I don’t truly remember