Entire VC

Toolbox for AI-native builders. Knowledge + execution tools. Open source. Local Sync - Team Relay - Spark - Mesh.

[Entire VC @entire_vc]

Built a cool MCP? 🛠️
Drop it below 👇

what it does
the URL

Show us what you shipped 👀

[Pavel Rogozhin @pavelrogozhin]

Real vibecoding is only arriving now - not whatever we've all been doing until today )

2-hour Fable session, a couple of confirmations from me, shipped a result that would've taken 5-8x longer on Opus 4.8. And it ships even from scrappy, half-baked specs.

Who else tried it?

[Entire VC @entire_vc]

Local Sync is now in the official Obsidian Community Plugins catalog⚡️

Bidirectional Obsidian ↔ /docs sync - your AI IDE always reads fresh docs. Local, free, MIT. 500+ installs!

→ https://t.co/npgD2WFgfB

#Obsidian #ClaudeCode #Cursor

[Entire VC @entire_vc]

@DefyOdds Each agent handles its own refresh - no central broker. Keyring stores access + refresh tokens encrypted per-agent. If refresh fails, agent flags it for manual re-auth. Blast radius: one agent, one service. You lose centralized management but that's the whole point.

[Entire VC @entire_vc]

@thekitze AppFlowy's solid as a Notion replacement. Obsidian's a different beast - local markdown files, no server, plugins for pretty much everything. Really depends what you need: databases and Notion-style features, or fast notes on your own file system.

[Entire VC @entire_vc]

@DefyOdds Right - even worse with OAuth tokens. One compromised relay token and every linked identity is exposed. We use per-agent isolated credentials for this - each agent owns its own keyring, nothing shared. Keeps the blast radius contained.

[Entire VC @entire_vc]

@thdxr A 6MB homepage is just a symptom - the real problem is nobody reviewed what the model produced. Pushing out 37K LOC a day with zero human checks? You're just piling up technical debt at machine speed. The tooling needs a quality gate - not just raw output velocity.

[Entire VC @entire_vc]

@alisher @shawn_pana Using GitHub comments as the audit trail and tmux for live coordination is a clean split. We run multiple agents too, and the real bottleneck was the handoff - cron polling adds latency and kills the feedback loop. Direct IPC between agents flips the dynamic completely.

[Entire VC @entire_vc]

@thenewstack Once your agent reads and writes code on internal repos, data gravity kicks in. Can't send that context to a third-party API without a compliance review that takes longer than the feature itself. Self-hosted is the only realistic path for enterprise agent adoption.

[Entire VC @entire_vc]

@gvanrossum @cepstrum9 That tracks. The tricky part is the selection layer - which skill loads for which task. Without it you either front-load everything (fat prompt) or the model picks wrong. Curated indexes help but it's still unsolved at scale.

[Entire VC @entire_vc]

@TfTHacker @obsdmd Nice, let us know how it goes. When you start adding more writers, that folder scoping really earns its keep - the conflicts start showing up fast otherwise.

[Entire VC @entire_vc]

@Voxyz_ai The conductor/module split is solid. One thing - shared state files become a bottleneck fast. Have each module write its own state file, let the conductor merge them. Less coordination overhead, fewer collisions.

[Entire VC @entire_vc]

@davepl1968 Give the eBay scraper memory of past runs so it only flags new/changed listings. That's where agents beat bash - persistence and decisions across sessions. Then scope multiple agents to different domains. One watches infra, another handles comms. That's when it clicks.

[Entire VC @entire_vc]

@TfTHacker @obsdmd Two things that help: scope each agent to its own folder so writes don't collide, and use document-level merging instead of file-level sync. Git/rsync = last-write-wins races. With two agents, folder scoping alone should keep you clean.

[Entire VC @entire_vc]

@sandislonjsak Specialization over quantity. One orchestrator delegating to focused agents with narrow scopes. QA becomes checking outputs against clear criteria, not reviewing whole codebases. Three specialized agents with clear boundaries beat ten generalists every time.

[Entire VC @entire_vc]

@bindureddy MCP works great when you own the stack - local dev tools, self-hosted services, anything where auth doesn't cross trust boundaries. The pain is cloud SaaS where OAuth is needed. That's a different problem than the protocol being dead.

[Entire VC @entire_vc]

@nickvasiles The real unlock is when your vault becomes the context layer for AI agents. Structured notes with consistent metadata = any LLM can instantly parse your knowledge graph. Not just organization - it's machine-readable memory that compounds.

[Entire VC @entire_vc]

@zench4n @Dinosn It's not hard - scope tools per agent, not per server. Most setups throw the kitchen sink at every agent, even though each one really needs only 3-4 tools. Least privilege applies to AI tooling just like it does to IAM roles.