llms.txt
Compare with other sites' llms.txt files →# oncell
> oncell (oncell.ai) is the enterprise agent platform: agents that get better
> in production. Developers write an agent in one TypeScript file
> (instructions, model, tools, budget, and the evals that say what good looks
> like) and oncell runs it through five stages: build, test, evaluate, run,
> improve. Every agent.* call crosses one runtime boundary, and that boundary
> is the run log, the trace, the cost meter per customer, the test harness,
> and the security perimeter at once. One agent per customer, team, or tenant,
> each in its own isolated cell that can be paused at ~$0, snapshotted, and
> forked. Buyers: enterprise AI platform teams, SaaS companies embedding
> agents, and AI-native products.
## Customers (agents in production, with real numbers)
- [JustCopy.ai](https://oncell.ai/customers/justcopy): JustCopy builds a small business its website, then the owner launches agents to run it: one that answers customers now, one that works in the background every day. oncell is the runtime for both, one agent per business, metered per business.
- [OpenCrew](https://oncell.ai/customers/opencrew): OpenCrew is the HQ where AI agents work as a team. Every agent in a crew runs in its own oncell cell behind OpenCrew's HQ: its own tools, its own budget, its own run log, paused at no cost when idle.
## Docs
- [Documentation](https://oncell.ai/docs): quickstart, the agent config, tools, skills, triggers, the loop, evals, the runtime, and the platform API
- [Evals](https://oncell.ai/docs#evals): declare what good looks like next to the agent; oncell eval scores it and diffs against the last run
- [Promoting a production run](https://oncell.ai/docs#evals-promote): turn any production run into an eval case
- [Budgets](https://oncell.ai/docs#runtime-budgets): daily budgets enforced by the runtime below the model
- [Isolation](https://oncell.ai/docs#runtime-isolation): one cell per customer, team, or tenant
- [Improve](https://oncell.ai/docs#runtime-improve): run logs are trajectories, evals are the reward
## Blog
- [What it costs to run a customer-facing agent per conversation](https://oncell.ai/blog/cost-per-conversation): JustCopy runs one agent per small business at $0.0006 a conversation, under a $1 per day budget. How the number is measured, and what a $9.99 plan covers.
- [The run that became a test: turning production conversations into evals](https://oncell.ai/blog/the-run-that-became-a-test): A plumber's agent invented a price. The conversation was already a run in the log, so it became an eval. How promoted runs guard every change to an agent.
- [One agent, one cell: isolating a crew of agents](https://oncell.ai/blog/one-agent-one-cell): OpenCrew runs every agent in a crew in its own cell. The tool allowlist is the cell, the blast radius is one agent, and shared work happens on forks, not turns.
- [Sync and async agents: the two kinds every business launches](https://oncell.ai/blog/sync-and-async-agents): Every business JustCopy serves launches two agents: a front desk that answers now and a follow-up that works every morning. Why a schedule is durable intent.
## Use cases (deployable agent templates)
- [oncellclaw — a Personal Assistant that Survives Your Laptop](https://oncell.ai/usecases/personal-assistant-oncellclaw): The open-source NanoClaw fork whose agents live in durable cells: persistent memory, real web research, Claude Code inside, ~$0 idle — talk to it from anywhere.
- [Backup Auditor that Demands a Monthly Restore](https://oncell.ai/usecases/backup-auditor): Verifies every backup daily for pennies — and once a month parks on a human until someone actually restores one.
- [Cert Sentinel with Deadline-Countdown Wakes](https://oncell.ai/usecases/cert-expiry-sentinel): Warns at 30/7/1 days before every cert expiry — each warning fired by a wake computed to land at the exact threshold instant.
- [Churn Detector on a Weekly Rubric](https://oncell.ai/usecases/churn-detector): Product events stream into SQLite for free; a weekly rubric pass scores every account and flags the ones drifting toward the exit.
- [Code Reviewer with a Real Shell](https://oncell.ai/usecases/code-reviewer): Paste a diff, get a senior-engineer review — produced in a sandbox where the agent runs grep on the patch instead of guessing.
- [Competitor Watcher that Reports Only Diffs](https://oncell.ai/usecases/competitor-watcher): Reads competitor changelogs and pricing pages daily, diffs them against memory, and digests only what actually changed.
- [Content Calendar with a Planner and an Executor](https://oncell.ai/usecases/content-calendar-planner): A weekly planning wake turns the backlog into a Mon-Fri plan file; a cheap daily wake reads it back and says what today is for.
- [Content Repurposer with a House Voice](https://oncell.ai/usecases/content-repurposer): Turns one blog post into a thread, a LinkedIn post, and a newsletter blurb — voice rules packaged as a versioned skill.
- [Relationship Cadence that Scores Decay Every Morning](https://oncell.ai/usecases/crm-touch-cadence): Scores your whole network for relationship decay at each daily wake and hands you the 3 most overdue people, openers drafted.
- [Self-Scheduling Morning Digest](https://oncell.ai/usecases/daily-digest): Compiles a crisp morning briefing from the team's notes, then schedules its own tomorrow — no cron box anywhere.
- [Data Cleaner that Shows Its Work](https://oncell.ai/usecases/data-cleaner): Drop a messy CSV on it, get back a normalized file plus a counted change report — cleaning rules as a skill, verification by real shell.
- [The Janitor that Cleans Its Own Cell](https://oncell.ai/usecases/data-retention-janitor): Applies declared retention rules to its own database and files every week, and publishes an exact accounting of what it purged.
- [Weekly Dependency Scout with Written Upgrade Notes](https://oncell.ai/usecases/dependency-update-scout): Sweeps your dependency list against the registry every week and writes one risk-ordered upgrade note — majors first, routine bumps after.
- [Expense Approver that Waits for Humans](https://oncell.ai/usecases/expense-approver): Auto-approves small expenses, parks at $0 waiting for a human on big ones — and survives crashes while it waits.
- [FAQ Builder Ranked by Evidence](https://oncell.ai/usecases/faq-builder): Ingests support transcripts one by one, clusters recurring questions in SQLite, and emits a FAQ ranked by how often customers actually ask.
- [Spaced Repetition as Per-Card Wake Chains](https://oncell.ai/usecases/flashcard-coach): Every flashcard owns its own wake chain — SM-2 computes each card's next review instant from your grade, and the runtime keeps 500 clocks.
- [Inbox Triager on a Haiku Budget](https://oncell.ai/usecases/inbox-triager): Labels, prioritizes, and suggests an action for every email — with skills as packaged procedures and a memory of every sender.
- [Invoice Chaser that Sleeps Between Nudges](https://oncell.ai/usecases/invoice-chaser): Politely escalating payment reminders that sleep 7 days at $0 between nudges — and stand down the moment payment lands.
- [KPI Watcher that Only Speaks When Numbers Get Weird](https://oncell.ai/usecases/kpi-anomaly-watcher): Metrics stream in for free; a daily wake rebuilds z-score baselines and alerts only past 2.5 sigma — silence is the deliverable.
- [Lead Qualifier that Escalates Only the Borderline](https://oncell.ai/usecases/lead-qualifier): Scores inbound leads against your ICP rubric — instant verdicts on the obvious, a $0 park for a human on the borderline.
- [Weekday Meeting Briefs that Skip the Weekend](https://oncell.ai/usecases/meeting-prep-briefer): Turns each weekday's calendar into one-page meeting briefs — and computes its own next weekday wake, so Friday books Monday.
- [Meeting Scheduler that Parks Between Rounds](https://oncell.ai/usecases/meeting-scheduler): Negotiates a meeting time over email, parking at $0 between counter-proposals — one run can span a week of back-and-forth.
- [Daily Brief that Learns What You Actually Read](https://oncell.ai/usecases/news-briefing): A daily topical brief whose emphasis drifts with your feedback — zero-LLM weight updates today, a differently-shaped brief tomorrow.
- [Onboarding Sequence that Skips the Activated](https://oncell.ai/usecases/onboarding-emailer): Runs a per-user onboarding email sequence — each touch schedules the next, and users who activate simply stop hearing from it.
- [On-Call Handoffs Written at the Rotation Boundary](https://oncell.ai/usecases/oncall-handoff-writer): Incidents stream into its log free all week; when the rotation turns, the wake compresses the week into a handoff doc for the next engineer.
- [Competitor Price Monitor with a Memory](https://oncell.ai/usecases/price-monitor): Watches competitor prices daily, keeps the full history in SQLite, and only speaks up when a move crosses your threshold.
- [Release Notes Writer with House Style](https://oncell.ai/usecases/release-notes-writer): Paste a commit log, get customer-facing release notes — grounded by real shell commands, styled by a versioned skill.
- [Weekly Report that Writes Itself](https://oncell.ai/usecases/report-generator): Metrics trickle in all week as zero-token SQLite inserts; every Monday a polished markdown report compiles itself and books next week.
- [Biweekly Retro as a Three-Phase Wake Cycle](https://oncell.ai/usecases/retro-facilitator): Opens a retro window, nudges mid-window, and synthesizes themes at close — one cycle spread across three phased wakes, no meeting.
- [Review Responder with a Human Gate on Anger](https://oncell.ai/usecases/review-responder): Drafts on-brand replies to app-store and G2 reviews — positives post instantly, negatives park for human sign-off first.
- [SEO Brief Writer with an Editorial Standard](https://oncell.ai/usecases/seo-brief-writer): Turns a keyword into a complete content brief — outline, entities, FAQs — grounded in SERP data and archived to a brief library.
- [Rank Tracker on Two Clocks](https://oncell.ai/usecases/seo-rank-tracker): Logs every keyword's search position daily for pennies, then writes one weekly trend report — movers first, wiggles ignored.
- [Social Queue with Two Clocks in One Agent](https://oncell.ai/usecases/social-scheduler): Publishes one queued post a day and writes a weekly engagement rollup — two interleaved wake chains running through a single agent.
- [PR Nagger that Remembers Who It Already Nagged](https://oncell.ai/usecases/stale-pr-nagger): Weekday-morning review reminders with per-reviewer fatigue memory — one bundled message per person, three per week, tone that de-escalates itself.
- [Standup Collector that Chases So You Don't](https://oncell.ai/usecases/standup-collector): Pings every teammate for their update, compiles one tidy standup, posts it — and books its own tomorrow.
- [Monthly SaaS Audit with a Human on the Trigger](https://oncell.ai/usecases/subscription-auditor): Sweeps the SaaS ledger every 30 days, flags zombie subscriptions, and asks a human before drafting a single cancellation.
- [Support Agent with Per-Customer Memory](https://oncell.ai/usecases/support-agent): Customer support that remembers every customer — and can't refund a cent without human approval.
- [Uptime Watchdog that Only Pages on Changes](https://oncell.ai/usecases/uptime-watchdog): Probes your endpoints every 5 minutes, remembers incident state, and alerts only on transitions — never twice for the same outage.
## Contact
- [Book a 30-minute working session](https://calendly.com/anup-oncell/30min?utm_source=oncell.ai&utm_medium=site&utm_campaign=docs&utm_content=llms-txt): bring one agent, leave with a loop