PH热榜 | 2026-08-11

Hey Product Hunt! Stephen here, Head of Product at Tines, and one of the makers of Tines 3B.

Last year, AI coding tools got good enough that anyone could build agents, apps, and automation fast. Just describe what you want, and pretty soon you have something that works. But what about what happens next? AI-generated software needs to connect to the real world, and it often does that by hardcoding API keys in plain-text files or databases. Handing credentials to models that can hallucinate or fall for prompt injection is a major risk. Run that code in a typical, unisolated container, and one bad step can contaminate the whole system.

We call this phenomenon wild code: it’s work that’s built fast with AI but runs ungoverned and unmonitored, while creating hidden security, financial, and operational risk.

With 8 years of intelligent workflow experience in supporting customers like Reddit, Coinbase, and Databricks, we knew how to solve this problem. And so we set out to build Tines 3B. 

What is Tines 3B?

A single, secure environment for everyone’s most important agents, apps, and automation. Tines 3B provides: 

• Freedom to build with AI anywhere

• Control to run and monitor that work

• An AI-native, code-first platform built to work with LLMs and AI coding tools

We built Tines 3B with safety as a north star. The AI never sees your credentials. They are injected at runtime by a proxy that sits entirely outside the code.

Why launch on Product Hunt now? 

We recently introduced Explore Edition, a no-commitment way to try Tines 3B yourself. This community of innovators is exactly who we want exploring it first. You’ll have unlimited users, spaces, and connectors, and can push 3 workflows live. This includes tunneling into your own home network or self-hosted services, no port forwarding required. Build, run and monitor your apps, agents, and automation with credential protection, isolated execution, and full auditability.

Who is Tines 3B for?

• Builders who want to go from ideas to workflows using natural language

• IT and Security teams who need visibility into what’s running 

• Anyone dealing with sprawl from AI-generated scripts, agents, and one-off tools

• Innovators who want AI to write the code for their projects, home labs, or personal agents

We’d love your feedback

We think you’re going to enjoy building in Tines 3B, and feel the control it gives you. Powering the world’s most important workflows is what we’re here to do, and we want to know what you think.

We welcome feedback from this community on how you and your teams are managing AI-driven work, and where Tines 3B fits in your day-to-day. 

I’ll be in the comments all day, ask away!

The credentials never reaching the AI/code itself is probably the part I'd want to test first.

This feels especially relevant for the little internal tools people build with Claude/Cursor and then somehow end up on six months later.

The credential isolation makes sense, but I’m still not clear on how the actual workflow is controlled. If AI-generated code can request credentials through the proxy, what prevents it from using those credentials to do something unintended? Is there granular control over what each workflow can access or do?

Tines 3B is such an exciting launch for me! I've gone from needing a fair amount of support when building, to being able to build fully functioning workflows myself. The blocker now is knowing what I want to build next 😅

This kind of freedom to experiment and find new ways of working has been really fun, and challenging, but one I'm delighted to have. Not being the most technical person, I'm also super happy to still have guardrails up so I'm not just pushing code out into the universe! I'm not blocked from building, but pushing a workflow live is still something that the IT and Security team have visibility over.

@Tines 3B has changed the way that our RevOps team works. Every use case is now solvable in a smart, secure, intuitive way.

@steobrien This looks really impressive! I especially like the idea of keeping agents, apps, and automations in one secure and auditable environment.
Good luck guys!

How are teams handling ownership when lots of small AI built automations start spreading across different departments?

Congrats team!

I like that you are not trying to stop people from building with AI. the safer approach seems to be letting them move fast, then putting guardrails around what actually runs.

As a marketer I've truly never experienced something like Tines 3B - connecting 3 totally different, unrelated systems to get an accurate picture of campaign performance, spend, and pipeline has been a GAMECHANGER.

@rosie_halpin is doing a live demo of Tines 3B on LinkedIn and Youtube right now for anyone interested!

Curious as to why you named it 3B. I thought Tines for was releasing a small language model at first lol.

I didn't quite understand the part about seeing the revenue generated by each workflow. Is that for entrepreneurs who create lots of small products that they bill independently?

@Tines The credential proxy answers above cover the read path well, so I'll ask about autofix instead. It was mentioned it applies a fix on a branch and then reruns the workflow to confirm the fix actually works. If the failing step was a write, posting to an API or sending mail, does that rerun hit the real connector with real credentials, or is there a dry run path so a fix attempt can't duplicate a side effect?

Hey PH, I'm Shabz, one of the people behind BetterClaw.

BetterClaw is a no-code platform for building AI agents that actually run on their own. Connect your tools, describe what you want, and it's live in 60 seconds.

Why we built it

I spent months in the OpenClaw subreddit helping people set up their agents. Hundreds of them.

And the pattern was always the same.

Someone discovers AI agents. Gets genuinely excited. Their eyes light up when their agent sends its first message.

Then they lose an entire weekend to Docker and config files.

And they're gone before week two.

The AI part was incredible. The infrastructure part was quietly killing it.

So we kept asking: why should anyone need to become a sysadmin just to use an AI agent?

They shouldn't. So we removed that part entirely.

What makes it different

• 60-second deploy - no Docker, no YAML, no VPS, no terminal. We timed it.

• 95+ one-click OAuth integrations - Gmail, Calendar, Slack, HubSpot, GitHub, Jira, Meta Ads, Linear and more. Real OAuth, not "paste a webhook and pray."

• Trust levels - every agent starts as an Intern that asks permission for everything. Promote it to Specialist, then Lead, as it earns your trust. Handing an AI full system access on day one is wild, and somehow that's the default everywhere else.

• Secrets auto-purge - AES-256 encrypted, and gone from agent memory after 5 minutes.

• BYOK, zero markup - bring your own LLM key. Pair it with a free model (Gemini, OpenRouter, Groq) and your total cost is genuinely $0.
  

Not "$0 but you need a server." Zero.

Who it's for

Founders, ops and support teams, recruiters, agencies, and tbh, anyone non-tech who's tired of doing the same thing 50 times a week.

What people are building

Meta + Google Ads spend vs Stripe revenue → daily ROAS in Slack
Gmail lead → scored, enriched into HubSpot, call booked
Zendesk tickets → answered from your product data, only edge cases escalated
Search Console → finds page-one pages losing clicks (grew our own site 1.6K → 7.9K clicks in a month)
Stripe declines → follow-up drafted, recovery tracked
PostHog drop-off → cross-checked in Salesforce → owner pinged pre-renewal

Huge thanks to @rohanrecommends for hunting us, and to the PH team for building a place where small teams get a real shot

And to Tina, Varsha, Aniket and everyone who poured months into this - this one's yours too. ❤️

🎁 For PH folks: use code PH3 for 3 months of Pro at $49.

We'd genuinely love your feedback - especially on what you'd want an agent to handle first. I'll be in the comments all day. Ask me anything, including the stuff we haven't figured out yet. 👇

→ betterclaw.io

Finally, a no-code AI agent platform that respects both my time and my wallet. The scheduled workflows (Gmail triage + Slack briefings) are exactly what I needed, and the fact that I can bring my own API key instead of paying per execution is refreshing. The UI is clean, setup was under a minute, and the "ask before acting" safety net is brilliant. Instant upvote — well done team!

Wishing good luck with the launch guys! :)

Laiba here, also on the BetterClaw team.

One thing worth calling out: every agent starts as an Intern that asks before it touches anything. You promote it to Specialist, then Lead, once it stops surprising you. That's the difference between a cool demo and something you leave running while you sleep.

We built it that way because we didn't trust our own agents at first either.

Fastest first agent is a morning briefing on Gmail or Slack. Takes about a minute.

What would you want an agent to take off your plate? I'm in the comments all day.

I like the idea of scheduled agent handling small repetitive task without needing another dashboard open all day.

The Intern model is a smart touch. Having the agent ask before acting makes automation feel a lot safer to try.

How do you decide when an agent has done enough to move from Intern to Specialist without giving it too much access too early?

Congrats @better_shaya & team!

This is intersting. Does it also clean unwanted notifications after summarizing..... like I love to see Linkedin notifications. But deleting all of them manually is a task. Can you do that?

BetterClaw is taking a practical approach to agentic AI — instead of just generating responses, the agents can actually run workflows on a schedule and handle repetitive tasks in the background.Love the “60 seconds & $0 forever” positioning.

BetterClaw looks like a really exciting way to make AI agents actually useful in the background-especially the scheduled workflows and no-code setup. Love the “60 seconds & $0 forever” positioning.

The "Interns that ask before acting" model is the right instinct, most agent tools skip that approval step entirely. Question on the other side of it: once an Intern acts on something external, an email sent, a Slack message posted, how do you verify the action actually landed rather than trusting the connector's own success response? I've spent today running browser automation and roughly one in three actions reported success while nothing actually happened, a click that never navigated, a submit that never posted. Curious whether BetterClaw checks real downstream state after an agent acts, or trusts the tool call's response.

As an agent gets promoted from Intern to Specialist/Lead, can you set granular custom rules for what actions trigger a permission prompt, or is it based on predefined roles? Super excited to see where you take this!

Congrats on the launch! The trust-level idea feels especially practical: starting agents as Interns before giving them more autonomy is a smart way to make non-technical users comfortable with scheduled AI workflows.

Not needing Docker or a VPS removes a pretty big barrier for people who just want an agent running. Congrats!

Someone on our team shared this and I set up an agent that pulls my calendar and Slack messages in the morning and sends me a briefing at 8am. That's it. That's my whole use case. Took about 2 mins

The Intern thing is smart. I would not have connected my work Gmail to something that could just do whatever it wanted. Knowing it has to ask me first made me actually try it instead of closing the tab.

Been burned by "no-code agent" tools twice before so I went in expecting to hit a paywall or a "contact sales" screen within 5 minutes. Neither happened.

Set up a Gmail triage agent on a free Gemini key. Took maybe 90 seconds, not 60, but close enough. It's been running for 11 days now and the only time I touched it was to promote it from Intern after day 4 because I got tired of approving the same correct classification over and over.

The BYOK part is what keeps me here. I switched from Gemini to Groq for one agent and to Sonnet for another, took about 10 seconds each time. No price increase on BetterClaw's end because they're not in the billing path.

Solid tool. Genuinely free. Would've saved me about three weekends I lost to Docker last year.

I like the idea of agents as Interns. That approval step could make this mucg easier to trust especially when connecting Gmail or Slack for the first time.

The Intern that asks before acting is the part I would build the whole product around, and it is also the part that quietly decays. We run approval-gated sending on our side, and the failure mode is never the agent doing something reckless. It is the human. Reviewing every action works until the volume goes up, then you start skimming, and the ones that get through are never the obvious ones.

So the thing I would watch is what keeps an approval meaningful on day ninety. An agent that asks rarely, and only where it is genuinely unsure, is worth more than one that asks every time and trains you to click yes.

Is it really going to be 0$ forever?

How does the agent 'earn' promotion, is that manual on your end or does it track some kind of track record automatically?

@shabb_katoch Congrats on the launch, Shabnam! Love that you doubled down on security from day one and also mentioning it right on the top of the fold in the website.

congrats on the launch you guys!

Congrats on the launch, Shabz! 🚀 60-second deploys with BYOK and zero markup is how no-code AI tools should actually work.

Really love the 5-minute secret auto-purge feature for enterprise/security peace of mind. Wish you guys a huge launch day!

I ran into what seems like a loop bug when retrying the agents. I initially tried it with a Groq API key, and even after switching to a Gemini API key, I kept getting the same error:

“The agent hit an unexpected error while responding.”

I’m not sure if the issue is with my API keys or something else in the retry flow, but sharing it here in case it helps you guys identify what’s going on

Nice product to have! Keep building!

This is Spotify's 100th launch on Product Hunt!

Suitable — considering most people think of Spotify as a music service, but really, from @Backstage to Xirp, they've become a formidable AI company in their own right.

Here's Spotify story about how it built Xirp, a vendor-neutral agentic development environment, to help engineers manage dozens of parallel AI coding sessions without losing context or getting locked into a single tool.

Paired with Portal, Xirp turns Spotify’s organizational knowledge into a shared baseline that improves agent effectiveness, reduces duplicate work, and makes it easier to switch models and preserve context across sessions.

Key Features

• Xirp lets Spotify engineers run 50+ parallel agent sessions across different harnesses while keeping worktrees and context separate but portable.

• The system is vendor-neutral, allowing Spotify to switch models mid-task and optimize for price-performance, including self-hosted open source models.

• When connected to Portal, Xirp adds organizational context like architecture, dependencies, ownership, and decisions, while session data flows back for visibility and reuse across teams.

From what I can tell, the nearest analog to Xirp is Craft Agents, which was acquihired by @Polymarket.

I didn't see this coming from Spotify

Hey Product Hunt! 👋

I’m Oren, co-founder and CEO of @EquityBee, alongside @oded_golan

Today, we’re excited to launch Equitybee Benchmark, a free tool that lets U.S. startup employees compare  equity grants by department, seniority, and company stage.

🎯 Why we built it

Companies have long used compensation benchmarks when structuring equity offers. Employees evaluating those offers rarely have access to the same market context.

We believe employees should have a benchmark too.

That context can help when evaluating a job offer, comparing opportunities, or preparing for a compensation conversation.

📊 How it works

Equitybee Benchmark is built on more than 9,000 verified new-hire option grants covering more than 2,500 U.S. startups.

Select your:

• Department
• Seniority level
• Company stage

So, if you’re a senior engineer evaluating an offer from a Series B startup, you can create a relevant comparison group and see the 25th percentile, median, 75th percentile, and average fair market value of grants within that group, alongside insights that help put those figures into context.

The goal is not to predict what those options may eventually be worth. It is to give employees a relevant comparison point for understanding the equity they are being offered.

Equitybee Benchmark is free and built specifically for startup employees, not their HR teams.

💬 We’d love your feedback

Explore the Benchmark and tell us: what additional information or comparisons would help you better evaluate startup equity?

Amazing teams build amazing products!
This is super valuable and helpful to startup employees.

Super excited to launch Equitybee Benchmark today! 🎉

Leveling the playing field for startup employees has always been at the heart of what we do at Equitybee. Having real, verified data to help people navigate offer letters or equity negotiations is a huge milestone.

Huge congrats to the team! Take it for a spin and let us know your thoughts!

I work at Equitybee, so I am biased, but this is the tool I wish existed years ago. I spend my days inside cap tables, and the one question employees can never answer is "how does my grant actually compare?" Companies have had that market context forever. Employees have not. Benchmark closes that gap: pick your department, seniority, and company stage and you see the real distribution, not a vague market median. Built from more than 9,000 verified grants.

Guy here, I lead marketing at @Equitybee and was one of the people working on this new product.

At Equitybee, our mission is to empower startup employees to participate in the success they helped build.

To deliver on that mission, we believe employees should have access to the data, knowledge, and tools they need to better understand the equity portion of their compensation.

I’m really excited to see the value this can bring to the startup builder community, and even more excited to learn from your feedback.

I’m here to answer any questions, thoughts, or concerns!

I’m Lior, VP R&D at @Equitybee

One of the biggest challenges with startup equity is that employees often have very little context for understanding how their grant compares.

That’s exactly why we built Equitybee Benchmark.

We’ve taken a complex set of inputs, data points, and comparison dimensions and turned them into an experience that feels simple, clear, and useful.

Now every startup employee can explore how their equity compares by role, seniority, and company stage.

Take it for a spin and let us know what you think. We’d love to hear your feedback and what you’d like to see next

Super excited to see Equitybee Benchmark live today! 🚀

It’s been really rewarding to work on this product and turn a large, complex dataset into something simple and useful for startup employees. Equity compensation can be hard to put into context, and our goal was to make it much easier to understand how your grant compares to the market.

Really proud of what the team built, and excited to hear your feedback. Give it a try and let us know what you think! 🙌

So great to see it in the air! Good luck to the Equitybee team!

Hey Product Hunt 👋 I’m Oded, co-founder and CPO at @Equitybee.

One of the things we’ve seen again and again is how little context startup employees have when it comes to their equity. You can know the number of options you were granted, but still have no idea whether that grant is actually competitive.

That’s a big part of why we built Equitybee Benchmark: to give employees access to real comparison data by department, seniority, and company stage. It’s built on 9,000+ verified new-hire grants across 2,500+ U.S. startups.

Really proud of what the team has built, and I’d love to hear what you think - especially what other data or comparisons would make this more useful for you.

Great job helping to bring transparency to the equity compensation space! Nice work!

Tried a few combinations of department, seniority, and stage. Looks good!

Are you planing to add more dimensions?  industry? company valuation?

Great work!

@orenbarzilai @guy_elbaz1 , This fills such a big gap to help hiring with startups a fair play where both startups and talent can negotiate a full cash + equity in a transparent and data driven way. Congrats on the launch!

So proud of our team for this launch! Benchmark makes it so easy to see how your equity grant stacks up against others in your role, seniority, and stage; no more guessing if your offer is fair. Super clear, intuitive, and genuinely useful for startup employees.

When it comes to stock option grants in the startup ecosystem, the information gap is real. Startup employees (myself included) have struggled with this issue for far too long. Equitybee Benchmark was launched to promote greater transparency across the ecosystem and empower startup employees to make better-informed decisions about what can easily become their most valuable asset: their equity.

Proud to be part of this company!

Really interesting tool! Equity grants can be pretty difficult to compare with one another, so having more context around how a grant stacks up across companies/industries seems like a great way to assess one's position.

This is the kind of transparency the equity comp space has needed for a while. Well done!

Congratulations on the launch :)

Are people using it more when weighing an offer, or later when deciding whether to exercise?

Everyone here is on the employee's side of the table. From the founder's: the option pool is the line item we see modelled worst — headcount goes in as cash, equity as "free".

Any plan for the reverse lookup? Feed it an 18-month hiring plan, get the pool it takes.

Hey everyone, my name is Alex Sima and I am the CTO of Bullet.

Like the post said, we built Bullet because Claude Code and Codex were too slow. Our YC partner said we should solve a problem we face ourselves, so we decided to go big or go home (guess we decided to go small and lightweight but ok).

We now only use Bullet internally...and personally, I can't go back to using Claude Code. If you hate slow coding agents, join the club: https://www.codewithbullet.com.

You can bring your subscriptions over, so it's free speed -- enjoy :)

Congrats! This is so cool :) Quick question — besides the Minecraft clone example in the video, what are some other real-world coding tasks where you think Bullet’s speed improvements would make the biggest difference?

Hi everyone, my name is @adi_kulkarni and I’m the CEO of Bullet.

Existing coding agents were too slow for our development workflows, especially when iterating on a feature synchronously. Even when they finished, there would be a mistake so countless hours were wasted in this loop. After a certain amount of context accumulates in a session, both the agent and I are confused about the actual state of the feature!

Thus, we built Bullet to make coding agents faster, more focused, and easier to iterate with. It automatically chooses the right model and reasoning level, runs independent work in parallel, and uses targeted code search instead of dragging unnecessary context through every step.

We built Bullet for ourselves first, and now we’re excited to share it with everyone. Would love to hear what you all think!

Bullet so fast I can't even doomscroll inbetween agent sessions anymore

Congrats on the launch@alex_sima26@adi_kulkarni! Does Bullet also help stretch usage limits, or is it just speed? Any benchmarks or stats on token savings?

Congrats on the launch! Waiting on coding agents is easily one of the most annoying parts right now, so cutting down iteration time is a big win.

congrats on the launch guys! looks very impressive, excited to see where this goes!!!

Congrats Alex and team, love this!! Looks fire 🔥

Great product, testing it rn.

Stumbled across this post as my claude code was taking 30 years. How timely

Best coding agent!! Lets go team!

Congrats Alex and team, this looks great!

I would love to download and start using this today, but can't justify burning tokens when claude code is $200 a month and I spend $3-4k worth of tokens :(

I'm so excited for this! Recently I've been noticing that the bottleneck with coding agents isn't quality anymore, it's just speed - so excited to have bullet accelerating my dev velocity!

I love love love BB. It just feels GOOD to use and the flexibility to build and manage agents is extremely freeing.

I'm genuinely enjoying using bb. Super slick experience and was a breeze to set up with my custom checkouts/worktrees setup. I'm looking for some more standardized paths for notifications for web ui, at the moment.

Congrats on tha launch, Sawyer! Love that this is oss and respects existing agent subscriptions

In spirit, bb feels like Emacs for coding agents: an environment intended to be shaped around how you work. bb is less like an AI chat bolted onto an editor and more like the beginnings of a programmable working environment for coding agents. I've been testing with version 0.36.0 and it has been an eye-opener. bb is evolving rapidly. It's not a settled, maintenance-free daily driver but it is extremely flexible and be warned you'll be dragged into exploring an insane multitude of possibilities!

Can bb be integrated into Lovable?

Hi, do you have plans supporting other platforms, than Mac? I'd give Linux build a try.

Congrats on the launch :) I really like that everything is local and private and the fading-strength idea feels just right to me.

Upvoted, really cool concept @rolandleth !

Just tell Vizard Agent what video you want to create, then sit back and let it surprise you. Congrats on the launch! 🚀

Been working with Vizard Agent for a while, and the part that feels different isn’t any single AI feature. It’s being able to say “make me this video” instead of manually deciding which tool to use for every step.

It’s still early, but watching it go from a brief -> planning -> creating/editing -> a finished video has made the “agent for video” idea feel much more concrete to me.

Very excited to finally see this out in the world!

Hey Product Hunt! 👋 I’m Alex, founder and CEO of Kubit.

When an AI feature fails, existing observability tools tell you what the agent did, and traditional analytics tell you if the user left. Neither tells you why because they don't talk to each other. You're left toggling between tabs, manually matching AI execution logs to front-end user sessions just to figure out what broke the experience.

We experienced this frustration first hand when building our own AI features. So we built Kubit to bridge this exact gap: a unified product analytics platform designed for both agents and users.

With Kubit, you can:

• Connect Agent Traces to User Behavior: Link backend agent traces directly to user actions to see why users re-prompt, abandon a flow, or convert.

• Tie Agent Performance to User Outcomes: Correlate P95 latency, token usage, and model costs directly to core metrics like DAU, retention, and LTV.

• Track User-Agent Funnels: Pinpoint the exact step where an agent’s hallucination or failed tool call disrupts a conversion funnel.

• Map AI User Journeys: Track re-prompts, rage clicks, user intent, and sentiment to uncover hidden UX dead-ends that standard APMs miss.

• Build Granular Cross-Domain Cohorts: Segment users using complex conditions that combine both backend agent interactions and front-end user behavior.

• Headless for Coding Agents: Leverage MCP and custom Skills for headless analytics designed for developer-facing AI tools.

Easy Integration & Flexible Data Architecture

• Quick Setup: Connect Kubit directly to your existing OpenTelemetry (OTel) infrastructure or CDP in minutes.

• Zero-Copy / BYOW: Prefer to keep data in your own stack? Our Bring Your Own Warehouse (BYOW) architecture ensures top-tier security, compliance, and control.

We built Kubit to help product and AI engineers optimize agent performance alongside user behavior to create AI products that actually stick. Try it out for free today at kubit.ai.

We’d love your feedback! How are you currently tackling agent observability and user analytics? Drop your thoughts, questions, or feature requests in the comments below, join the conversion, or reach out directly at alex@kubit.ai.

Happy building! 🚀

I talk to teams building AI features every week, and it's always the same story. They can tell you what the agent did. They can tell you if the user bailed. What they can't tell you is why, because those two things live in separate tools and nobody's connected them.

That's the whole reason Kubit exists. Agent traces and user behavior, same place, so you can actually trace a bad tool call to the drop off it caused instead of guessing.

Would mean a lot if you checked it out and gave us an upvote. And if you're wrestling with this problem yourself, tell me how, I'm curious.

Hi Product Hunt!

Large companies are extremely complex systems. As soon as a company scales past fifty or so employees, it becomes totally impossible for an executive to know every single thing that happens at a company. This means that, since everyone at a company only understands a small slice of what's going on, it's essentially inevitable that inefficiencies arise on a global level that aren't visible at smaller scales. Before Cerenovus, this was a very difficult problem to tackle, because there was just too much information for any one person to have the whole picture. Cerenovus is built to fix this.

Here's how Cerenovus works:

First, Cerenovus ingests the company's information. This looks differently for every company based on their priorities and existing systems, but typically this means hooking into documents, communications and databases that the company wants to analyze, and bringing all the information together in one place.

Next, Cerenovus runs an AI-powered synthesis pipeline to organize the information and map out all existing systems and workflows. Once the system has mapped out how things work at a company, it scans for a wide variety of common failure modes, and does open-ended analysis to find places where processes can be streamlined. Finally, its findings are compiled into polished, cited briefs, reviewed and validated by humans, and then shipped back to our clients.

We're so excited to show everyone what Cerenovus can do - book a demo today: https://cerenovus.ai

the adversarial reviewer checking sources is a good answer to the accuracy question, but the part that'd worry me more as a customer is access, not accuracy. to find inefficiencies across a whole company you need to ingest documents, comms and databases pretty broadly, and "where the money is being lost" reports often trace back to a specific team or person underperforming. that's a report with real career consequences for someone, generated from access to their emails and docs that they didn't necessarily know would be read this way. who inside the client company actually sees the finished brief, is it scoped so a manager only gets findings about their own team, or does it land somewhere with company-wide visibility by default?

Really like the session persistence. Congrats!

Hey everyone, Vladimir here, one of the makers of Octomind.

We built this out of a simple frustration: AI agents that die the moment you close your laptop. Now you just hand off a task, shut the lid, and pick the session up later from any device.

Really proud of what the team shipped today. If you give it a try, honest feedback — good or bad — is the best thing you can leave here. Happy to answer any questions. Thanks for checking us out!

Hi everyone!

Awesome to co-hunt Gitar with André @conduit_design!

Now part of @Sonarsource, Gitar reviews PRs, diagnoses CI failures, applies fixes, and validates those fixes against the same pipeline.

The review doesn’t end with a finding and another comment for devs to deal with. Gitar stays with the change through the fix and back through CI.

André had a good way of putting it:

Most people think this space is saturated. It is, but in the low end. High-end code reviews is blue ocean. Maybe Gitar could be the first one to enter.

P.S. If you maintain an open-source project, you can apply for free access to Gitar.

agent-manager is the fastest way I have found to work with coding agents, and everything in it is one keypress.

space on a group starts an agent in the right directory from a single sentence. space on a session answers one that is blocked, without attaching. ctrl+r opens what it changed as whole files, where a comment on a line goes back to that agent as a review prompt. f forks a conversation into a named sibling, T pins a plain shell next to the agents, and yesterday's release added the arrow keys for stepping in and out of a pane.

It runs on tmux rather than replacing it. Sessions are ordinary tmux sessions on their own socket, so quitting the manager leaves every agent running and your own tmux config untouched. Claude Code, Codex, OpenCode, Gemini CLI, Grok and Pi come with status rules, and any other CLI runs as a session right away.

Honest about the scope: cost tracking is not there, and the wheel scrolls but everything else is keys. Go, Apache-2.0, one binary, no daemon.

Happy to hear what breaks for you.

Looks cool! I’m gonna give it a try! For some reason, though, the website link doesn’t seem to work.

One thing I did not expect: review mode became the reason I keep it open, more than the list did.

It started as a way to see which agent was blocked. Then I kept attaching just to read diffs, so ctrl+r grew into whole-file diffs where a comment on a line goes back to the agent as one review prompt. Yesterday's release added stepping in and out of a pane with the arrow keys, which sounds trivial and is now the thing I press most.

If you run more than one agent: where does your time actually go, watching them or reviewing what they wrote?

This feels much more helpful than regular AI chat apps on mobile. Excited to see how it improves!

open source and local is the right call for a dictation tool specifically, since the whole point is you're speaking things you'd never type into a random cloud service. curious how the accuracy holds up on technical vocabulary though, variable names, library names, that kind of thing tends to be where local models fall over compared to the big hosted ones. is there a custom vocabulary or fine tuning option for that or is it running as-is out of the box.

Love that this runs fully local and open source, so my voice never leaves my machine while I dictate. That privacy first approach is a rare and refreshing take for a dictation tool.

Finally, a tool for this. Is the switch shortcut configurable or fixed? Been burned before by tools that hardcode something that clashes with whatever app I have open.

Very cool, the companion app remote really elevates this to a uniquely practical tool