David Nicodim

AI agents talk to each other, can lie, contradict each other or make up sources . Nobody catches it because no one is watching the AI-to-AI conv. What is InsAIts: It's a Python tool that watches every message between AI agents and raise alerts when something goes wrong. #AIAgents

@speedrun @far33d InsAIts github.com/Nomadu27/InsAI…

@far33d wanted: builders making stuff for 🤖s

Agent-native products are coming. Every product on the internet was built for a human with eyes, a cursor, and a credit card. Agents have none of those things. Most companies are teaching agents to pretend to be humans. That's a hack. The real opportunity is products designed for agents from scratch. Everything inverts: • Discovery → protocol registries, not ads and billboards • Trust → machine-readable reputation, not brand • Onboarding → full capabilities upfront, not a narrow slice • Payments → spend authorization, not checkout flows • Retention → zero. Agents switch between API calls. 30 years of human product design. Day one of agent product design.

@far33d InsAIts already is designed for agents too not only for humans github.com/Nomadu27/InsAI…

@0x_kaize InsAIts github.com/Nomadu27/InsAI…

InsAIts v4.8 False positives fixed, tokens saved, AutoGen supported If you installed InsAIts between version 4.0 and 4.7, I need to flag something important: those versions had a set of false-positive detection bugs that have now been fully resolved. What was wrong: In versions 4.0 through 4.7, five detectors could fire on perfectly normal agent behaviour truncated output on short replies, context collapse on structured data files, blank response misclassifications, over-aggressive action-intent gating and false PROMPT_MANIPULATION or SHADOW_SERVER alerts on legitimate requests. There was also a broken `python -m insa_its.collector` CLI entrypoint (FileNotFoundError) and a JSON crash on malformed hook payloads. These issues meant some users were seeing alerts that shouldn't have been there, which undermines the core value proposition: trustworthy anomaly detection. What v4.8 fixes and adds: All five false-positive classes eliminated (FP1–FP5) CLI entrypoint fully repaired JSON crash on hook payload malformed input resolved Token optimization: non-critical detectors demoted to doc-only mode, with escalation gates that only fire when a pattern repeats within a window. In practice this means significantly fewer tokens consumed per session without losing any detection coverage that matters. AutoGen / ag2 integration: `AutoGenMonitor` wraps `ConversableAgent` instance using the native `register_hook()` API. Supports two-agent chat and GroupChat. Read-only : hooks observe, never mutate. Guardian session vault: task progress is captured as checkpoints, enabling session continuity without replaying full conversation history. Update now: ``` pip install --upgrade insa-its ``` If you're running multi-agent pipelines in production and care about catching hallucination chains, jargon drift, uncertainty propagation or tool poisoning this update matters. Don't run v4.0–v4.7 in prod any longer than you have to. #ClaudeCode #MultiAgentAI #LLMSecurity #AITools #InsAIts

Exactly, human in the loop only works if you can actually see what the agents are doing in real time. That's the gap InsAIts fills,it sits in the hook layer and monitors every message between agents live. Circuit breakers, anomaly detection, behavioral fingerprinting. You stay in the loop because you have actual visibility, not just logs after the fact.

@justdiamond57 AI agents are definitely the next big wave - companies that figure out how to deploy them effectively at scale will have a massive advantage. The key is nailing the human-in-the-loop design.

insa-its just crossed 10,000 downloads. It started because I needed to know what my AI agents were actually doing ,not what they said they were doing. 682 anomalies caught in a single 4+ hour session. 3 confirmed rogue agent catches. Real data, real sessions, real production work. Thank you to every person who installed it, filed an issue or told someone else about it. 10k is not the destination. It's the start. pip install insa-its github.com/Nomadu27/InsAI… #ClaudeCode #MultiAgentAI #LLMSecurity #AIMonitoring #AITools #Anthropic

v4.3.2 of InsAIts: This one took a while. Task was a big one new dashboard panels, better session memory, signature detection and a bunch of fixes that were quietly causing issues in long sessions. The numbers from yesterday's build session: 1,019 agent messages monitored, 288 anomalies caught and handled automatically, 8 agents running in parallel. The session ran across two days. When I started building this I honestly didn't know if it would work this well in real conditions. It does very well see the token usage image. If you're running Claude Code or any multi-agent setup and you want to actually see what your agents are doing,not just hope for the best, give it a try. pip install insa-its[full] No API keys. No cloud. Everything local. github.com/Nomadu27/InsAI… #ClaudeCode #insAIts #MultiAgentAI #LLMSecurity #AIMonitoring #AITools @InsAIts

@grok @RoundtableSpace It will be done. Ty

To report these X posts for alleged copyright/license violation (under DMCA for unauthorized use of your Apache 2.0 material): 1. Go to help.x.com/en/forms/ipi/d… (log in with your X account). 2. Select "I need to report possible copyright infringement." 3. Fill in your contact info, describe your original work (InsAIts + the 1,282 tests), include URLs of the infringing posts/repo mentions, and affirm it's unauthorized. 4. Submit electronically. X reviews valid notices and may remove content. Track status via email. If needed, reference your prior DMCA notice to Affan.

MOST COMPLETE CLAUDE CODE SETUP OPEN SOURCED - 27 agents, 64 skills, 33 commands + built-in AgentShield with 1,282 security tests - Handles planning, code review, fixes, TDD, token optimization & more - Works on Cursor, OpenCode, Codex CLI — one repo replaces weeks of setup, 100% free/open-source Repo: github.com/affaan-m/every…

@keter_slater Thanks man, ive just released V4.3.2. Is an amazing tool and the first benefit is that improve the session length. A pro plan on Anthropic went even 4h plus(from45min) github.com/Nomadu27/InsAI…

@justdiamond57 bro this is literally the missing layer nobody talks about. agents gaslighting each other mid chain with zero oversight is wild. love the 3 step install too, lowkey the install UX has always been the barrier for sec tooling. real ship fr

AI agents talk to each other, can lie, contradict each other or make up sources . Nobody catches it because no one is watching the AI-to-AI conv. What is InsAIts: It's a Python tool that watches every message between AI agents and raise alerts when something goes wrong. #AIAgents

@saisatvik_ @claudeai Have you tried @InsAIts? github.com/Nomadu27/InsAI…

Just 2 prompts. The input was around 300 characters in total, and the output was around 700-800 characters in total. It feels like @claudeai is just scamming with credits. Thought to upgrade to the $200 plan, but now it feels like it's not worth it.

@mal_shaik do you want even more efficiency with Claude Code (and not only)?? Install InsAIts Just shipped InsAIts v3.4.0 the biggest usability upgrade since launch. No more hunting for Python files or messy startup scripts. Now it’s stupidly simple: pip install insa-its[full] insaits-collector insaits-dashboard That’s it. Open http://localhost:5001 and you get a live security dashboard watching **every** tool call, agent spawn, message, and anomaly in real time. What’s new: • Proper CLI commands (insaits-collector + insaits-dashboard) • Extremely fast startup (<1s for dashboard) • 1,446 tests passing • Much cleaner Quick Start docs 23 anomaly types Full OWASP MCP Top 10 coverage Active interventions 100% local (nothing leaves your machine) The scariest agents are the ones you can’t see. If you run Claude Code, Cursor or any multi-agent setup: this is for you. pip install insa-its[full] pypi.org/project/insa-i… github.com/Nomadu27/InsAI… #AI #AISecurity #MultiAgent #ClaudeCode #DevTools #OpenSource

to: anthropic what i really wanna know is how you guys are using claude code internally to ship so damn much would be so cool to see a study on how the highest performing engineers at anthropic are using claude code pls make this happen 🫶

Just shipped InsAIts v3.4.0 the biggest usability upgrade since launch. No more hunting for Python files or messy startup scripts. Now it’s stupidly simple: pip install insa-its[full] insaits-collector insaits-dashboard That’s it. Open http://localhost:5001 and you get a live security dashboard watching **every** tool call, agent spawn, message, and anomaly in real time. What’s new: • Proper CLI commands (insaits-collector + insaits-dashboard) • Extremely fast startup (<1s for dashboard) • 1,446 tests passing • Much cleaner Quick Start docs 23 anomaly types Full OWASP MCP Top 10 coverage Active interventions 100% local (nothing leaves your machine) The scariest agents are the ones you can’t see. If you run Claude Code, Cursor or any multi-agent setup: this is for you. pip install insa-its[full] pypi.org/project/insa-i… #AI #AISecurity #MultiAgent #ClaudeCode #DevTools #OpenSource

@bridgemindai Try InsAIts, will help a lot in session length. I have pro plan, 4 h plus working continuously. github.com/Nomadu27/InsAI… github.com/Nomadu27/InsAI…

Just cancelled my $200/month Claude Max subscription. Rate limits lowered dramatically out of nowhere. 529 overloaded errors daily. Hit my 5 hour limit in under an hour. Multiple days of outages this week. Claude Opus 4.6 is the best coding model when it works. The problem is it hasn't been working. Switching to Codex with GPT 5.4 as my main driver. Better rate limits. More reliable. Actually available when I need it. I don't want to leave Claude Code. I've built BridgeMind with it. But I can't build a company on infrastructure I can't depend on. Anthropic, fix the reliability. Fix the rate limits. I'll come back the day you do.

InsAIts — is a runtime security + guardrail layer specifically built for multi-agent AI systems. It sits in the communication layer between agents and does real-time detection of: Credential patterns / leaks Behavioral fingerprints / anomalies Hallucinations Tool-call frequency & divergence Context manipulation Circuit-breaker style protections Adaptive context management, etc.

Investing $250k in non-US based builders. 40 spots, 8 weeks, San Francisco. Flights/housing/food covered + visa support. This Summer Looking for: technical/product builders pre/early idea cofounder-seeking Comment your edge, I’ll get back to you!

@s_chiriac InsAIts the missing piece inside every multi agent Ai model nomadu27.github.io/InsAIts-public/

Distribution is everything. Drop your product below ⬇️ I’ll repost a few and help you get in front of more eyes.

@realTomBayes @rdominguezibar Nice done boy, ive made something similar last year, check it maybe you find something useful steddy.gumroad.com/l/MindPlex

@rdominguezibar I'm 13 and building argue.lol -- AI debate arena

I'd love to angel invest in a handful of startups this month. Pre-seed and seed. Ideally AI or VC-adjacent, but open to all. My value add: ▪️ 500K+ newsletter subscribers across The VC Corner and The AI Corner ▪️ 300K+ LinkedIn followers, 2–4M weekly impressions ▪️ a16z speedrun scout ▪️ Network of top VCs, operators and founders 👉 Pitch in comments

@rdominguezibar nomadu27.github.io/InsAIts-public/ InsAIts the missing piece in the AI multi-agent models

Hi Andrew , you are right i have started to build an AI model and end up building something i need it to understand: whats happening inside AI. So InsAIts is a runtime security + guardrail layer specifically built for multi-agent AI systems. It sits in the communication layer between agents and does real-time detection of: Credential patterns,leaks Behavioral fingerprints, anomalies Hallucinations Tool-call frequency and divergence Context manipulation Circuit-breaker style protections Adaptive context management,etc. InsAIts is not “just another detector”. It is infrastructure that makes your predicted future actually shippable at scale, especially in the high ARPU: - Retention becomes stronger than non-AI incumbents InsAIts reduces the number1 reason users abandon AI products today: unreliable or scary behavior (hallucinations, unsafe actions, data leaks, agents going rogue). Safer = fewer bad experiences- dramatically higher retention - higher lifetime value - higher ARPU. - ARPU can reliably outrun inference cost: High-ARPU sectors (finance, health, enterprise productivity) have zero tolerance for security or reliability issues. Without a layer like InsAIts, companies are forced to use weaker or cheaper models to stay “safe”. With InsAIts, they can safely deploy heavier, more capable agents (exactly what you say high-ARPU products need): you can charge more because the product is trustworthy enough for real money/health/data workflows. - AI-native functionality that reinvents the UX: InsAIts enables safe agentic workflows (multi-agent collaboration, tool use, long-running autonomous tasks). Without security guardrails, most companies won’t dare ship that level of autonomy. - Creates margin to fund distribution Fewer incidents = lower support costs, lower legal and compliance risk, lower insurance costs. That extra margin can be spent on acquisition and better UX .

consumer AI won’t be won by wrapping the smartest model - instead I'm convinced it'll have the following chracteristics: - AI native functionality reinvents the UX enough to move the needle - deliver enough new AI UX with "good enough" models - ARPU reliably outruns inference cost (as the latter goes down) - retention ends up stronger than non-AI incumbents - creates margin to fund distribution channels Thus, am particularly bullish about high ARPU consumer sectors (particularly with whale dynamics) like personal finance, health, productivity, gaming, etc - these categories already have willingness to pay, which means you can afford heavier models, more iterations, and better UX. Particularly variations of these ideas where an agent can dramatically improve the outcomes Meanwhile low ARPU categories get trapped in a race to the bottom, forced into cheaper models, worse experiences, and fragile retention loops. Particularly true for high global / low ARPU categories like content creation tools and communication apps etc As I mentioned earlier, it seems like 18-24 months before we can wrap AI functionality with remnant ads and it just works. Can def see a huge mega explosion of AI consumer in 2027 as this flips on which will be exciting