For Game Studios & Indie Game Devs

NPCs that remember.
Worlds that prove themselves.

Persistent NPC memory across sessions. Cryptographically-signed anti-cheat audit trails. Replayable world state for adaptive questlines and tournament-grade speedrun verification. Infrastructure for AAA-grade game integrity, built for studios shipping AI-driven gameplay.

Unity / Unreal SDK in development. HTTP API + memory proxy available today.

The integrity layer modern games are missing

What this gives a game studio: a single piece of infrastructure that captures every NPC decision, every player action, and every world state transition into a tamper-evident chain. NPCs remember the player across sessions and shards. Cheats produce broken chains that the server detects instantly. Speedruns and tournament wins carry cryptographic proof. Adaptive AI-driven questlines can replay any decision point to design what-if scenarios deterministically.

The same product that compliance teams use for AI audit trails happens to solve four of the hardest unsolved problems in modern game design: persistent NPC memory, tamper-proof anti-cheat, deterministic replay for adaptive content, and verifiable speedruns. Same engine. Different application surface.

SteelSpine's hash-chained event log (HMAC-SHA256 + Ed25519) was originally built for AI agent audit trails. The same primitive that proves an AI agent's decision wasn't tampered with proves a player's actions weren't tampered with. The same memory proxy that gives an LLM persistent context gives an NPC persistent memory across game sessions.

What this looks like in your game

Two scenarios. Same SteelSpine infrastructure underneath. Both real CLI flows a game developer would execute.

steelspine — npc memory + anti-cheat flow

# Scenario 1: NPC remembers a player across sessions

$ steelspine start   # memory proxy on port 11435

✓ Auto-detected: local NPC LLM at port 11434 · proxy now injecting entity memory

# Player approaches NPC "firegiant" (session 4827, 12 days after session 247)

$ curl localhost:11435/v1/chat -d '{"model":"npc-firegiant","messages":[{"role":"user","content":"Greetings, FireGiant."}]}'

✓ Memory context injected: 18 past interactions with player_id=alice_4271

  → "Alice. Last time you tried to kill me with the flame-axe. I'm prepared this time."

# Scenario 2: Verify a tournament run for tamper-proof speedrun submission

$ steelspine verify-run player_session_891

✗ TAMPERED at event 1847: input timing chain broken

✗ Speed-hack detected: 3 inputs in 12ms (mechanically impossible)

  → Tournament submission rejected

  Clean speedrun submission would pass independently: pack-verify with public key

Memory and anti-cheat run on the same signed event chain. NPC interactions persist across sessions automatically. Cheat attempts produce broken chains that the server rejects before they affect game state.

Five use cases gaming has been trying to solve

Each one was previously expensive, fragile, or impossible. Each one is a first-class feature of SteelSpine's existing infrastructure.

01 / NPC Memory

NPCs that remember every player choice

Every interaction with an NPC is captured into a per-entity memory store. The NPC's context is automatically injected into every subsequent LLM call. Memory persists across sessions, server restarts, and player disconnects.

"FireGiant remembers you tried to kill him last week. He's prepared this time." "The blacksmith recognizes you from the tournament. He gives you a discount." "The dragon refuses to negotiate. You burned his hoard three patches ago."
How: steelspine memory-agent on port 11435. Any OpenAI-compatible NPC LLM call routes through. Entity context auto-injected. Per-entity disk snapshots at ~/.prime/entities/<npc>.json.
02 / Anti-Cheat

Cheats break the cryptographic chain. Instantly detected.

Every player action is sealed into the hash-chained event log on emit. Server-side verification of the chain proves nothing was modified, removed, or reordered. Cheats that modify client-side state produce broken chains the server rejects at the next sync.

"Speed hack? The timing chain doesn't hash-verify." "Inventory injection? The transaction chain breaks at the inserted item." "Aimbot? The input-event chain proves the inputs were synthesized."
How: steelspine emit on every player action. Server runs steelspine verify-run to confirm chain integrity. Tampering produces specific event-ID failure with no false positives.
03 / Evolving Questlines

AI-driven content that branches deterministically

Capture every player decision into the event chain. Replay from any decision point to simulate alternate paths. Build adaptive questlines where the AI dungeon master can deterministically explore "what if the player had chosen mercy instead of vengeance" to design content that responds to actual play patterns.

"The DM AI replays the player's path through Act 1. Designs Act 2 questlines that reference their specific choices. No two playthroughs the same."
How: steelspine replay-branch <event_id> reconstructs state at any decision point. steelspine branch-evaluate scores alternative paths against criteria. Use offline; no live API calls needed.
04 / World State Replay

One world. Many shards. One audit chain.

Cross-shard world state aggregation via the cross-adapter system. Every world event across every server lives in one canonical signed chain. Roll back, replay, or audit any time window. Disputes resolve cryptographically.

"Did the guild raid actually break the boss spawn? Replay the world state from 03:47 UTC. Verify the event chain. Make a final ruling."
How: steelspine cross-adapter-aggregate fuses events across data sources. state-at <event_id> reconstructs world state. Per-event public-key signatures survive any server going offline.
05 / Speedrun Verification

Tournament-grade proof, no client-server trust required

Every input + frame + game state transition cryptographically signed with the player's private key plus optional eIDAS-accredited RFC 3161 timestamp. Speedruns become mathematically verifiable. Tournament judges run steelspine pack-verify on the submitted run.

"The world record was set 14:32:08 UTC. Signed with the runner's key. Timestamped by Sectigo. Replays bit-for-bit. Accepted without dispute."
How: steelspine pack-create exports the run as a portable signed packet. Any third party with the public key runs steelspine pack-verify on their own machine. No trust in the runner or the tournament organizer required.
Bonus / AI Playtesters

Run agents through your game. Capture everything.

Wrap AI playtester runs the same way: steelspine run python3 my_playtester.py. Compare runs to find divergence. Replay specific failure points. Branch simulate alternate strategies. Eval gate the agent on regression before shipping the patch.

"1000 AI playtester runs against the new boss. The compare report shows three exploit paths. The patch ships with eval gating. CI fails if any of the three return."
How: Same steelspine run as any agent. compare --strict for CI gating. Plus Cursor / Claude Code MCP integration for game dev AI assistant access to playtest history.

Current state and SDK roadmap

SteelSpine's core infrastructure is shipping today. The gaming-specific developer experience layers are at different maturity stages. Honest map:

Capability Status How to use today
NPC persistent memory Available Memory proxy on port 11435. Any OpenAI-compatible NPC LLM call.
Cryptographic event chain Available HTTP API POST /emit on port 8765, or steelspine capture-pipe
Replay + branching Available CLI: steelspine replay-branch, simulate, state-at
RFC 3161 timestamping Available Auto-on with compliance_mode. Sectigo TSA default.
Unity SDK Roadmap Use HTTP API directly today. SDK shipping with studio partnerships.
Unreal Engine plugin Roadmap Use HTTP API directly today. Plugin shipping with studio partnerships.
Asset Store / Marketplace listings Roadmap Coming with SDK release.
Honest scope: if your studio wants to be a launch partner for the Unity SDK or Unreal plugin, schedule a consultation below. We're prioritizing real game integrations over speculative SDK polish. Studios that pilot get input on the SDK shape.

Schedule a studio consultation

30 minutes. No demo, no pitch. Tell us what your studio is building, what the integrity and persistence pain looks like at your scale, and which use case you'd start with. We'll tell you honestly whether SteelSpine is the right fit before either of us invests further time.

We respond within one business day. Your information is used only to follow up on this inquiry. Not added to any marketing list.

Prefer email? Reach us directly at hello@steelspine.ai.