# Product feature inventory and evidence ledger

> **Evidence scope:** The website review includes the designated application repository, its README and help files, and selected Pascal units. Application paths in the ledger are repository-relative. Source/help inspection supports the implementation descriptions below; it does **not** establish that a capability is release-ready or user-tested on every platform. On 2026-10-29, the canonical latest-release API `https://api.github.com/repos/olbrichattila/lazycodingagent/releases/latest` identified published, non-draft, non-prerelease tag **v0.1.39**, published at `2026-10-29T21:24:49Z`. Tag API: `https://api.github.com/repos/olbrichattila/lazycodingagent/releases/tags/v0.1.39`; release page: `https://github.com/olbrichattila/lazycodingagent/releases/tag/v0.1.39`; comparison/changelog: `https://github.com/olbrichattila/lazycodingagent/compare/v0.1.38...v0.1.39`. The six uploaded asset filenames and GitHub-published SHA-256 digests are recorded in `README.md` and `download.html`; no asset was downloaded and independently hashed. This is checked-in fallback evidence, not a guarantee it remains latest. The latest-release API still handles optional browser-time refresh, but browser/API behavior, application runtime/install behavior, and deployment remain unverified. No live application run, external GitHub community verification, or current DigitalOcean deployment verification was performed. Application sources were not changed.

## Status meanings

- **Implemented in reviewed source/docs** — source wiring and/or help describe the capability. This is not a release, runtime-test, or download claim.
- **In Development** — use only for an owner-confirmed current status. The README says AI/model execution is in active integration; because that statement's currency is unconfirmed, public copy must not claim overall readiness.
- **Planned / Coming Soon** — an explicitly future item or roadmap placeholder, not available functionality and not a delivery commitment.
- **Unverified** — no adequate evidence was checked for the claim, or a live/release/third-party state remains unchecked. This does not mean absent.

## Evidence inventory

Application paths below are relative to the separate `Lazarus/lazycodingagent` repository unless marked as website paths.

| Capability / fact | Classification and safe public wording | Evidence reviewed | Limits / further verification |
|---|---|---|---|
| Desktop application and technology | Implemented in reviewed README/source. Describe as a Free Pascal, Lazarus and LCL desktop coding workspace, not a browser application. | `README.md` (“Technology Stack”, “Architecture & Unit Structure”); `lazycodingagent.lpr`; `mainunit.pas` | No independent release/runtime check. |
| Workspace UI, editor and panels | Implemented in reviewed source/docs. Main window includes navigation, workspace project tree, chat, a single-file editor, Markdown preview, guidelines/utility panels and a terminal panel. | `README.md` §§ “Docked panels”, “Vertical Navigation Rail”, “Primary AI Agent & Chat Workspace”; `mainunit.pas`; `uNavigationPanel.pas`; `uMainFileEditor.pas`; `help/workspace-layout.md` (“Center area”, “Right and utility panels”) | Describe reviewed implementation, not parity across every target or screenshot availability. |
| Ask / Plan / Agent modes and permissions | Implemented in reviewed source/docs. Ask is read-oriented, Plan can save plans, and Agent exposes write/patch/shell tools. Writes and shell cwd are workspace-bound by default; outside-workspace modification is a per-session opt-in. | `help/chat-and-modes.md` (“Composer modes”, “Writes outside the working directory”); `help/how-the-agent-works.md` (“Modes vs file changes”, “Safety and sandboxing”); `help/tools-reference.md` (“Mode summary”, “Path rules”); `uAgentTypes.pas`; `uAgentTools.pas`; `uToolWriteFile.pas`; `uToolRunProcess.pas` | This is application-level path policy, **not** an OS security sandbox. Shell/write actions execute on the user's machine. Agent mode requires appropriate trust. |
| Agent execution | Implemented architecture in reviewed source/docs: per-session engine, background run thread, model/tool loop, cancellation and step limit. | `uAgentEngine.pas`; `uAgentRunThread.pas`; `uChatSessionHub.pas`; `help/how-the-agent-works.md` (“What runs when you send a message”, “The ReAct loop”) | Do not imply multiple collaborating agents. Describe recorded actions, not private hidden chain-of-thought. |
| Multiple sessions and execution detail | Implemented in reviewed source/docs: per-workspace sessions, per-session engines, persisted history, tool errors and execution detail/log UI. The interface can record prompts/context, model responses, tool arguments and results. | `README.md` §§ “Primary AI Agent & Chat Workspace”, “Architecture & Unit Structure”; `uChatSessionHub.pas`; `uChatSessionStore.pas`; `uAgentExecutionLog.pas`; `help/chat-and-modes.md` (“Debugging mode vs live trace”); `help/workspace-layout.md` (“Tools & Logs”) | No guarantee of exact replay, complete durable trace, or sanitized export. Execution Detail is not hidden reasoning disclosure. |
| Providers, models and image attachments | README lists model examples including OpenRouter slugs for Claude 3.7 Sonnet, Hermes 4 70B/405B, GPT-4o and DeepSeek R1. Help describes configured models and image attachment support (PNG/JPEG/WebP/GIF, up to four images, 5 MiB each, possible downscaling). | `README.md` §§ “Primary AI Agent & Chat Workspace”, “HTTPS for LLM providers”; `uAgentTypes.pas`; `uLlmClient.pas`; `uChatImageStore.pas`; `frmSettings.pas`; `help/chat-and-modes.md` (“Image attachments”, “Model selection”) | Name only cited examples; configured provider/model availability, capabilities, API costs and service behavior depend on configuration and third parties. Image handling requires a vision-capable model. |
| Workspace tools | Read/search/list, diagnostics, git inspection, plan/todo, patch/edit/write and shell tools are described with mode restrictions. | `uAgentTools.pas`; `uAgentToolBase.pas`; `uAgentToolPaths.pas`; `uTool*.pas`; `README.md` tool table; `help/tools-reference.md` | Do not call the tool system risk-free or fully sandboxed. Git inspection is read-only; shell and writes run locally. |
| Integrated terminal | Documented/implemented real PTY terminal, started in workspace root. | `README.md` “Integrated terminal”; `mainunit.pas`; `uTerminalPanel.pas`; `help/workspace-layout.md` (“Terminal”) | Windows requires Windows 10 1809+ for ConPTY. Linux needs PTY/libutil packages as detailed in README. No universal runtime guarantee. |
| Themes, project rules, debugging/learn options | README documents dark/light theme, per-workspace `.rules/*.md` text injected into mode prompts, debugging mode and Learn mode. | `README.md` §§ “Modern Dark-First Design”, “Debugging mode and project rules”; `help/chat-and-modes.md` (“Learn mode”); `uWorkspaceState.pas`; `uAgentEngine.pas` | Published descriptions should stay within documented settings; no claims about unsupported theme/plugin extensibility. |
| Platform build configuration | README and CI workflow configure Windows, Linux and macOS builds. CI publishes packaged releases (see Downloads row). README documents target-specific UI/terminal/build nuances. | `.github/workflows/build.yml`; `README.md` “Platform verification checklist”, “Building & Running”, “Integrated terminal” | Published packages and per-machine install/runtime parity have not been verified by this website review. |
| Multi-agent workspace, execution plans, workspace knowledge/RAG, The Brain | **Planned / Coming Soon**, with roadmap placeholders called out as having no backend. | `help/workspace-layout.md` (“Future Features”); `future-plan/Lazarus_AI_Agent_Future_Plan.md` | Proposed future work only; no dates or commitments. |
| Embedded SQLite/vector retrieval | **Future planning**, not implemented. | `future-plan/Lazarus_Embedded_Storage_and_Vector_Plan.md` (plan explicitly says not implemented) | Do not market as a current feature or dated commitment. |
| Product integration maturity | Unverified/current owner confirmation required. README says “Multi-window workspace layout implemented. AI/model execution is in active integration.” | `README.md` status line; implementation sources listed above | Source exists, but the README maturity statement is ambiguous/staleness is unknown. Owner must confirm before any readiness or release claim; no application run was performed. |
| Downloads and releases | **Verified fallback snapshot; current status may advance.** Checked-in page values are v0.1.39 links and six GitHub-published SHA-256 digests. On 2026-10-29, the canonical latest-release endpoint identified published, non-draft, non-prerelease v0.1.39 (`published_at: 2026-10-29T21:24:49Z`) and listed six uploaded platform artifacts. Tag API: `https://api.github.com/repos/olbrichattila/lazycodingagent/releases/tags/v0.1.39`; release page: `https://github.com/olbrichattila/lazycodingagent/releases/tag/v0.1.39`; comparison/changelog: `https://github.com/olbrichattila/lazycodingagent/compare/v0.1.38...v0.1.39`. The six verified artifacts are Windows installer `LazyCodingAgent-0.1.39-win64-Setup.exe`, macOS Intel `LazyCodingAgent-0.1.39-x86_64.dmg`, macOS Apple Silicon `LazyCodingAgent-0.1.39-aarch64.dmg`, Linux amd64 Debian `LazyCodingAgent-0.1.39-amd64.deb`, Linux x86_64 AppImage `LazyCodingAgent-0.1.39-x86_64.AppImage`, and Linux arm64 Debian `LazyCodingAgent-0.1.39-arm64.deb`; exact direct URLs and GitHub-published digests are centralized in `README.md` and `download.html`. | Website `download.html`, `assets/latest-release.js`; latest API `https://api.github.com/repos/olbrichattila/lazycodingagent/releases/latest`; tag API and release page above; six uploaded artifact names, URLs, states and GitHub-published digests recorded in `README.md` / `download.html` (checked 2026-10-29) | GitHub API release metadata/assets/digests were checked, but files were not downloaded and rehashed. Latest status may have changed since the 2026-10-29 snapshot. The browser-time latest-release API refresh and deployed-site behavior remain unverified. Release metadata does not establish artifact install/runtime success or deployed-site/browser behavior. Installation, runtime, and code-signing beyond GitHub digest metadata remain unverified. |
| License and community channels | **Unverified.** Do not claim a license, open-source status, contribution process, Issues/PRs/Discussions availability. | No license/contribution evidence or current GitHub feature state verified in this review. | Direct repository and enabled-channel verification required. The website's repository URL is owner-provided, not externally checked. |
| Website privacy and hosting | Authored pages are static HTML/CSS/local artwork; the download page also loads a small script that optionally requests `api.github.com` latest-release metadata. No authored forms or analytics/advertising scripts were observed. Production host behavior is unverified. | Website source: `privacy.html`, `download.html`, `assets/latest-release.js`, `assets/site.css`; `DEPLOYMENT.md` is conditional guidance. | No guarantee about GitHub data handling or host/CDN/logging, DNS, cookies or retention. Confirm privacy contact and legal requirements before launch. |

## Verification gates before launch

1. Owner confirms whether the README's active-integration status is current and approves a precise maturity statement.
2. Recheck the current GitHub latest release, artifact names/download URLs and SHA-256 digest values before future fallback changes, then test deployed page/API behavior. The v0.1.39 fallback is a dated GitHub API snapshot from 2026-10-29, not proof it remains latest or that browser refresh, installation/runtime, or deployment work.
3. Verify GitHub repository status, license, contribution guidance and enabled community channels directly.
4. Confirm rights to publish derived product descriptions/artwork and whether curated help material may be reproduced.
5. Confirm privacy contact and production hosting behavior, plus actual DigitalOcean repository/branch, source/output settings, domain, HTTPS and 404 routing.

Record dates, scope, and evidence when an owner performs these checks. Keep every unverified gate open until then. This ledger belongs to the website project. No application files were changed. Evidence indicates implementation in reviewed code/docs only, not shipped/released/user-tested status.