Roadmap

Monitoring

CPU, GPU, memory, storage, and temperature, sampled continuously through a single telemetry pipeline and reflected live on the dashboard.

Monitoring

Correctly identifies the real CPU package temperature sensor across both Intel and AMD hardware, instead of guessing from a single sensor name pattern.

Monitoring

Detailed processor, graphics, memory, storage, motherboard, BIOS, and Windows information, gathered on demand via a hardware scan.

Intelligence

A single authoritative 0-100 health score and rating, computed from live temperature, memory, storage, and load readings.

Intelligence

Deterministic, evidence-based recommendations. Each one states what was detected, why it matters, the recommended action, and a risk level. Nothing is ever applied without your explicit confirmation.

Intelligence

Per-component CPU and GPU temperature classification, plus cross-metric reasoning about whether a workload is CPU-bound, GPU-bound, or balanced, and whether cooling looks adequate for the current load.

Intelligence

Extending the same cross-metric reasoning approach used for thermal data to storage health/capacity and memory pressure.

Intelligence

Reasoning about Windows Update state, requiring new local-data capture comparable to the driver-version work behind the NVIDIA advisor.

INGRID

INGRID receives the system’s current live telemetry and health state directly, so her answers reflect what’s actually happening on your machine right now.

INGRID

Cross-source lookups for things local telemetry can’t answer alone, such as current NVIDIA driver advisories, layered on top of INGRID’s local reasoning.

INGRID

Anything outside INGRID’s recognized commands and reasoning modules is sent to a locally running Ollama model. If Ollama isn’t reachable, INGRID says so plainly.

INGRID

Talk to INGRID out loud instead of typing. Push-to-talk speech recognition (OpenAI Whisper, with a local Windows fallback) transcribes what you say; INGRID replies in Marin, a natural OpenAI voice, through the same reasoning pipeline as typed chat. Live-verified end to end with real hardware — real microphone input, real recognized text, real spoken replies.

INGRID

INGRID's reactor emblem and avatar react in real time to what she's actually doing — idle, listening, thinking, speaking, or creating — driven by one shared live state bus rather than separate, potentially-inconsistent visual layers. Same signal also drives the Studio launch cards' glow.

INGRID

INGRID keeps a running, queryable memory of what she just told you, and a separate live record of what she just searched — its sources, and which one you're currently focused on — so a real follow-up like "tell me more" or "open it" resolves against what actually happened, not a guess.

INGRID

A failed request — an unreachable model, a search that didn't come back — is tracked apart from conversational memory entirely, so it can never be recalled later as something that actually happened. Surfaced honestly in the moment, then retired the instant you move on to something else.

INGRID

A first slice of natural-language follow-up understanding — recognizing rephrasings like "give me the short version" or "go deeper on that" that don't match INGRID's original exact recognized phrases, without ever inventing a target she isn't confident about. Runs only as a fallback behind every one of INGRID's existing deterministic commands and reasoning stages, never ahead of them. Built and covered by automated tests; not yet in a shipped build.

INGRID

A camera input pipeline and Vision panel — live device enumeration, preview feed, resolution and measured FPS, and honest Offline/Connecting/Active/Error status. A new Observing avatar state reflects when INGRID is actively attending to the feed. Object recognition, scene understanding, facial tracking, text recognition, and emotion recognition are clearly labeled as coming soon, not yet implemented. Verified end to end against real webcam hardware, including correcting the image orientation the capture layer delivers.

INGRID

Real-time two-way voice conversation with INGRID from a paired phone, reached over INGRID Link on a private, encrypted network link — never the public internet — with her live camera feed staying up for the whole call. This is a call with INGRID herself, not a messaging or calling feature between other people. Live-verified end to end on real hardware: the call connecting, two-way audio, and both the call and its video automatically recovering after the phone locks or switches apps. Not yet part of a public build.

INGRID

The rest of the conversational-fluidity work beyond the first slice above — natural topic switching and correction, broader semantic-understanding coverage, and a full paraphrase-matrix validation pass with a live conversation smoke test. Not yet started.

INGRID

Letting you interrupt INGRID mid-response — voice or typed — instead of needing to wait for her to finish. Not yet built.

INGRID

A dedicated review pass over INGRID's conversational and reasoning pipeline — direction only; no design work has started yet.

Intelligence

INGRID's engineering core. A structured task/result model — objective, evidence, findings, confidence, verification status, risk-rated recommendations — with a provider-neutral interface behind it and a persistent engineering log. Claude is the first connected provider today; the architecture never hardcodes it. Live- verified end to end against a real Anthropic API call.

INGRID

INGRID recognizes an engineering or diagnostic request in ordinary chat — "why is my CPU temperature wrong," "review this code and find the bug" — routes it to C.O.R.V.U.S. automatically, and explains the reviewed result back in her own voice. Provider identity never surfaces to the user; findings preserve their real confidence and verification status rather than being smoothed into certainty.

Intelligence

Read Only, Edit, and Commit modes, chosen by the operator and binding on the session. In Read Only the editing, build and commit tools are never offered to the model at all, and nothing can escalate past the selected mode. The choice persists between sessions.

Intelligence

Forge is the agent; the model is a backend. It runs on a local Ollama model with no account or network access required, or on Anthropic's Claude or OpenAI when a harder task justifies it — selected from a menu, with an explicit priority order on Auto. The tools and safety rules are identical whichever is chosen.

Intelligence

An answer has to be earned. If a task names a file, that file must actually be opened before an answer is accepted — every file, when several are named. A read that fails counts as evidence too, so a question about a file that does not exist gets "it does not exist" rather than a confident description of its contents.

Intelligence

Search by content or file name, read files, list directories, and inspect Git status, log and diff — all confined to the opened workspace. After making a change, Forge builds the project and runs its tests before treating the work as done, and takes a checkpoint before it edits anything.

Intelligence

For the work a language model should not do in its head — arithmetic over a file, parsing a CSV, reshaping JSON, a quick chart — Forge can run Python in a restricted environment: no network, no launching other programs, file access limited to the workspace and a scratch directory, a time limit and a cap on output.

Intelligence

A provider-neutral seam for requesting concept art, mockups, icons, textures and diagrams, with generated files landing in one approved directory and never silently overwriting an existing asset. It currently produces local placeholder images; no hosted art service is connected yet.

Studio

Conversational image generation entirely through chat with INGRID — generate, restyle, vary, remove background, and resize for wallpaper — with no separate generate or style buttons; creative actions only happen through the conversation. Includes save/export and project/history management.

Studio

Connect your own OpenAI account so INGRID can route Art Studio image requests through it instead of the built-in placeholder — no manual per-request provider picking. Credentials are stored securely on your device, and if nothing is connected (or a request fails) INGRID falls back to the placeholder automatically. Real generation through a connected account has been verified end to end against a live OpenAI account.

Studio

Ask INGRID for a track (e.g. "create a song about summer nights") and play it back in a dedicated tab. Generation routes through your own connected ElevenLabs account; until you connect one in Creative Services, requests fail with an honest explanation rather than a placeholder result. Real generation through a connected account has been verified end to end against a live ElevenLabs account.

Studio

Architecturally scoped as the narrative counterpart to Art and Music Studio — its reactor/presence integration point is already documented — but no generation workflow is built or user-facing yet.

Studio

An early idea for a future creative module; no design or implementation work has started yet.

Intelligence

Extending C.O.R.V.U.S.'s engineering reasoning to real Windows-side evidence — logs, process and service state, duplicate-file analysis — as read-only input to an engineering task. No design work has started yet.

Intelligence

Extending the execution and rollback pipeline beyond temporary-file cleanup to Windows-side changes — services, startup entries, scheduled tasks. The pipeline itself is built and proven; what each of these still needs is its own validated, genuinely reversible implementation, added one at a time rather than as a generic "run something" hook.

Intelligence

Optimization History is currently in-memory and session-only, so a restart severs every history entry from the real backup it points at, leaving the backup orphaned and unreachable. Persisting history durably would let a backup remain rollback-eligible across restarts instead of only for the session that created it. Not designed yet.

Intelligence

The rollback direction's own durable half: keeping each OptimizationRollbackDescriptor valid and resolvable once history itself survives a restart, and giving BlueCrow a way to find and reclaim backups whose history entry no longer exists, rather than leaving them as silent disk usage. Depends on reboot-safe optimization history landing first. Not designed yet.

Intelligence

BlueCrow can now carry out a Safe-level Optimization Advisor recommendation, starting with temporary-file cleanup. Every run is one pipeline: your explicit confirmation, a safety check, a restore point, the change itself, then verification — and it is only reported as successful if that verification actually confirms it. A refused, cancelled, or failed run is recorded as exactly what it was.

Intelligence

Anything BlueCrow applied can be undone from Optimization History, through the mirror of the same pipeline — confirmation, restore, then verification. Rollback is offered only while the backup it depends on is genuinely still intact, so the button is never a dead end, and a rollback only counts as successful once it has been verified too.

Distribution

An application icon and a real installer, so BlueCrow Studio can be distributed outside of building it from source.

Platform

A cinematic public home for BlueCrow — introducing INGRID and C.O.R.V.U.S., an honest live roadmap, a development journal, and a real (not decorative) Early Access signup with email verification and a permanent Founder Number. Live in production, with the full signup and email round trip verified against it.

Platform

A premium /support page offering one-time and monthly contributions across five tiers, plus a custom-amount option, with PayPal Orders v2 wired server-side for order creation and capture. One-time contributions are live and verified with real captured payments; monthly/subscription contributions are built but not yet verified end to end.

INGRID

A richer character-animation layer beyond today's reactor/avatar glow states — natural expressions, eye movement, breathing, and gesture — plus an always- available overlay experience. Direction only; no implementation started.

Intelligence

Additional engineering and creative providers behind C.O.R.V.U.S. and the Creative Engine, coordinated the same provider-neutral way Claude is today. Direction only; no specific provider work has started.