01 / Question
Start with a real constraint
A slow workflow, an unreliable tool or a question from actual use. Record who needs the outcome and what would count as an improvement.
What stays behindProblem + success criteria
Engineering build journeys
Build log.
Latest builds, lessons and progress across my projects.
Subscribe via RSS →About these build logs
Inside the process / 01—04
Build something useful.
Keep what it teaches.
ChaseInTech is the public record. ChaseOS provides the context and governance behind the work. Each build connects a real problem to a result you can inspect.
- 28
- build journeys
- 10
- projects
- 113
- skills indexed
The working loop
From a question to a reusable lesson.
Illustrated workflow · select a stage to explore it
02 / Research
Find evidence before answers
Read primary sources, inspect existing systems and compare approaches. Keep dates, source links and uncertainty attached to each claim.
What stays behindSources + competing hypotheses
03 / Build
Make the smallest useful version
Turn a hypothesis into a bounded implementation. Preserve the decisions and trade-offs so the next change has a starting point.
What stays behindCode + decision record
04 / Review
Test the result against the need
Check behavior, inspect the rendered experience and record failures. A successful local test and a verified production release are different milestones.
What stays behindTests + visual proof + limitations
05 / Retain
Leave something the next build can use
Write the milestone, connect it to its project and keep the reusable lesson. Private knowledge promotion and public publication each need their own review.
What stays behindBuild log + linked lesson
How the pieces fit
One brief. Bounded work. A human decision.
The products below exist; this complete cross-product flow is a proposed pattern, not a live run.
01 / DirectionHuman brief
Outcome · constraints · evidence needed
02 / CoordinationChaseOS
Retrieve context · scope the task · check permissions
Research workInspect sources
Claims and uncertainty
Implementation workBuild and test
Changes and test evidence
Review workChallenge the result
Failures and acceptance checks
03 / DecisionHuman review
Accept · revise · stop
Approved action + receiptReviewed lesson + source
Existing foundation
ChaseOS Core
Runtime profiles, scoped permissions and an explicit approval gate.
Inspect the architecture ↗Existing projectChaser Agent
Source-linked work with claims, evidence, uncertainty and an activity record.
Inspect the project ↗Existing projectAgent Review Studio
Review recorded runs and preserve ratings, decisions and corrections.
Inspect the review loop ↗Use one worker for a narrow task. Split independent work only when coordination helps. Retrieved context informs a decision; it does not grant permission to act.
A map with provenance
Connect the lesson to the evidence.
A mind map explains the relationships. The linked records make them inspectable.
01 / Where it beganProblemUser need + constraint
02 / What informed itResearchPaper + date + claim
Build recordProject · decision · source
03 / What changedImplementationCommit + test + screenshot
04 / What carries forwardLessonTrade-off + limitation
Current project evidence
From a visual map to useful context.
ChaseOS project records show a graph of documents, projects and runtime profiles. A future worker can use reviewed, relevant records to orient its work.
The proposed next step is a repeatable path from build evidence to reviewed knowledge. Automatic ingestion of every build log is not verified here.
Capture→Review→Retain→Retrieve
Evidence behind the direction
Research informs the build. Tests decide.
Selected foundations and recent directions · sources checked 9 September 2026 · not an exhaustive survey
Research paper · NeurIPS 2020
Give generation a source to consult.
Lewis et al. combine retrieval with generation for knowledge-intensive tasks. This motivates source-linked context; it does not establish the performance of ChaseOS.
Read the RAG paper ↗Engineering practice · Dec 2024
Choose the simplest useful workflow.
Anthropic distinguishes predefined workflows from model-directed agents. Use parallel workers where the task separates cleanly; extra coordination has a cost.
Read the architecture patterns ↗Engineering direction · Apr 2026
Separate context, reasoning and execution.
Managed Agents separates the session record, harness and execution environment. The design lesson here is recoverable records and replaceable workers—not an integration claim.
Read the engineering report ↗Research report · 2026
Evaluate the team, not only each agent.
Anthropic’s AI-organization experiments found effectiveness and alignment trade-offs in tested settings. Outcomes depend on the model and setup; evaluate the complete workflow.
Read the study and limitations ↗How a trend earns a place in the workflow
Source → hypothesis → small experiment → measured result → review. A paper or provider report is a reason to investigate, not proof that the same result holds in my systems.
Opus Vela Space: from a storage map to a signed Windows release
Building a local storage explorer meant fixing the map, the restart path and the delivery path—not just drawing a polished interface.
Open the build journey →
Visual proof LaunchReady: checking a vibe-built app before real users arrive
How Product 4 became a working app launch check kit, a prominent commercial walkthrough, and a live £59 buying route without pretending a public listing proves paid-buyer delivery.
Open the build journey →
Visual proof Inside LaunchReady: making 111 launch checks easy to navigate
How a 111-deliverable release system was made navigable for agentic builders while keeping the design, executable harness, publication and buyer-delivery proof states separate.
Open the build journey →
Visual proof ChaseOS Studio V1.1.0: binding the interface, installer and update feed to one release
How V1.1.0 made knowledge intake inspectable and tied the public download, updater, installed app and visual evidence to the same signed Windows release.
Open the build journey →
Visual proof ChaseOS Studio V1.0.9: making knowledge review and local control inspectable
How V1.0.9 tightened the Intake and Research review flow, made approval origins legible, and reconciled local runtime state without presenting observation as a live connection test.
Open the build journey →ChaseInTech: turning a portfolio into a proof-led operating surface
How the site grew from a project list into a connected system for products, services, case studies, responsive media and evidence-backed enquiries.
Open the build journey →
Visual proof 20 Aug 2026 → 25 Aug 2026
ChaseOS Studio V1.0.8: proving the updater from an older install
Why a new version number was not enough—and how installer identity, public delivery and the real in-app update notification were verified as one release contract.
Open the build journey →
Visual proof Solo Operator AI Workspace: replacing promo graphics with product proof
How a terminal-heavy storefront identity became an editorial product story built from real customer artefacts.
Open the build journey →
Visual proof 21 Aug 2026 → 24 Aug 2026
Opus Vela Voice 0.6.2: one contract from installer to taskbar
How Voice joined signing, immutable downloads, updater verification, installed-payload parity and Windows taskbar identity into one release acceptance path.
Open the build journey →
Visual proof 20 Aug 2026 → 24 Aug 2026
MarginFlip: separating business performance from market research
Why a reseller's measured business performance and external market research needed different routes, data meanings and proof standards.
Open the build journey →
Visual proof ChaseInTech products: letting people inspect the work before buying
The step from static product cards to an inspectable, keyboard-operable proof reader with real downloadable samples.
Open the build journey →
Visual proof ChaseInTech services: from seven broad lanes to twelve bounded offers
How the service catalogue became twelve outcome-led offers with linked project proof and honest commercial boundaries.
Open the build journey →Visual proof ChaseOS Studio V1.0.7: requiring rendered evidence before release
How V1.0.7 changed packaged QA from a screenshot exercise into a rendered-page gate and proved the signed clean-install and updater path.
Open the build journey →Opus Vela: replacing placeholders with the products as they actually looked
A six-route media refresh that made Web, Studio, Voice, Pricing, Tools and Security tell one current product story.
Open the build journey →
Visual proof Project catalogue: making status, identity and responsive proof agree
The pass that reconciled public identities, honest release labels, real routes and 320-pixel project layouts.
Open the build journey →
Visual proof Mobile project media: turning screenshots into an inspectable gallery
How project evidence gained swipe, zoom, focus restoration and a usable phone-sized inspection surface.
Open the build journey →
Visual proof Mobile navigation: fixing the menu that disappeared after page changes
A small mobile-menu failure exposed a larger lesson about client navigation, replaced DOM and resilient event binding.
Open the build journey →
Visual proof ChaseOS Web: making Early Access a technical boundary
How the ChaseOS front door exposes a real waitlist and product story while keeping downloads, checkout, admin export and managed-agent claims behind explicit gates.
Open the build journey →GreyTheory Research: a signal is not a vulnerability
How GreyTheory separates raw signals, deterministic validation, model-assisted reasoning and human authority—before anything can become a reportable finding.
Open the build journey →TradeSync: making the preview more important than the execute button
How TradeSync separates scoring, risk review and execution authority—and why its public cockpit had to distinguish demo data from paper operation before live trading could even be discussed.
Open the build journey →Opus Vela Studio: designing the operation before the interface
Why Opus Vela Studio starts with a semantic kernel, typed operations and state-based verification instead of teaching an agent to click around a creative editor.
Open the build journey →Discovery and observability: proving the site could be found and measured
The operational step from a deployed website to a domain property with a readable sitemap and verified privacy-aware analytics.
Open the build journey →CI repair: the 18 pixels that only Linux could see
How an anonymous CI failure became a reproducible 18-pixel mobile overflow and a stronger browser-reporting system.
Open the build journey →Visual proof ChaseOS Cloud: designing the meter before the product
Cloud is planned, not live — but its billing architecture is already specified. Notes on the single-chokepoint meter, the reserve-execute-settle state machine, and why the local path is free by construction.
Open the build journey →Hypelist: one app, two storage modes, no cloud requirement
Why an inventory app for resellers supports both Firebase and SQLite behind the same Flask backend, and what the dual-mode abstraction costs.
Open the build journey →Chaser Agent: designing a harness that refuses to be trusted
Why the first version of Chaser Agent deliberately has no authority, and what a review-first output artifact has to separate to be worth reading.
Open the build journey →ChaseOS: treating agent governance as the product, not the paperwork
The approval boundary is the most-used surface in ChaseOS. Notes on why the governance layer earned first-class status instead of being bolted on.
Open the build journey →20 Mar 2026 → 13 Jul 2026
ChaseOS: turning build activity into native working knowledge
How build logs, runtime activity, evidence and reviewed lessons stay distinct—then become useful context for future agent work without silently becoming truth.
Open the build journey →Visual proof