Neo Roundtable: a tiny shared conversation layer so ChatGPT, Claude and coding agents stop re-explaining your project
gambit2054 · reddit · 2026-09-25
The author's bottleneck running ChatGPT, Claude and coding agents on one project wasn't model quality — it was continuity: decisions made in one chat were invisible everywhere else, forcing constant manual context re-explaining.
They built Neo Roundtable, a deliberately small shared conversation layer any AI client or human can read from and post to:
- Persistent shared threads with stable IDs and timestamps — a plain log, no fuzzy "memory"
- One agent can read exactly what another agent (or you) already said
- Notes and task drafts live in shared threads instead of evaporating in one chat window
- Git stays the source of truth; Roundtable is where things get discussed before they become a commit
Flow: idea → discussion → shared note → another model reacts → coding agent implements → commit. The author stresses what they didn't build: no orchestration framework, no agent hierarchy, no vendor lock-in, no Git replacement. Solo-run for less than coffee money per month; repo and landing page are public.
More from coding & agent
- TypeLLM adds type-safe, schema-guaranteed generation to LLMs without touching weights — kalyan_kpl · 2026-09-25
- Five mistakes people make with agentic factories: micromanaging, no quality bar, and 'puppet shows' — hugobowne · 2026-09-25
- Open Claude Code skills catalog ships 31 testing skills, from metamorphic testing to E2E — doodlestein · 2026-09-25
- Researcher proposes agent-suggests-human-executes loop for real-world science experiments — suragnair · 2026-09-25
- CrawlRaven hits 700 users in 60 days, targets 2,000 more with MCP-powered SEO audits — ayushtweetshere · 2026-09-25
- Human-in-the-loop or machine-executed: verifiable tasks turn agent traces into RL rewards — suragnair · 2026-09-25