By Peak Studios · Open Source

FRIM

FiveM Runtime Intelligence Model

An open-source, local-first AI reliability engineer for FiveM servers. It observes your server, understands its architecture, detects problems, identifies root causes — and verifies the fix.

View on GitHub Join Discord

8–14B

Model Parameters

80+

Prometheus Metrics

Release Phases

How it works

Not just a fine-tuned chatbot

FRIM combines deterministic analyzers with a domain-specialized language model. The analyzers handle facts. The model handles reasoning. Every claim cites a source.

Sources

Evidence & Graphs

Reasoning

Output

FiveM Server

Logs · Prometheus · Perfetto

Static Analyzers

Lua · JS · TS · C# · SQL

Cfx Knowledge

Docs · Natives · Changelogs

Evidence Store

Normalized · Local-first

Resource Graph

22 node types · 13 edge types

Knowledge Graph

Versioned · Platform-aware

Incident Engine

Root-cause ranking

FRIM LM

8–14B · Tool-augmented

Server + CLI layer

Graph + Evidence

Knowledge layer

LM handles reasoning

Natural-language reasoning, hypothesis generation, cross-file explanation, patch generation, and report writing.

Analyzers handle facts

Parsing, dependency resolution, native validation, event-flow analysis, data-flow analysis, metric calculations, trace extraction.

Local-first by default

Source code, logs, and database rows stay on your machine. Remote inference and telemetry are opt-in.

Evidence-linked output

Every concrete claim cites a source line, log entry, metric reference, Perfetto trace span, or graph edge. No hallucinated facts.

Capabilities

A foundation model for FiveM

Inspired by DeepSeek, Qwen, and Mistral — open weights, community-first, API for convenience. FRIM is multimodal, conversational, and tool-augmented.

Natural Language

Ask in plain English. 'Why did the server crash?' or 'Can players dupe items?'

Multimodal Input

Feed screenshots, Perfetto traces, or crash images alongside your log files.

Code Generation

Generates minimal, verified patches. Every patch must pass 5 gates before being marked safe.

API Access

REST API for hosting providers, CI/CD pipelines, and community tools. Self-host or use our cloud.

Live demo

See it in action

The CLI is the first interface. Five commands, five jobs — each one dramatically more useful than a FiveM chatbot.

frim — bash

frim scan

Parse all resources. Build dependency graph. Detect invalid manifests.

frim audit security

Trace untrusted values from client events to money, inventory, and permission operations.

frim audit enhanced

Apply 8+ Enhanced compatibility rules. Know your migration score before upgrading.

frim diagnose

Correlate logs, Prometheus metrics, and Perfetto traces. Rank root causes by evidence score.

frim patch + verify

Generate minimal patches and gate them through 5 automated verification stages.

5 core capabilities

Built from the ground up for FiveM

Resource Architecture Graph

Every resource, file, function, event, export, native call, state-bag key, and database table mapped as a structured property graph. Answer questions no generic AI can.

Runtime Diagnosis

Ingest Prometheus metrics (80+ in Enhanced), Perfetto traces, server logs and resource lifecycle events. Build incident timelines and rank root causes with evidence scores.

Defensive Security Audit

Trust-boundary graph from client-controlled sources to sensitive sinks. Every finding cites exact file and line. Defensive analysis only — no exploit tooling.

Enhanced Compatibility

YAML rule engine applied before the LM. State-bag replication, Mumble API deprecation, .NET 10 sandbox, OneSync big-mode, pure-mode — 8+ built-in rules.

Verified Patch Generation

5-gate pipeline: parse → FiveM semantics → security regression → resource contracts → incident replay. A patch is only marked verified when it passes every gate.

FMRB Benchmark

The FiveM Reliability Bench — an independent standard for measuring whether any AI system genuinely understands FiveM. Compare FRIM across leading frontier and open models.

Roadmap

6 phases to a fully intelligent FiveM AI

First public release of Phase 1 (Static Scanner): Q4 2026. FRIM Core model: Q2–Q3 2027.

PHASE 0

Benchmark & Schemas

In ProgressQ3 2026

Foundational schemas, incident formats, and the FMRB benchmark.

Resource graph schemaIncident formatFMRB benchmarkSynthetic fixtures

PHASE 1

Static Scanner

PlannedQ4 2026

CLI tool: frim scan, graph, audit security, audit enhanced.

Lua / JS / TS parsersResource graph builderSecurity taint analyzerEnhanced compatibility rules

PHASE 2

Runtime Doctor

PlannedQ1 2027

Live server observability — Prometheus, Perfetto, incident timelines.

frim-agent sidecarPrometheus ingestionPerfetto trace parserIncident timelines

PHASE 3

FRIM Core Model

PlannedQ2–Q3 2027

Domain-specialized open-weight LLM (8–14B). Multimodal. Tool-augmented.

Training pipelineGraph-conditioned reasoningPatch generationAPI access

PHASE 4

Verification Sandbox

PlannedQ3 2027

5-gate patch pipeline: parse → FiveM semantics → security → contracts → replay.

Resource contractsIncident replay enginePatch validationStaging deployment

PHASE 5

Ecosystem

PlannedQ4 2027

VS Code extension, GitHub Action, Discord bot, txAdmin companion.

VS Code extensionGitHub ActionDiscord botFramework adapters

Access model

Open weights. API for convenience.

Self-host for free forever, or use the managed API. Inspired by DeepSeek and Mistral — community-first, commercially friendly.

Community

Freeforever

Self-host FRIM Core on your own hardware. Full capabilities, no limits.

Open-weight model (self-hosted)
Full static scanner (frim scan, audit)
Resource graph builder
FMRB benchmark access
Community Discord support

View on GitHub