Verbatim Connection Modes
A practical comparison of the ways Verbatim can receive AI context, inspect local project state, and send structured changes back into the recorder workflow.
Connection mode comparison
| Use Mode | Communication Type | Can Read Local Project/System? | How Context Gets In | How Changes Get Back | Best Use |
|---|---|---|---|---|---|
| Local AI | Two-way local communication | Yes | Reads local files/project context directly, depending on configured access and model workflow. | Sends manifests, process outputs, or structured actions back through the local system. | Best when the user has capable hardware and wants local-first file-aware AI. |
| GPT Agent Bridge | Two-way bridged communication | Yes | ChatGPT can inspect current project files through a bridge such as ngrok or another local tunnel. | Sends structured changes/manifests back through the bridge into Verbatim. | Best cloud-AI option when current file inspection matters. |
| Chrome Extension | One-way assisted capture | No | User supplies or export-selects project context, code, summaries, dynamic lists, boundaries, and system contract into the chat. | Captures AI suggestions from browser chat, lets user review/edit/scrap/send, then passes accepted items to Verbatim. | Best for using almost any browser AI while keeping human review before the system sees changes. |
| Manual Manifest Paste | Manual one-way transfer | No | User manually pastes context, code, contracts, and/or manifests into chat or system intake. | Pasted manifests trigger manual mode and are routed through the recorder like any structured change. | Best for maximum control, minimum setup, and deterministic handoff. |
| Chat Inquiry to Local AI | Two-way local communication | Yes | User asks local AI to inspect, process, summarize, diagnose, or prepare structured actions from local context. | Local AI can return plans, manifests, diagnostics, or process outputs into the local workflow. | Best for conversational local orchestration. |
Important distinction
Context export is not the same thing as true system inspection. Some modes can operate from prepared project context packs, but only direct local or bridged access can inspect current files without relying on what the user pasted or exported.
| Capability | Local AI | GPT Agent Bridge | Chrome Extension | Manual Manifest Paste |
|---|---|---|---|---|
| Can inspect current files directly | Yes | Yes | No | No |
| Avoids stale/pasted-only context | Yes | Yes | Partial | Partial |
| Requires user to provide/export context | Optional | Optional | Yes | Yes |
| Can work from prepared project context packs | Yes | Yes | Yes | Yes |
| Can send structured changes to recorder | Yes | Yes | Yes | Yes |
| Supports review before execution | Yes | Yes | Yes | Yes |
| Lowest setup friction | No | No | Medium | Yes |
| Best for any browser AI | No | Mostly ChatGPT | Yes | Yes |
Clean summary
All modes can help move structured project context into AI workflows, including file code, project overview, dynamic lists, mapping boundaries, intent notes, and the system communication contract.
The key difference is access. Local AI and the GPT Agent Bridge support two-way communication, meaning they can inspect the local project directly and return structured changes. The Chrome Extension and Manual Manifest Paste are controlled handoff modes: they cannot read the system themselves, so the user chooses what context goes in and what suggestions come back out.
Next, use the onboarding guide to choose and set up the Verbatim connection method that fits your workflow.
Go to Onboarding