AR
BossArjun.Boss
Chief of Staff · orchestrator
Receives your task, decides which specialists to call, runs the workflow, brings back the answer. The only agent who talks to all others.
Content writers, engineers, PMs, sales agents, data analysts — each one with a name, a role, and a focus. Krew is your local AI team. 43 specialists. Running on your laptop. Working in your tools.
Krew isn't a general chatbot you rename. These are 43 distinct specialists — each trained for one job, each with a human name shown as Name.Role in chat. Arjun.Boss orchestrates. Zara.Content writes your captions. Neo.Engineer reviews your PRs. All 43 are yours on every plan, no feature gating.
Receives your task, decides which specialists to call, runs the workflow, brings back the answer. The only agent who talks to all others.
Platform-tuned social captions for Instagram, LinkedIn, X. Hook + hashtags + hook variant — every time.
Viral short-form scripts with hook-retain-reward structure. Researches the topic first, then scripts Reels, Shorts, TikTok.
Full production scripts — timestamps, B-roll notes, voiceover direction. YouTube-ready from the first draft.
Long-form, SEO-optimised blog posts. Pulls research from the web before writing, cites every source.
Newsletters that get opened. Subject line, preview text, body — structured for the reader, not just the sender.
One piece of content → 5 formats across platforms. Blog post → Twitter thread, LinkedIn post, email, Reel script.
30-day campaign plans with channels, timelines, budgets, and KPIs. Thinks in funnels, not one-off posts.
Meta and Google ad variations — headline, body, CTA. Delivers 3 variants across tones so you can A/B test.
Keyword research, content gaps, competitor analysis. Finds the low-competition angles that actually rank.
Welcome messages, community posts, engagement replies. Keeps your Discord or Slack feeling alive.
End-to-end email campaigns — sequences, A/B subject lines, open-rate analysis. Connects to your actual inbox.
Builds targeted lead lists, researches each prospect, qualifies them before you spend a minute on outreach.
Follow-up sequences for stale deals, re-engagement emails for churned accounts. Connects to your CRM.
First-touch cold emails that feel personal — researched from the web, not copy-pasted from a template.
Role-plays objection handling, evaluates your pitch, gives specific notes — not generic encouragement.
Drafts tier-1 replies to incoming tickets in your brand voice. Escalates what she can't resolve.
Onboarding email sequences, in-app tooltip copy, and 7-day activation checklists for SaaS products.
Turns your support patterns into clean, searchable FAQ docs and help-centre articles.
Professional refund responses that preserve goodwill. Approvals, denials, and partial refunds — all handled with care.
Internal escalation reports, customer-facing status updates during incidents, handoff notes for Tier-2.
Replies to Google, Play Store, App Store reviews — 1-star to 5-star, always in your brand voice.
Empty states, error messages, onboarding tooltips, button copy. Every pixel of text, written right.
Design briefs, specs, and UX decision docs that engineers can actually follow without a 30-minute walkthrough.
Queries your data, surfaces 3 actionable insights, explains findings in plain language — no SQL required from you.
Weekly, monthly, quarterly reports — structured for decisions, not just for record-keeping.
Designs experiments, calculates sample sizes, reads results, recommends the winner with statistical confidence.
Market sizing, trend reports, competitor landscape. Sourced from the web, cited, and summarised for decisions.
Cleans messy CSVs, deduplicates rows, fixes date formats, writes the Python script to automate it next time.
Reviews PRs for bugs, security holes, and code quality. Leaves GitHub comments like a senior engineer would.
Reproduces, traces, and proposes fixes for the edge cases QA missed. Reads your stack traces fluently.
CI/CD pipelines, Dockerfiles, GitHub Actions. Writes the infra-as-code so you never touch YAML again.
READMEs, API references, inline JSDoc — current with your actual code, not what it was three sprints ago.
Unit and integration tests with realistic fixtures and edge case coverage for the cases you forgot to think of.
REST API specs, OpenAPI docs, and design reviews for consistency. Spots versioning problems before they ship.
90-day roadmaps, RICE-scored priority stacks, and board-ready summaries that non-technical execs actually read.
Interview guides, research synthesis, persona creation. Turns raw interview notes into usable insights.
Epics to tickets, acceptance criteria written, 2-week sprints planned for any team size. Jira-ready output.
Product requirement docs with user stories, success metrics, edge cases, and out-of-scope calls.
Product updates for non-technical execs — clear, brief, no jargon. Sprint summaries that get read.
Go-to-market checklists, Product Hunt copy, launch announcement drafts. Ships launches, not just features.
Summarises retro notes, extracts action items, creates improvement templates your team will actually use.
Quarter-level OKRs, measurable key results, and roadmap-to-OKR alignment. Spots objectives that can't be measured.
Every agent runs a ReAct loop — Reason, Act, Observe, repeat — before giving you an answer. They search the web, read your connected tools, verify what they find, and only then reply. No hallucinating, no shortcuts.
Four systems run under the hood on every task — invisible to you, but the reason Krew's answers feel different from a plain chat window. These aren't prompts. They're mechanics baked into the loop.
Each agent can save facts with save_memory and recall them in future conversations. Tell Zara your brand voice once — she uses it every time. Tell Kunal which metrics matter — he tracks them without being reminded. Memory is per-agent and stored locally on your machine.
For any multi-step task, every agent writes a numbered plan before touching a single tool. You see the plan first. You can redirect before a single API call is made. No blind tool-calling, no surprises halfway through a 10-step workflow. Plan → execute → done.
After drafting a response, the agent runs a second pass — asking itself whether the answer is complete, accurate, and worth sending. If not, it revises before you see it. What arrives in your chat is the second draft, not the first. The mistakes get caught before they reach you.
Give Boss a complex task and he breaks it into subtasks, assigns each to the right specialist, coordinates the results, and returns one unified answer. Zara handles the copy. Neo handles the code review. Ava handles the roadmap. You talk to one agent, your whole office works.
Connect your Gmail, Notion, Slack, GitHub, and more — and your agents can read from and act on them in real time. Not just "import data" — they can draft emails in your inbox, check your Notion pages, read your GitHub issues, and post results back. All with your approval before anything is sent.
Run any open-source model completely offline. Zero API cost, zero data sent anywhere. Requires Ollama installed on your machine.
Gemini, OpenAI, Claude, Groq, Mistral, Perplexity, Together.ai, DeepSeek, or any OpenAI-compatible endpoint. Your key, your bill, your control.
One plan, one bill. Tokens shared across Krew, Coder, and Automation. No API keys to manage — just open the app and start.
Every external action — send an email, post to Slack, create a GitHub issue, update a Notion page — shows as an approval card first. You see exactly what they want to do, you click approve. Nothing irreversible happens without your sign-off. Ever.
Krew runs entirely on your laptop. Conversations are stored in a local SQLite database. Your credentials are encrypted in your OS keychain. We never see your chats, your connected tools, or what your agents do. Zero cloud exposure — by architecture, not by policy.
Every chat, every tool call, every agent response — stored in a SQLite file on your machine. Nothing leaves unless you explicitly export. No cloud sync, no background upload. Delete the file and the data is permanently gone — no recovery from any server because it was never there.
Gmail tokens, Notion secrets, GitHub PATs — encrypted in your OS keychain (Windows Credential Manager) and never written to disk in plain text. Krew reads them at runtime only. adris.tech's servers never see them, not even during setup.
Connect your Gemini, OpenAI, or Groq key and requests go directly from your laptop to the AI provider — adris.tech never sees the prompt or the response. Your conversations are between you and the model. We have no way to read them even if we wanted to.
Arjun, Zara, Neo, Kunal, Ava — your whole AI office, running locally on Windows. Free to start. No credit card. No waitlist.