# Phanindra Reddy and MagicTeams.ai > MagicTeams.ai builds AI agents that do real work: voice agents that answer the phone, operations agents that work the inbox and ask before they act, and the open-source controls that make every agent visible, owned, and safe to switch off. Founded and led by Phanindra Reddy. Working with US teams. Canonical page: https://phanindra.magicteams.ai/ Privacy: https://phanindra.magicteams.ai/privacy MCP server (read-only): https://phanindra.magicteams.ai/mcp ## Quick facts - Company: MagicTeams.ai (legal entity: Saphaare Labs Private Limited) - Founder: Phanindra Reddy, Founder and AI Engineer (MagicTeams.ai since November 2024) - Experience: 5 years building production software, from enterprise ML pipelines at Accenture to founder-led voice AI and agent platforms - What we do: production voice agents, operations agents, and agent governance (inventory, identity, tracing, cost, evals, kill switch) - Where: founder based in Hyderabad, India; customers in the United States - Email: phanindra.reddy@magicteams.ai - Book a call: https://cal.com/champ1288/aios - LinkedIn: https://www.linkedin.com/in/phanindra-ai - GitHub: https://github.com/everyai-com - Company site: https://magicteams.ai ## Startup programs Member of NVIDIA Inception, AWS for Startups, Claude for Startups (Anthropic), Microsoft for Startups, and Cloudflare for Startups. These programs provide credits and technical support; they are not endorsements or audits. ## Live demos - Flare CI dashboard (demo mode): https://flare-actions.everyai-com.workers.dev/dashboard?demo=1&tour=1#/live - Flare Forge (agent coordination, demo repo): https://flare-actions.everyai-com.workers.dev/app?repo=bookshelf - agentprofile: https://agentprofile.everyai-com.workers.dev - Argus (browser verification console): https://argus-cloud.everyai-com.workers.dev - PlainSync: https://plainsync-open-markdown.tradephani.chatgpt.site - AIOS desktop download: https://magicteams.ai/download - EveryAI: https://every-ai.com - Idea Atlas: https://idea-atlas.everyai-com.workers.dev - Camp Revolution (client site, Sonoma County, CA): https://camprevolution.org - Muse use cases: https://muse.every-ai.com - TallTrack: https://talltrack.everyai-com.workers.dev - MCP servers (Streamable HTTP, live): https://sahadeva.magicteams.ai/mcp, https://festival-finder.magicteams.ai/mcp, https://invoice-gen.magicteams.ai/mcp, https://report-raja.magicteams.ai/mcp, https://sheet-shift.magicteams.ai/mcp - Voice platform and client systems: private; live demo on request ## Operating principle Products that work. Useful AI systems need durable memory, explicit approvals, verifiable receipts, and a kill switch. Software should carry the work forward, show its receipts, and ask for a human only when judgment matters. ## What we build ### Voice agents Phone agents that answer, qualify, and book. Speech-to-speech over real telephony, with cloned voices and interruption (barge-in) handling, responding in under 400 ms on live calls. Human takeovers are counted and reported, never hidden inside an automation rate. ### Operations agents Agents that work the inbox, CRM, messages, and finance data. They open with what needs attention, not another dashboard, and every outbound action waits for a person to approve it. Built on AIOS, our operating layer for a company. ### Agent controls The plumbing that makes agents safe to ship: MCP tools with typed schemas and approval tokens, one identity per agent with a tested kill switch, tracing and cost per agent, and verification and coordination when many agents share one codebase. ## How every system works: Agent, Job, Ask, Receipt - Agent: a teammate with a role, tools, and limits. It reads what it needs and nothing more. - Job: durable work with steps, state, and cost. It survives restarts. - Ask: a human decides where judgment matters. Money, customers, and anything irreversible stop at an Ask. - Receipt: proof of what happened, which tools ran, what it cost, and a trace you can replay. Illustrative example: a dispatch inbox receives "26,000 lb, Dallas to Houston, pickup Friday". The ops agent opens a Job to price the route, check trucks, and draft a quote. It stops at an Ask: "Send a $4,800 quote to Lone Star Freight?" After the ops manager approves on mobile, the quote is sent and a Receipt records the tools used and the trace. ## In production - Logistics: a 33-branch logistics company runs a unified inbox and opportunity engine. 57 mailboxes flow into one pipeline from enquiry to quotation to truck to trip. In production. - Voice: the MagicTeams.ai voice platform runs speech-to-speech phone agents with telephony, cloned voices, and interruption handling on real business calls. In production. - Manufacturing: a Nevada solar recycler runs a production run log with mass balancing across staged, WIP, and offtake, certificate issuance, and badge and PIN operator sign-in. Fixed-fee pilot. - Go-to-market: at AcquisitionX (Phanindra is part-time CTO and Partner), a pipeline goes from LinkedIn engagement to qualified leads, with ICP research, drafted outreach with approval and deduplication, and Smartlead campaign sync. - MCP: five small MCP utility servers submitted to Meta's Muse connector directory: InvoiceGen (invoices), ReportRaja (weekly reports), SheetShift (spreadsheet cleanup), FestivalFinder (festival dates), and Sahadeva (an astrology engine). Each answers a standard MCP handshake. Client names are left out where the work is private. Details can be walked through on a call. ## How we work with your team Most agent programs stall after the pilot. This model is built so your team can see every agent, prove it is worth running, and keep it going without us. Delivered so far: a production inbox and opportunity engine at a 33-branch logistics company, production voice agents on our own platform, a fixed-fee pilot for a manufacturer, and one-week production installs of AIOS for founder-led businesses. The full assessment, pilot, and scale model below is how we would run a larger rollout. Team: founder-led, with a two-person engineering core. Every engagement names who works on it and how much of their time it gets, in writing, before it starts. US hours: a daily overlap with US business hours for pairing, and a named on-call engineer during US peak hours for every live agent. Principles: - You own what we build. Agents, tests, dashboards, and runbooks belong to you. Your engineers pair with ours from week one, and the exit plan is written into the statement of work. - Baseline before build. We measure today's numbers first, with a method your finance team signs off, so any lift is provable against control sites. - Controls before autonomy. Read and write agents are split, approvals bind to the exact action, credentials stay in a gateway, and every agent has a tested kill switch. - Honest numbers. Metric definitions are agreed up front. Human takeovers are counted and reported. How an engagement runs: 1. Assessment (fixed fee, 4 to 6 weeks, quoted after a 30-minute scoping call): an inventory of every agent and AI tool, risk tiers, a visibility architecture, and a costed 90-day plan. 2. Pilot (2 to 3 agents): live in a few sites next to matched control sites, with a go, iterate, or stop decision on a date agreed up front. 3. Scale (in waves): rollout site by site, with handover to your team and hypercare until they run it without us. Security and data: we do not hold a SOC 2 report today. To keep client exposure low, agent work runs inside the client's environment (their identity provider, cloud, and data warehouse). We hold none of their data and never train models on it. We complete the client's security questionnaire before any access is granted. Voice data path: for each voice deployment we disclose, in writing, which telephony, speech, and model providers sit in the call path, where audio and transcripts are stored, and for how long. Our own product: AIOS, our product for founders, keeps local work on the owner's computer; its optional cloud jobs run on our infrastructure. It is not part of a client engagement unless the client chooses it in writing. Fits your stack: open standards (OpenTelemetry, OAuth, MCP), so the controls sit alongside the platforms you already run and travel with you if you change them. ## Open source - [Flare](https://github.com/everyai-com/flare-actions): CI that runs on your own Cloudflare account, with an API and MCP server agents use to verify their own changes. - [Flare Forge](https://github.com/everyai-com/flare-forge): many agents, one codebase. Agents declare intent before they edit, overlaps surface early, and every line answers why. - [Agent Ready](https://github.com/everyai-com/agent-ready): a capability manifest that decides what an agent may read and write, with locked fields and two-step confirmation on writes. - [agentprofile](https://github.com/everyai-com/agentprofile): one agent identity, skills, and memory, synced to every agent tool through a single MCP server. - [Argus](https://github.com/everyai-com/argus): parallel browser testing on Cloudflare, so a release is checked the way a customer would use it. - [PlainSync](https://github.com/everyai-com/plainsync): a local-first Markdown workspace for people and agents. ## Products - AIOS: a durable operating layer for a company, built on Agent, Job, Ask, Receipt. Desktop, cloud, and mobile. Download: https://magicteams.ai/download - AIOS for Home Services: missed calls, stale estimates, and dormant leads, ranked for the owner each morning. - AIOS Mobile: Brief, Needs you, Ask, and Pulse on iOS and Android. - EveryAI: one place to discover and use text, image, audio, code, and research models. - Magic Voice Teams: build, test, and deploy voice agents in minutes. - Also: TallTrack, Idea Atlas, Camp Revolution, Muse use cases, Lazystudio, and private builds (Callcraft, Founder OS, Video OS, and the Glow, Alarm Maker, and Swing Coach mobile prototypes). ## Skills and stack - AI systems: LLM orchestration, tool calling, RAG, MCP, memory systems, evals, approval-gated workflows, audit trails - Voice and realtime: speech-to-speech agents, telephony, streaming audio, barge-in, STT and TTS, Whisper, latency optimization - Languages and frameworks: TypeScript, JavaScript, Python, React, Electron, Node.js, SQL, SQLite, PySpark - Infrastructure: Cloudflare Workers, D1, R2, KV, Durable Objects, Workers AI, Convex, Azure, Google Cloud, GitHub Actions ## Founder background - MagicTeams.ai: Founder and AI Engineer, November 2024 to present - AcquisitionX: CTO and Partner (part-time), September 2025 to present - Every AI: Founder, April 2023 to present; the AI practice that grew into MagicTeams.ai, and the every-ai.com tools directory - Web3 Wibe: Founder, August 2022 to January 2024 - Accenture: Data Engineering Analyst and ML Engineer, July 2021 to November 2022 - B.Tech, Electrical Engineering, National Institute of Technology (NIT) Silchar, 2017 to 2021 ## About this profile - Public facts only. Client names are left out where the work is private, and no private account or customer data is published. - Last updated: 23 October 2026. - Questions: phanindra.reddy@magicteams.ai or https://cal.com/champ1288/aios.