Problem The default breaks down
Most people use multiple AI tools as isolated assistants. That creates duplication, contradictions, weak provenance, and repeated setup overhead.
Cross-model operating system
A shared source of truth for multi-model AI work.
Hellespont turns a spreadsheet into a disciplined operating layer for ChatGPT, Gemini, and Perplexity, so projects are built from structured evidence instead of drifting chat history.
Most AI work breaks when memory becomes vague, sources become detached, and each tool starts operating from a different version of reality. Hellespont solves that by using Google Drive and a structured spreadsheet as the canonical working layer. Models can think, research, verify, and build, but the system keeps the facts, decisions, and project structure grounded.
Operating principles
Five properties that distinguish Hellespont from prompt-only work.
-
01
Canonical storage
A single structured artifact replaces fragile chat memory as the working substrate.
-
02
Role separation
Each model is used for what it does best, with clear responsibilities and handoffs.
-
03
Traceable provenance
Sources are logged. Facts are promoted only when confidence and provenance are clear.
-
04
Questions stay visible
Unresolved issues and decisions are tracked separately from established truth.
-
05
Automation-ready
Designed to layer scripting and tooling later without becoming brittle.
How it works
Problem, solution, and how the three models divide the work.
Solution One shared memory layer
Hellespont uses one shared spreadsheet as the system memory layer. Sources are logged. Facts are promoted only when confidence and provenance are clear. Open questions stay visibly unresolved. Decisions are captured explicitly. Model-specific tasks are routed through a messenger layer.
Roles Three models, one contract
ChatGPT
Synthesis
Structuring, schema design, clarification, and controlled writing.
Perplexity
Research
Live research, discovery, external validation, and design and reference gathering.
Gemini
Stewardship
Google-native stewardship, Drive synthesis, sheet analysis, and later Apps Script or AppSheet support.
System architecture
The Hellespont Master workbook. Ten tabs, each a discrete system component.
Tab
Role in the system
Code
Control Tower
Top-level priorities and next actions.
01
Portfolio Map
Major domains and linked projects.
02
Project Registry
Concrete work items and their blockers.
03
Canon Facts
Approved, reusable truths.
04
Open Questions
Unresolved issues kept visible.
05
Source Register
Provenance and source records.
06
Decisions Log
Architectural and strategic choices.
07
Workflow Rules
Operating contract across tools.
08
Data Dictionary
Schema definitions and required fields.
09
Model Messenger
Structured handoffs between models.
10
Example workflow
Building a website about Hellespont, end to end.
- 1
Define the project
Register the work in Project Registry.
- 2
Gather sources
Collect internal sources and architecture notes.
- 3
Promote facts
Move only verified structural claims into Canon Facts.
- 4
Record open questions
Keep unresolved site questions visibly unresolved.
- 5
Capture decisions
Log scope and sequencing choices in Decisions Log.
- 6
Route research
Hand research tasks to Perplexity through Model Messenger.
- 7
Synthesize copy
Use ChatGPT to synthesize the copy and structure.
- 8
Review in Drive
Use Gemini to review Drive artifacts and suggest automation paths.
- 9
Publish
Produce a build brief and publishable website copy.
Why this beats prompt-only work
Prompt-only workflows are fast at the beginning but unreliable over time. They depend too heavily on transient context. Hellespont creates continuity, auditability, and cleaner collaboration between models. The result is slower to set up once, but much more stable over repeated use.
Prompt-only
- Implicit memory
- Weak provenance
- Contradictions between tools
- Repeated setup overhead
- Fast start, fragile over time
Hellespont
- Externalized, canonical memory
- Logged sources, promoted facts
- Clear contract across models
- Questions and decisions tracked separately
- Slower setup, stable at scale
Next step
Interested in building a cross-model operating system for your own projects, research, or portfolio?
Start with three commitments:
- Define a canonical source of truth.
- Assign a role split across models.
- Adopt a disciplined path for promoting sources into facts.