Changelog

Prisma native breakthrough — sub-millisecond decode & parse

• AddedDecoding and parsing breakthrough — Prisma native GGUF keeps both steps under 1 ms for final completions and ~5 ms on average during streaming, versus typical industry stacks that treat decode (115 ms) and parse (78 ms) as separate tooling passes.
• ChangedBenchmarks page updated with the new decode/parse results — integrated pipeline performance now reflected in per-step industry comparisons and summary callouts.

Guard Focus — AI environment security

• SecurityNew developer security methods and tools help ensure your environment remains safe and confidential from hackers, criminal enterprises, and other maliciously acting entities.
• AddedDeveloper security methods — /scan, /sanityCheck, /networkProtect, and /interface security collectors. /sanityCheck audits the last 60 minutes of file activity (PID, program, read/create/update/delete) across workspace, /Library, /System, and kernel logs, and detects user activity monitoring (mouse/keyboard, app launches, browser navigation, input-monitoring grants). /scan traces hidden files, directories, and their connections.
• AddedExpanded /interface scan — hidden screen sharing, data smurfing, hidden screenshots, and export/exfil detection.
• FixedGuard mode reply fixes — addressed empty thread output when the model returned only reasoning tokens; improved visible scan reporting.
• Fixed/scan stuck at “Starting system scan…” — slash menu and Methods sidebar now trigger the collector; added shell-io readiness, workspace resolution, and subprocess timeouts.
• Fixed“Could not append scan request” — broken fallback for slash labels (/scan, etc.) now resolves correctly.

Footer chrome & timeline — status bar, 90-day audit

• Added90-day development timeline compiled — Apr 11–Jul 10 window, 625 file touches, 15 active days.
• AddedComprehensive changelog audit — phase map from Foundation (May 1) through Prisma GGUF (Jul 10) reconciled to filesystem and transcript evidence.
• AddedNative inference engine — in-process GGUF backend via llama-cpp-4; GGUF set as default LLM backend.
• FixedFast-launch fix — renderer build checks, startup scripts, and model warm-path preload for quicker cold starts.

Prisma / inference-engine — relay-protocol & GGUF benchmarks

• AddedRelay protocol — Gemma 4 stream relay for the Prisma inference stack.
• AddedInference engine — in-process GGUF native backend; MO-Engine thread pool, decode, parsing, HF prep, and tensor accelerators.
• AddedGGUF benchmarks — Prisma native vs. stock llama-server baseline comparison.
• FixedRemote GPU deploy path hardened — server, deploy, and pod-sync scripts on the inference stack.
• AddedToken server auth — API-key validation, tokenizer seeds, and gemma-protocol binary on codeapp-server.
• SecurityPhyziro proprietary license; no prompt database, monitoring, or training on user data.

Localized LLM — always-on private AI on device

• AddedLocalized LLM capabilities for always-on AI and private AI development on the local device — run models on your machine without relying on remote inference for day-to-day coding.
• SecurityLocal inference helps protect against data exfiltration, remote entity project rebuilds for training, and other malicious intent — and removes the AI prompt training attack vector from your development workflow.

Architecture sprint — threading, shell-io, Monaco, accelerators

• AddedThreading — UI operations off main thread; Monaco WASM with dedicated Rust intensive thread.
• AddedShell I/O stack — safe storage, PTY, and Rust file-read fallback; shell UI panic fix.
• Fixedshadcn z-index stacking — settings dropdown placement and threads panel sash overlap.
• FixedMonaco editor — WASM loader path and drag-and-drop code blocks from agent chat.
• AddedAccelerators workspace — MO-Engine thread pool scaffold under Prisma utilities.
• SecurityProfiled hardening — evidence guard alerts, deny-delete guards, offline blind-spot detection.

Gemma-protocol & remote Rust — RPC layer

• AddedGemma-protocol RPC on codeapp-server — stream relay wiring for Prisma-to-Gemma 4.
• AddedGemma 4 message framing on remote GPU inference server.
• AddedHuggingFace tokenizer and inference prep in the gpu-rs Rust driver.
• AddedSettings fields framework — reusable input, checkbox, textarea, and select field components.
• FixedDiagnose wire complete — request-file-fixes command and tokenizer warm path on remote server.

Early agents & UI — tabs, chat panel, workspace flows

• AddedAgent tabs — multiple concurrent agent conversations with per-tab state.
• AddedChat panel — agent message stream, block preview, and connection error handling.
• AddedWorkspace flows — open-editor context, @-references, and focus-mode routing over agent-chat HTTP.

Public beta & profiled security complete

• AddedPublic beta — AI-native IDE with agent chat, inline completions, and interactive terminal.
• AddedPhyziro model attribution directive; multi-chat and 500k Collaborate context window (began Jun 9).
• SecurityProfiled stack complete — permission-escalation, file-tampering, Wi-Fi monitor, espionage scan, snoop-expose, evidence guard, opensnoop, protected paths.

Profiled monitor & containerization

• AddedProfiled monitor — JSON digests, Rust audit chain, and LLM token stream parsing; first development token logged.
• AddedContainerization — per-app POSIX users, 50 GB APFS volume, platform user, and per-service scaffold; developed with the beta on macOS devices.
• SecurityCoding focus diagnose-before-edit and apply-code workspace scan.

Platform — auth, web-search & LLM helpers

• AddedAPI-key validation on remote token server with encrypted payloads.
• AddedWeb-search parsing, LLM model calls, and search helpers; auth dashboard and remote environment provisioning UI.
• AddedLanguage selection — per-file language detection, routing, and content-type inference.
• SecurityAES-256-GCM payload encryption between IDE and remote token server.

Early agents & UI — IDE bridge; GPU scheduler

• AddedChat panel bridge — block preview artifacts and LLM connection error panel in agent UI.
• AddedWorkspace flows — Debug with AI five-phase pipeline; web-search approval and chat activity status.
• AddedGPU scheduler — partition and scheduling layer refactored on codeapp-server.

Platform & language refs — tokenizer paths

• AddedTokenizer paths on GPU server — encode/decode cache extending May 11 Prisma Protocol.
• AddedAgent task decomposition — shared task-splitting helper on the Prisma inference path.
• AddedGemma thinking-policy scaffold — per-lane thinking stream parsing (designed Jun 1–20; landed Jul 9).

Remote GPU server — inference worker complete

• AddedGPU inference worker subprocess on remote token server — completes May 11–Jun 11 remote server phase.

Platform & language refs — syntax references

• Added29-language syntax reference library — platform-wide diagnostic and linting reference covering syntax errors, logical errors, and other code-environment issues; runs automatically while the user is actively coding and performs an auto-review every 5–30 minutes to keep code up to date.

Early agents — multi-chat & context

• AddedMulti-chat per-focus caching (300 completions/thread); 500k Collaborate context window.
• AddedFull-file review lane began — completed July 6 diagnose wire.

Early agents — build automation & grounded chat

• AddedBuild step automation, Mermaid/UML diagrams, RAG verified-sources, and reasoning guards.
• AddedAuto-Solve — system method for having AI automatically solve problems, bugs, or issues surfaced in the terminal.
• AddedInspection panel; terminal approve-before-run.

Prisma Protocol — RunPod handoff

• AddedPrisma Protocol RunPod handoff atop May 11 remote token server.

Foundation & IDE shell — merge detection

• AddedEditor merge and conflict detection in the renderer.

Foundation & IDE shell — syntax/LSP, search, settings gutter

• AddedSyntax and LSP — language server session, JSON-RPC stdio bridge, manager, and terminal syntax hints.
• AddedSearch UI — find-in-workspace with configurable options panel.
• AddedSettings gutter — editor autosave and live settings input sync.

Remote GPU server — Prisma Protocol begins

• AddedPrisma Protocol for Prisma-to-Gemma began May 11, 2026 — OpenAI-compatible remote token server.
• AddedGPU inference server on remote server refined through Jul 9–10.
• AddedDeploy scripts — remote provisioning pipeline for GPU.
• AddedRemote Sync — workspace push to remote GPU environment.
• AddedRust gpu-rs driver — HuggingFace prep, tokenizer warm path, and inference lib on codeapp-server.
• AddedGPU inference worker subprocess — on disk Jun 11; completes May 11–Jun 11 phase.

Foundation & IDE shell

• AddedFile tree and workspace explorer (May 1–13) — left-sidebar project navigation with fuzzy search, symbol outline, and folder browsing; interactive tree UI completed July 9.
• AddedLanguage selection — per-file language detection and routing in the workbench (expanded Jun 20).
• AddedElectron + Monaco IDE shell — integrated terminal, Problems panel, custom syntax diagnostics, and Rust launcher.

Genesis — patent pending

• AddedWebFlows.ai patent pending July 23, 2025 — formal genesis of the product line.
• AddedPrior research — private agent orchestration, containerized inference, IDE-integrated AI; Genie cloud platform.
• SecurityCore IP posture: user prompts, code, and inventions remain customer property.