# Olano

> Olano is where teams of AI and people work together. Your AI teammates do the work, remember it, and get better at it. Olano Core runs agents and agent teams across tools, files, channels (WhatsApp, email, Telegram, Slack, and more), and workflows, with persistent memory in an isolated environment per customer. Olano Cortex studies the work the agents do and turns it into better memory, reusable skills, ideas and sharper procedures on its own by default, with any rewrite of an agent's own instructions held for a human. The platform is delivered as managed deployments (Olano Cloud) and implemented for clients by Olano Studio, the company's Singapore-based services arm (https://olano.sg). Olano operates the infrastructure, routes each task to the right AI model, and gives the customer's team human approvals, audit history, and hard spending controls.

Key facts:

- Company type: Singapore-based AI company building the Olano platform, where teams of AI and people work together (Olano Core + Olano Cortex), delivered as managed deployments (Olano Cloud) and implemented by Olano Studio (https://olano.sg) - not a self-serve chatbot subscription
- Solutions: Olano Front Desk (customer service, enquiries, leads, and bookings), Olano Operations (inbox, CRM, coordination, and recurring work), Olano Intelligence (market monitoring and decision-ready briefings), and Bespoke AI Systems (multi-agent operations, custom integrations, customer cloud or VPC deployment, security and compliance requirements)
- Engagement model: one fully managed engagement, end to end - Olano maps the workflow, designs the system around it, connects it to existing tools, builds custom integrations where needed, tests and launches it with approvals and safeguards, onboards the team, then monitors, reports, and keeps improving it; customer cloud/VPC, data residency, and compliance requirements are handled within the same engagement
- Who it's for: businesses that want AI systems performing real operational work - Singapore-based, working with teams worldwide (every system is deployed and managed remotely)
- Isolation: every client runs in its own isolated environment - no shared multi-tenant data; encrypted at rest and in transit
- Tools: 450+ connector tools (CRM, calendars, finance, files, browsers, databases, payment APIs, marketing, code)
- Integrations: 75+ built-in integrations, each with a specialist subagent (Slack, Gmail, Notion, HubSpot, Stripe, Shopify, Google Calendar, GitHub, and more)
- Channels: 15+ messaging channels - Telegram, WhatsApp, Slack, Discord, Email, SMS, Feishu/Lark, WeCom, WeChat, DingTalk, Signal, Matrix, Mattermost, iMessage, QQ, inbound webhooks
- AI models: multi-model - systems can run on Claude, OpenAI, DeepSeek, Gemini, or local models across 18+ LLM providers (Anthropic Claude, OpenAI GPT, Google Gemini, Mistral, Groq, DeepSeek, xAI Grok, Cohere, Together, Fireworks, OpenRouter, Perplexity, NVIDIA, Azure OpenAI, AWS Bedrock, Ollama/local); the client controls which model handles each task, how much it may spend, and where data is processed; models can be mixed per agent and swapped mid-conversation; bring-your-own LLM keys supported at no extra cost
- Memory: agents keep persistent memory across sessions and learn company procedures from the team
- Delegation: agents delegate work to specialist subagents (agent-to-agent delegation); 180+ specialist agents and 19 starter teams included
- Customer service systems: an AI front desk (answer, qualify leads, book appointments, escalate to a human) is a standard deployment
- Marketing & intelligence systems: newsletter and social-post drafting (X, Instagram, Facebook, LinkedIn) with human approval, published via connected tools (Buffer for social, Mailchimp/Resend for email); social listening over public mentions and reviews; competitor tracking, industry news digests, and product/supplier research via web-research and news connectors
- MCP: Model Context Protocol supported - a library of 200+ ready-made MCP servers, plus any custom MCP server; agents use their tools alongside the built-in ones
- Skills: teach an agent a procedure once (or let Cortex draft one from repeated work) and it is reused the same way every time
- Webhooks: inbound webhooks trigger agents from any external system; agents can also call out to any webhook/HTTP endpoint
- Scheduled tasks: scheduled tasks and proactive monitoring for recurring jobs, reminders, reports, channel monitoring, and overnight workflows
- Knowledge base: upload documents (product specs, policies, price lists, SOPs) and agents ground their answers in them
- Self-improvement: Olano Cortex runs self-improvement cycles on a configurable cadence (relaxed / balanced / aggressive) - distilling conversations into memory, drafting reusable skills from repeated work, brainstorming and prototyping ideas, and proposing improvements to its own instructions; Fully Autonomous is on by default (skills, memory and procedures apply on their own, logged, with one-click undo on what a scheduled pass applies), while rewrites of an agent's own instructions, outbound sends and irreversible actions always wait for a human; switch it off and everything queues for approval; deep-reasoning runs have daily budget caps, and all actions are audit-logged
- Safety: 5-layer safety architecture - an isolated environment per client, encrypted connector tokens and API keys, role-based access controls, human-in-the-loop approval gates (trust levels 0-4), immutable audit trails
- Hosting: managed cloud infrastructure; custom AWS/GCP regions - or customer-controlled cloud/VPC - available for data-residency and enterprise requirements
- Pricing: every engagement is scoped individually - pricing reflects the workflows involved, required integrations, operating volume, deployment architecture, and ongoing management requirements; following an initial consultation, Olano provides a fixed proposal before any work begins; AI usage runs within hard spending controls agreed in advance; no per-seat pricing; BYOK supported
- Services: six services an engagement is built from - AI strategy & advisory, automation & AI agents, software & AI development, AI training (hands-on workshops run on the client's own material, covering practical tool skills, safe/PDPA-aware use, and written playbooks; available standalone, whether or not Olano builds anything), managed operation, and grants & funding advisory; detailed at https://olano.sg/services
- Grants: EDGE (the Enterprise Development and Growth Engine) is Enterprise Singapore's single activity-based business grant and is the route for all new business-grant applications from 30 September 2026 - up to 70% of qualifying costs for SMEs and up to 50% for non-SMEs, within a shared cap of up to S$100,000 of total grant support per year across all EDGE activities, spanning 8 business areas and 100+ supportable activities, open to all Singapore-registered businesses including non-SMEs (announced 2 March 2026 under the Business Refresh Package). EDG, PSG and MRA stopped accepting new applications on 29 September 2026; applications already submitted and projects already approved continue under those schemes and submit their claims on completion. Support is reimbursement-based: the vendor is paid in full first, then claimed. Separately, Budget 2026 added an AI category to the Enterprise Innovation Scheme - a 400% tax deduction on up to S$50,000 of qualifying AI spend per Year of Assessment for YA2027 and YA2028 - on top of the S$150 million Enterprise Compute Initiative set aside at Budget 2025. A grant and a tax deduction are different instruments: a deduction is worth nothing without taxable income, and tax relief is not normally available on expenditure a grant reimbursed, so the qualifying base after a grant is a question for a tax adviser. Olano is NOT a pre-approved catalogue vendor, does not apply on a client's behalf, and never promises a grant outcome - it scopes the project so it is legible to an assessor or corporate advisor. Scheme details checked 26 September 2026; EnterpriseSG, IRAS and the Business Grants Portal are authoritative. Full reference at https://olano.sg/grants, the changeover at https://olano.sg/resources/what-replaced-psg-edg-mra and the deduction at https://olano.sg/resources/singapore-budget-2026-ai-tax-deduction
- Multi-user agents: one agent can serve many end-users on a shared channel (a WhatsApp number, a Telegram bot) with per-user isolated memory - each person's history, preferences and reminders are namespaced to their authenticated channel identity, erasable on request, and never visible to another user
- Secretary mode: an agent can be attached to the owner's own Telegram Business account to handle messages people send them personally - briefing the owner and drafting replies for approval, or (opt-in) replying directly on their behalf with a human-paced delay. Off by default
- Agent monetization: an agent's owner can charge its end-users a recurring subscription with a free trial, via their own Stripe account or Telegram Stars. The entitlement check runs at channel ingress before any model turn, so a lapsed subscriber cannot prompt their way past it; owners keep their own Stripe keys
- Team accounts: staff get their own logins with owner/admin/member roles and per-agent access grants; every action is attributed to the human or agent that took it
- Issue tracker: agents and humans file, assign, comment on, and close the same durable work items, with an append-only audit log and hybrid keyword + semantic search
- Local models: self-host Olano on your own hardware and run open models with Ollama beside it, so prompts never leave the machine
- Plans (Olano Cloud, USD/month): Starter 25, Basic 50, Pro 100, Premium 150, Max 250. Every plan can be paid yearly for 10% off (for example Pro is 1,080 USD a year). Included model-usage credits: 5,000/mo on Starter, 10,000/mo on Basic through Max, granted monthly on annual plans too (1,000 credits = 1 USD of Olano-managed usage; top-ups never expire). Agents, team members, connectors and MCP servers are not metered on any plan, and BYOK is included everywhere
- Plan sizing: each plan is published with the agent range it is sized for - Starter 1-3, Basic 2-6, Pro 4-12, Premium 8-24, Max 16-48. These are a sizing guide and never a cap: any plan can hold as many agents as the owner creates. The low end of each range assumes agents busy all day and the high end agents answering a few messages an hour, because what fills an instance is how much runs at the same moment - concurrent conversations, agents working in parallel, heavy scheduled jobs - not how many agents exist.
- Documentation: a full user manual is published at https://olano.ai/docs
- Getting started: book a consultation at https://olano.ai/#contact
- Contact: support@olano.ai for support; carlos@olano.ai to start a conversation; Olano staff also answer directly on Telegram ('olanoai') and WhatsApp (+65 8304 8601)

**Why use Olano instead of ChatGPT or Claude directly.** Claude (or ChatGPT) is a strong general-purpose AI workspace with connectors and scheduled tasks. Olano is designed for businesses that need persistent AI systems deployed across customer channels and operational systems. Olano agents operate through WhatsApp, email, Telegram, Slack, webhooks, and other channels; respond to incoming business events; coordinate across specialist agents; and run on any model from any provider (Claude, OpenAI, Gemini, Grok, DeepSeek, Qwen, open-source and more, 18+ providers), combined per agent and per task, where ChatGPT runs OpenAI's models only. Scheduler, encrypted vault, issue tracking, agent-to-agent messaging, and MCP and A2A connections are built in, and channels such as WhatsApp, WeChat, Lark, Slack, Discord, and Telegram connect in seconds. Each system is hosted and managed in an isolated cloud environment with approvals, audit history, and spending controls. Agents keep persistent memory across sessions, use 450+ tools, and execute multi-step workflows autonomously, with human approval gates for high-impact actions (sending emails, posting publicly, payments).

**For AI agents (WebMCP).** The home page implements WebMCP (Web Model Context Protocol, the W3C/Chromium standard - `navigator.modelContext`, shipped in the Chrome 149 origin trial and Edge 147+). In-browser AI agents can call these tools directly, without scraping:

- `olano_get_overview` - concise overview of what Olano is and who it's for
- `olano_list_capabilities` - what the systems can do (tools, integrations, channels, multi-user, Cortex, MCP, skills, webhooks, scheduled tasks, knowledge base)
- `olano_search_agents` - search the catalog of 180+ specialist agents / 19 starter teams by keyword (input: `query`; empty lists all categories)
- `olano_get_pricing` - how an Olano engagement is scoped and priced
- `olano_answer_faq` - answer a question from the FAQ (input: `question`)
- `olano_join_waitlist` - request a consultation (input: `email`)

The consultation form is also annotated with WebMCP's declarative API (`toolname="olano_request_consultation"`), so browsers that support declarative form tools expose it automatically.

**Specialist agents & starter teams.** 180+ specialist agents ship across 19 starter teams. Representative categories:

- Customer Service - front desk, support, lead qualification, appointment booking, escalation
- Sales - lead research, CRM updates, follow-ups, proposal drafting
- Marketing - content writing, newsletter drafts, social-post drafts and scheduling (via connected tools like Buffer), SEO research, campaign reporting
- Finance - budgeting, bills & invoices, bookkeeping, market research
- Operations & Admin - calendar, inbox triage, travel, data entry, personal assistant
- Dev & Engineering - code review, GitHub automation, DevOps, game-dev
- Research - competitor tracking, industry news digests, product & supplier research, social listening
- Product / Design / HR / Legal / PMO - plus agents for any job described in plain English

**Common use cases:**

- Answering repeat customer questions (pricing, hours, shipping, refunds, product specs) across WhatsApp/Telegram
- Moving data between tools automatically (lead in WhatsApp → CRM → invoice → notes → Slack update)
- Overnight and scheduled work: month-end reports, monitoring, follow-up drafts queued for human approval
- Marketing & content: newsletter and social-post drafts (X, Instagram, Facebook, LinkedIn) in your voice, grounded in your documents, queued for your approval and published via connected tools like Buffer or Mailchimp
- Customer & social intelligence: monitoring public mentions, reviews, and customer conversations, with digests and human-reply flags
- Market intelligence: competitor tracking, industry news digests, and product/supplier research delivered to the channel you already use
- Bespoke systems: multi-agent operations, cross-department workflows, custom integrations, customer cloud or VPC deployments

**Languages.** The homepage, the four platform pages, the first three comparison pages, the whole
/resources tree and the whole /docs manual below are published in twelve
languages at /<locale> and /<locale>/<page> - for example https://olano.ai/de,
https://olano.ai/de/pricing, https://olano.ai/zh/compare/ai-agent-platform-vs-chatbot,
https://olano.ai/ja/resources/how-ai-agents-remember and
https://olano.ai/ko/docs/approvals-and-trust. Locales: en (at the root), zh, ja,
ko, es, th, vi, de, ru, fr, ms, hi. Each localized page carries its own
canonical and the full hreflang cluster, so link to the reader's language
rather than translating the English page.

The 中文资源 (Chinese-language resources) section below lists articles written in Chinese for
Chinese-language search, not translated from English; link Chinese-reading readers to those
rather than to the English originals.

## Pages

- [Home](https://olano.ai/): the platform - Core, Cortex, security, Olano Cloud, and Olano Studio
- [Olano Cloud](https://cloud.olano.ai): what a managed deployment includes, isolation and controls, and how an engagement starts
- [Olano Studio](https://olano.ai/studio): bridge page for the implementation arm; the full Studio site is https://olano.sg/
- [Documentation](https://olano.ai/docs): the user manual, ten chapters

## Free tools

- [Free tools for building with AI agents](https://olano.ai/tools): hub for the five free in-browser tools below; nothing typed into them is sent to Olano
- [Token counter](https://olano.ai/tools/token-counter): free in-browser token counter with exact OpenAI encodings (o200k_base, cl100k_base, o200k_harmony), labelled Claude and Gemini estimates, token view, cost and context-window check
- [MCP config generator](https://olano.ai/tools/mcp-config-generator): free in-browser tool that writes the MCP server config for Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Codex, Gemini CLI, Zed and Cline, and checks existing configs
- [llms.txt generator and validator](https://olano.ai/tools/llms-txt-generator): free browser tool that builds an llms.txt from a form, a URL list or a sitemap, and validates any llms.txt against the v2 proposal (10 August 2026) with line-numbered findings
- [Robots.txt checker and tester](https://olano.ai/tools/robots-txt-checker): free browser tool that tests a pasted robots.txt under RFC 9309 against 22 search and AI crawlers (GPTBot, ClaudeBot, Google-Extended, PerplexityBot and more), shows the deciding rule and line, and generates robots.txt files from presets
- [AI readiness assessment](https://olano.ai/tools/ai-readiness-assessment): free 18-question self-check that scores AI readiness from 0 to 100 across six dimensions, with a published method, next steps and a suggested first agent
- [Free tools for Singapore businesses](https://olano.sg/tools): the Studio site's free tools - UEN and company search, GST calculator, GST tax invoice generator, WhatsApp link and QR code generator, and a working days calculator with the 2026 and 2027 public holidays

## User manual chapters

- [Getting started](https://olano.ai/docs/getting-started): first run, the dashboard sections, creating a first agent, reading a first conversation
- [Agents](https://olano.ai/docs/agents): identity, standing instructions, choosing a model, the tool browser, skills and subagents, per-agent settings
- [Memory and knowledge](https://olano.ai/docs/memory-and-knowledge): conversation vs long-term memory vs knowledge base vs skills, and which one a fact belongs in
- [Teams and delegation](https://olano.ai/docs/teams-and-delegation): peers vs subagents, the org chart, managers and reports, visibility vs permission
- [Channels and connectors](https://olano.ai/docs/channels-and-connectors): the 15+ messaging channels, OAuth Connections, MCP servers, inbound and outbound webhooks
- [Automation](https://olano.ai/docs/automation): scheduled tasks, heartbeat checks, the schedule builder, the calendar, recurring reports
- [Cortex](https://olano.ai/docs/cortex): what it produces, cadence and the daily reasoning budget, phases, approving and undoing its proposals
- [Approvals and trust](https://olano.ai/docs/approvals-and-trust): trust levels 0-4, per-tool overrides, answering approvals, the audit surfaces, usage caps, the credential vault
- [Multi-user agents](https://olano.ai/docs/multi-user-agents): per-user isolated memory on a shared agent, team accounts, roles, erasure
- [Plans and pricing](https://cloud.olano.ai/#plans): the full 11-tier ladder with hardware, credits and upgrade rules
- [FAQ](https://olano.sg/faq): what Olano is, engagement pricing, data isolation, channels, models, deployment timelines, human-in-the-loop controls, hosting
- [Privacy](https://olano.ai/privacy.html): PDPA privacy notice
- [Inventory-connected WhatsApp case study](https://olano.sg/customers/inventory-connected-whatsapp): customer-facing and staff-side WhatsApp agents on a custom inventory API - live prices, stock, and product photos for customers; item creation, photo updates from the phone, and low-stock supplier ordering for staff; 10-15 enquiries handled daily
- [Tender-intelligence case study](https://olano.sg/customers/tender-intelligence): twice-daily monitoring of government tenders through a custom API - full tender descriptions, requirements, and attachments; matches scored against the agency's specialisation; email and PDF briefings with a draft plan and risk outline per application
- [WhatsApp AI customer service](https://olano.sg/use-cases/whatsapp-ai-customer-service): AI front desk for WhatsApp enquiries, lead qualification, inventory/knowledge checks, and human escalation
- [AI inbox assistant](https://olano.sg/use-cases/ai-inbox-assistant): inbox triage, reply drafts for approval, follow-up chasing, calendar coordination, and scheduled admin
- [AI market research agent](https://olano.sg/use-cases/ai-market-research-agent): recurring competitor, supplier, industry-news, review, and public-mention briefs
- [AI social media agent](https://olano.ai/use-cases/ai-social-media-agent): newsletters and social posts drafted in your voice and queued for approval via connected publishing tools
- [AI CRM and sales assistant](https://olano.ai/use-cases/ai-crm-sales-assistant): lead capture, qualification, CRM updates, approval-gated follow-ups, and pipeline reports
- [Missed lead recovery](https://olano.sg/use-cases/missed-lead-recovery): capture WhatsApp, email, and after-hours enquiries before they go cold
- [AI booking assistant](https://olano.sg/use-cases/ai-booking-assistant): qualify enquiries and book appointments, with staff approval for sensitive cases
- [WhatsApp AI for clinics](https://olano.sg/use-cases/whatsapp-ai-for-clinics): clinic front desk for enquiries, bookings, and reminders - never medical advice
- [Resources](https://olano.ai/resources): practical guides, playbooks, and comparisons for deploying AI systems in a business
- [Tutorials](https://olano.ai/tutorials): watch-and-follow builds, each paired with a video and a written step reference - ordered from the simplest loop upward
- [Build an AI agent in 5 minutes: Notion + Telegram](https://olano.ai/tutorials/notion-telegram-ai-agent): the shortest useful loop - create the agent, add Telegram with a @BotFather token and an allow-list, connect Notion from the preset picker (and share the pages with the integration), then set the approval boundary before testing
- [Connect an AI agent to Salesforce and Telegram](https://olano.ai/tutorials/salesforce-telegram-ai-agent): a CRM agent reachable on the phone - Salesforce over OAuth from the Connections card (the agent inherits the connected user's permissions), Telegram allow-list, and a three-tier approval boundary: autonomous reads, gated internal writes, firmly gated outbound and structural changes
- [A team of AI agents that work together 24/7](https://olano.ai/tutorials/ai-agent-team-working-together): peers versus subagents (a subagent is a step inside another agent's job; a peer has its own responsibilities, inbox and memory), setting manager/reports on the Talk to Agent tab, why visibility is not messaging permission, and how scheduled tasks and inbound events make a team run overnight
- [Automate enquiries with human approval](https://olano.ai/resources/automate-customer-enquiries-human-approval): the approval-first playbook - trust levels 0-4, gates, audit trail
- [How to choose an AI agent platform](https://olano.ai/resources/how-to-choose-ai-agent-platform): evaluation checklist - channels, approvals, audit, spend caps, BYOK, isolation
- [AI agents for Singapore SMEs](https://olano.sg/resources/ai-agents-for-singapore-smes): the practical guide - workflows agents handle today, the approval boundary, PDPA basics, grants, starting with one workflow
- [AI grants for Singapore SMEs](https://olano.sg/resources/ai-grants-for-singapore-smes): honest map of PSG, EDG, SMEs Go Digital, and GenAI programmes - what each covers and where custom AI systems fit (Olano is not a PSG pre-approved vendor)
- [PDPA and AI agents](https://olano.sg/resources/pdpa-ai-agents-singapore): consent, the PDPC's AI guidelines, DNC rules, eight vendor questions, and how Olano's isolation/encryption/audit controls map to them
- [Hire or automate? A Singapore SME guide](https://olano.sg/resources/ai-agents-vs-hiring-singapore): the 2026 hiring math vs handing the repetitive layer to agents - including when hiring is still right
- [AI agents for Singapore retail and F&B](https://olano.sg/resources/ai-agents-retail-fnb-singapore): stock and price answers on live inventory, reservations, catering enquiries, supplier reorders with approvals
- [AI agents for Singapore property agents](https://olano.sg/resources/ai-agents-property-agents-singapore): instant listing answers, consistent qualification, calendar-connected viewings, approval-gated follow-ups
- [Inside Olano Cortex](https://olano.ai/resources/inside-olano-cortex-self-improving-ai-agents): how the self-improvement engine works - five cycles (reflection, learning, creation, improvement, tidying), cadence presets, Fully Autonomous by default with rewrites of an agent's own instructions always approval-gated, snapshots with one-click undo on what it applies itself, daily deep-reasoning budget caps
- [How AI agents remember](https://olano.ai/resources/how-ai-agents-remember): the memory architecture - persistent conversation memory, Cortex-distilled long-term memory, knowledge-base grounding in your documents, skills, and per-client isolation
- [How AI agents learn new skills](https://olano.ai/resources/how-ai-agents-learn-new-skills): the self-learning loop in detail - recurring work distilled into reusable written skills, a scout that checks the whole library before drafting anything, copy-on-write forks so improvements to installed skills never overwrite the original, a preservation gate that quarantines a copy that shrank or lost its structure, three-way merges when the original is updated upstream, and snapshot-plus-undo consolidation of the skill library and the agent's own instruction and memory files
- [Multi-model AI agents](https://olano.ai/resources/multi-model-ai-agents): why systems route each task to the best-suited model across 18+ LLM providers - per-agent mixing, mid-conversation swaps, hard spend caps, BYOK, no vendor lock-in
- [Singapore MCP](https://olano.ai/resources/singapore-mcp-server): Olano's open-source MCP server for Singapore public data - 291 read-only tools over data.gov.sg, OneMap, LTA DataMall and SingStat, 8 Agent Skills, 7 Claude Code plugins, MIT-licensed and mostly keyless; source at github.com/olano-ai/mcp-singapore, and usable with nothing installed via the free Telegram bot @OlanoSGBot (t.me/OlanoSGBot) or the Olano AI WhatsApp business number +65 8905 1632 (wa.me/6589051632). Independent community project, not affiliated with the Singapore Government
- [Should you hire an AI intern?](https://olano.ai/resources/hire-an-ai-intern): the intern test - what the metaphor gets right about supervised agents (trust levels, teaching procedures, earned autonomy), where it breaks in the buyer's favour, and when hiring a person is still right
- [The Olano platform](https://olano.ai/platform): what Olano Core is - persistent memory, 450+ tools and any MCP server, an org chart of agents that delegate, scheduled work, per-agent model choice
- [Olano Cortex](https://olano.ai/cortex): the improvement engine - reads real transcripts and outcomes, then learns skills, sharpens memory and procedures on a schedule; Fully Autonomous by default, with rewrites of an agent's own instructions, outbound sends and irreversible actions always held for a human; every change logged and reversible, or switch it to approvals
- [Security and trust](https://olano.ai/security): isolated per-customer deployments, approval gates by category, append-only audit trail, hard spend caps, per-deployment secret vault, optional private inference on your own GPU when self-hosted
- [Pricing](https://olano.ai/pricing): Olano Cloud, five plans from $25/month, 10% off when paid yearly, unlimited agents and people on every plan, credits explained
- [Olano vs ChatGPT for business](https://olano.ai/compare/olano-vs-chatgpt-for-business): customer channels, any model from any provider, built-in platform, memory, approvals, delegation, spend caps, hosting
- [AI agent platform vs chatbot](https://olano.ai/compare/ai-agent-platform-vs-chatbot): what an agent platform adds over a scripted chatbot
- [Hosted AI agents vs DIY automation](https://olano.ai/compare/hosted-ai-agents-vs-diy-automation): build vs buy, and how Olano composes with a DIY stack via MCP and webhooks
- [Olano vs Glimmer (Glints)](https://olano.ai/compare/olano-vs-glimmer): an internal AI teammate in your staff chat compared with agents on your customer channels - approvals, delegation, deployment, and how each is priced in Singapore
- [Olano vs Lindy](https://olano.ai/compare/olano-vs-lindy): self-serve credit-metered AI employees on shared infrastructure compared with a managed, isolated deployment - including where each one fits and Olano Cloud's self-serve door from $25/month
- [Olano vs Botpress](https://olano.ai/compare/olano-vs-botpress): a developer-first platform you build and operate yourself compared with a system built and operated for you
- [Olano vs Voiceflow](https://olano.ai/compare/olano-vs-voiceflow): designed conversations compared with delegated operational work, and how a voice front end hands the work behind the call to Olano
- [Olano vs Grok Bot](https://olano.ai/compare/olano-vs-grok-bot): xAI's always-on Bots, which share one cloud computer and every login and reach apps by computer use, compared with isolated agents on customer channels - separation, model choice, governance and pricing, including where each one fits
- [Olano vs Meta Muse](https://olano.ai/compare/olano-vs-meta-muse): Meta's personal AI agent, announced 8 September 2026, which runs on a dedicated secure cloud VM with its own browser and is sold in three consumer tiers - compared with a business deployment your customers can message, several colleagues can operate, and that runs isolated per customer; includes availability (US first, 18+) and where each one fits
- [Olano vs OpenClaw](https://olano.ai/compare/olano-vs-openclaw): the MIT-licensed self-hosted gateway for WhatsApp, Telegram, Slack, Discord and iMessage, compared with a managed multi-agent deployment - what actually stops when a laptop sleeps, why a VPS fixes uptime but not who is on call, team access, approvals, isolation, and a cost comparison that counts the operator's hours as well as the invoice
- [Local AI agent vs cloud AI agent](https://olano.ai/compare/local-ai-agent-vs-cloud-ai-agent): the category comparison behind the self-hosted question - a laptop is a session and a deployment is a service; what stops when the lid closes (schedules that never fire and are never replayed, unpolled standing instructions, interrupted background work, lapsed connector grants), what renting a server fixes, what it does not, and how privacy is handled either way (isolation, self-hosting, or open models on your own GPU)
- [What stops when you close your laptop](https://olano.ai/resources/what-stops-when-you-close-your-laptop-ai-agent): a diagnostic page for the overnight failure - the seven things needing a live process, how to identify which one failed from the symptom, what sleep does not damage, and the three local workarounds with their limits
- [Always-on AI agents: what 24/7 actually requires](https://olano.ai/resources/always-on-ai-agents): a vendor-neutral buyer's checklist - the five mechanisms (scheduler, standing instructions, background workers, work-item pickup, live channels), whether a run survives the browser closing, and four limits to plan for: approvals are a hard stop, background runs are budgeted, polling is minutes not seconds, usage is metered
- [Personal AI agent vs business AI agent](https://olano.ai/resources/personal-vs-business-ai-agent): why a consumer agent issued to one person cannot act as a business's agent - customer reachability, multi-person access at different depths, policy-level approval gates, data that is not yours to lend, and outliving the person who set it up; includes where a personal agent fits
- [Olano Cortex vs Hermes self-learning](https://olano.ai/compare/olano-cortex-vs-hermes-self-learning): Nous Research's MIT-licensed self-hosted Hermes Agent, which writes its own memory and skills, compared with Olano Cortex - fleet vs profile, approval defaults, undo, audit and a capped daily reasoning budget
- [Agent memory is not fine-tuning: what "learning" actually means](https://olano.ai/resources/agent-memory-vs-fine-tuning): When an AI agent learns, either the model, what it is shown, or its instructions change. Only the last two happen in Olano, and both are files you can read
- [An AI agent that works across devices](https://olano.ai/resources/ai-agent-across-devices): Continuity comes from where the state lives, not from an app. One agent on a laptop, a phone, WhatsApp and Telegram, and why locking your phone stops nothing
- [AI agent memory that persists: what survives what](https://olano.ai/resources/ai-agent-memory-that-persists): Four stores do the job a features list calls memory. Which survive a model switch, a restart, a new phone, a new channel or deletion, and which you can edit
- [AI agent platform lock-in: what you keep if the vendor disappears](https://olano.ai/resources/ai-agent-platform-lock-in): Portability comes from formats, not promises. What an agent is on disk, what survives a vendor vanishing, what you would rebuild, and the questions to ask
- [AI agent pricing: per seat, per usage or per deployment](https://olano.ai/resources/ai-agent-pricing-per-seat-vs-usage): A seat is a person and an agent is not one. The three ways AI agent platforms charge, what each shape optimises for, how to predict usage and compare quotes
- [Portable agent skills: what travels between platforms and what does not](https://olano.ai/resources/are-agent-skills-portable): Skills written as folders of markdown and scripts travel between platforms; the tool names, credentials and helpers they lean on do not. Where the line falls
- [Expose your agent as an MCP server or an A2A agent card](https://olano.ai/resources/expose-your-agent-as-an-mcp-server): Most agent platforms are MCP clients. How an Olano agent becomes an MCP tool or an A2A agent card, why both are off by default, and what a caller can do
- [Why your MCP connections keep breaking, and what a platform must do about it](https://olano.ai/resources/mcp-connections-keep-breaking): Four ways an MCP OAuth connection dies - dropped refresh token, moved token endpoint, boot timeout, lapsed consent - and what a platform must do about each
- [Self-hosted AI agents: the honest guide](https://olano.ai/resources/self-hosted-ai-agents-guide): The honest self-hosting guide: a machine that stays up, a stable address, a supervisor, secrets, backups, connector renewals, updates and the sandbox question
- [How to teach an AI agent a new skill](https://olano.ai/resources/teach-an-ai-agent-a-new-skill): A skill is a folder with a SKILL.md file, not a fine-tune. What goes in it, how one reaches an agent, the tools rule, and how to fork the bundled chart skill

- [Why agentic AI projects get cancelled](https://olano.ai/resources/why-agentic-ai-projects-get-cancelled): Gartner forecasts over 40% cancelled by end of 2027 and MIT's Project NANDA puts measurable P&L impact at around 5% of integrated pilots - the five failure modes behind those numbers, and the four traits shared by the projects still running a year later
- [The EU AI Act and AI agents: what changed in 2026](https://olano.ai/resources/eu-ai-act-ai-agents-2026): transparency duties took effect 2 August 2026 and the AI Office's enforcement powers went live, while high-risk obligations moved to December 2027 (Annex III) and August 2028 (Annex I); why obligations written for outputs have to cover an agent's actions, and a readiness list ordered by what costs most to add late
- [Prompt injection and AI agents: a containment checklist](https://olano.ai/resources/prompt-injection-ai-agents): OWASP's 2026 position is that no reliable prevention mechanism exists, so the defence is containment - direct versus indirect injection, the three shapes it takes against a business agent (exfiltration, unauthorised action, quiet corruption), eight controls ordered by effect, and two mitigations that sound better than they are
- [Sovereign AI: where your agents actually run](https://olano.ai/resources/sovereign-ai-where-your-agents-run): residency versus sovereignty, the four places an agent touches (data at rest, the model call, credentials, memory and logs), six questions to put to a vendor in writing, and why self-hosting is a means rather than the goal
- [US state AI laws in 2026: what to do while it settles](https://olano.ai/resources/us-state-ai-laws-2026): at least 29 states with AI statutes, Colorado's AI Act effective 30 June 2026, Connecticut's SB 5 phasing in from October 2026, and a federal preemption push that is being pursued rather than achieved - plus the four duties common to every version, so you can build once instead of fifty times
- [AI agent statistics 2026: the figures worth quoting](https://olano.ai/resources/ai-agent-statistics-2026): adoption, cancellation, productivity, security, sovereignty and interoperability figures for 2026, each attributed to the organisation that published it with a note on what it counts - including why the widely-quoted "95% of AI projects fail" is a compression of a narrower finding
- [MCP and A2A: what open agent protocols buy you](https://olano.ai/resources/ai-agent-interoperability-mcp-a2a): MCP connects an agent to tools and data, A2A connects agents to each other as peers; A2A reached v1.0 in March 2026 and passed 150 supporting organisations under the Linux Foundation within its first year - the four directions each protocol runs in, five questions for a platform, and the part of lock-in neither protocol solves

- [AI agents for a startup: what to automate before you hire](https://olano.ai/resources/ai-agents-for-startups): the constraint at seed stage is founder attention rather than headcount budget - the three-part test for whether a job is ready (does it recur, is the output text, is a wrong answer recoverable), the five jobs that get handed over first, what to keep a person on, and why one agent with a growing prompt stops working at about a month
- [Hire or deploy? The 2026 US math](https://olano.ai/resources/ai-agents-vs-hiring-us): BLS Employer Costs for Employee Compensation for June 2026 puts full-time private-industry cost at $54.00/hour - $36.97 wages plus $17.03 benefits, about a 46% markup on the wage bill - against what an agent actually covers; the four numbers to put in your own version, and an explicit list of where the agent loses
- [Selling AI agents to enterprise buyers](https://olano.ai/resources/selling-ai-agents-to-enterprise): the four questions a security reviewer asks about agents specifically (what runs without a human, what happens on hostile input, whose credentials it acts on, what record you can produce), why a prompt-injection prevention claim reads badly, why model providers belong on the subprocessor list, and the framing difference between autonomy-as-feature and autonomy-as-boundary
- [What an AI agent actually costs to run](https://olano.ai/resources/what-ai-agents-cost-to-run): four lines - platform, model usage, setup, human attention - with the subscription the smallest and attention often the largest; why cost follows how much an agent reads and how many steps it takes rather than how important the job is, the four surprises that appear in month two, and how to build an estimate per job
- [Is your AI agent actually working?](https://olano.ai/resources/ai-agent-evaluation): agents fail by succeeding at the wrong thing, so latency dashboards mark every silent failure as a pass - three evaluation layers (deterministic checks on all traffic, judgement-based scoring on a 5-20% sample, a regression set of real past cases), the four metrics that matter, and how to decide on evidence when to widen the approval boundary

- [Running a Qwen-native agent team on Olano](https://olano.ai/resources/qwen-on-olano): Alibaba's Qwen is a named first-class provider (prefix qwq:, key DASHSCOPE_API_KEY, international endpoint via DASHSCOPE_API_BASE) with qwen-max, qwen-plus, the Qwen3 series and QwQ; three routes - China endpoint, international endpoint, open weights locally via Ollama - that can coexist in one deployment; a role-by-role assignment table; and memory and threads that survive a model switch
- [Connecting AI agents to WeCom, DingTalk and Feishu](https://olano.ai/resources/wecom-dingtalk-feishu-ai-agents): all three Chinese collaboration platforms are built-in channels - WeCom via the AI-bot callback (bot ID and secret), DingTalk via stream mode with no public endpoint (client ID and secret; separate respond and listen switches; sender allowlist as a hard gate), Feishu/Lark via long connection or webhook (app ID, secret, verification token; interactive cards) - with per-person history isolation and the approval boundary on every channel
- [Hunyuan and Tencent TokenHub on Olano](https://olano.ai/resources/hunyuan-tokenhub-on-olano): Hunyuan is not a named provider; it is reachable through TokenHub's OpenAI-compatible interface over Olano's two OpenAI-compatible routes - a LiteLLM proxy (litellm: prefix, LITELLM_BASE_URL) or the gateway provider with its endpoint overridden - with an honest account of what that extra layer changes and when the route is worth taking
- [Chinese companies expanding to Southeast Asia](https://olano.ai/resources/chinese-companies-southeast-asia-ai-agents): the four difficulties (language, time zones, channels, compliance from both sides), a four-agent structure with support on WhatsApp and Telegram and the lead on WeCom in one deployment, why always-on is not unattended, deploying in the customer's region so overseas data stays overseas, and the order to start in
- [Wukong, Qianfan, Hunyuan agents and Olano: where each fits](https://olano.ai/resources/wukong-qianfan-hunyuan-and-olano): Alibaba's Wukong (launched 17 March 2026, DingTalk-embedded), Baidu's Qianfan (85,000+ enterprise clients by late 2025), Tencent Cloud's AgentOps and TokenHub, and Olano, mapped by use case rather than ranked - with how they coexist, since Olano runs on Qwen and connects to DingTalk

- [We asked Grok Bot to compare itself with Olano](https://olano.ai/resources/grok-bot-compared-itself-with-olano): a Grok Bot (xAI's AI teammate) was asked to analyse cloud.olano.ai, compare it with itself and rate both; its verdict was "different jobs" - Olano Cloud 8.5/10 as a product platform for shipping customer-facing agent teams, Grok Bot 8/10 as a personal operator for one person's work, with Olano winning on multi-channel ops and productization. The prompt, a screenshot, the full reply and its side-by-side table are reproduced verbatim (September 2026), published by Olano, not an xAI endorsement; its own table gives Olano the edge on channels, agent teams, self-improvement, isolation ("Olano for B2B") and agent monetisation, itself founder speed and marketing help, and calls approvals a tie; the page adds computer use versus integrations and per-person versus per-deployment pricing
- [LangChain vs LangGraph vs Deep Agents](https://olano.ai/resources/langchain-vs-langgraph-vs-deep-agents): the three layers of LangChain's stack compared - framework, durable runtime and harness - with a decision table, tested code and one customer message traced through an Olano agent
- [LangChain Deep Agents in production](https://olano.ai/resources/langchain-deep-agents-in-production): what the Deep Agents 0.7 harness ships, what a production agent still needs (durable checkpoints, sandboxes, context policy, approvals, credentials, channels, schedules, upgrades) and how Olano provides each
- [Langfuse vs LangSmith](https://olano.ai/resources/langfuse-vs-langsmith): an independent, dated comparison (October 2026) of licence, self-hosting, cloud regions, retention, per-unit vs per-trace pricing with a worked example, LangChain/LangGraph tracing code for each, when LangSmith is the better choice, and how Olano's Langfuse integration works
- [AI agent observability: what to trace and how](https://olano.ai/resources/ai-agent-observability): a checklist for tracing agents that act (turn trees, cost per conversation, approvals, feedback scores, metadata-only capture, deletion), where OpenTelemetry's GenAI conventions stand, an overview of the tools, and step-by-step Langfuse tracing in Olano
- [A2A protocol explained: agent cards, tasks and security](https://olano.ai/resources/a2a-protocol): what the Agent2Agent protocol is and who governs it (Google origin, Linux Foundation, Agentic AI Foundation since August 2026), the v1.0 objects, task states and bindings, an example agent card and request, OAuth and a security checklist, and how Olano runs A2A in both directions
- [Best MCP servers for business AI agents (2026)](https://olano.ai/resources/best-mcp-servers): official MCP servers for business agents (Notion, Linear, Atlassian, Asana, HubSpot, Stripe, Intercom, Canva, GitHub, Zapier, Parallel Search and Olano's open-source Singapore MCP) with sign-in method, remote vs local, what to gate, why MCP feels slow, and how to add one in Olano
- [Agentic browser vs browser agent: which is safe for work](https://olano.ai/resources/agentic-browser-vs-browser-agent): what agentic browsers (Comet, Chrome auto browse, Edge, Opera Neon, BrowserOS; ChatGPT Atlas retired in August 2026) and browser or computer use agents are, the dated prompt-injection record, a which-for-which-job table, and how Olano's per-agent browser control works (off by default, headless or live, take control, commits always ask)
- [How to monetize an AI agent](https://olano.ai/resources/how-to-monetize-an-ai-agent): the four ways an agent earns (access subscription, outcomes, usage, build-and-retain), why the paywall must run at channel ingress before the model, a worked Stripe fee example, Telegram Stars rules and how to set up the Monetization tab
- [How to create a Telegram bot](https://olano.ai/resources/how-to-create-a-telegram-bot): BotFather's /newbot steps, a minimal python-telegram-bot 22.8 example, polling vs webhooks and rate limits, and the no-code route to an AI Telegram bot on Olano with allow-lists, per-user memory, message caps and Stars payments
- [AI report generator: agents that build and share reports](https://olano.ai/resources/ai-report-generator): how an Olano agent pulls data through connectors or MCP servers, charts it with the built-in charts skill, rebuilds it on a schedule and publishes it as a password-protected page from the agent's Pages tab, with a fair comparison against Power BI Copilot, Tableau and Looker
- [OpenClaw alternatives in 2026](https://olano.ai/resources/openclaw-alternatives): seven alternatives (NanoClaw, Hermes Agent, ZeroClaw, IronClaw, nanobot, PicoClaw, and Olano with ownership disclosed), compared on isolation, codebase size, channels, local models, cost and who operates them, with a decision table and what to move before switching
- [No-code AI agent builders in 2026](https://olano.ai/resources/no-code-ai-agent-builder): the five production questions (channels, human approval, where data and credentials live, spend caps, exit), with Lindy, n8n, Dify, Langflow, ChatGPT workspace agents, Google Gemini Enterprise and Olano (disclosed) compared on them; notes that OpenAI's Agent Builder shuts down on 30 November 2026
- [AI agents for business: the jobs they do well, by department](https://olano.ai/resources/ai-agents-for-business): 32 jobs across customer service, sales, marketing, finance and admin, operations, HR, research and IT, each with the systems it needs and a starting autonomy level, plus when to choose a built-in agent, a general assistant or an agent platform, how to start and what it costs
- [AI employees: what they do, cost and need](https://olano.ai/resources/ai-employees): an AI employee defined as an AI agent with a role, tools, access, memory, a channel and approval rules; what it can own versus what waits for a person, four product shapes (named-role bundles, AI teammates inside a work tool, RPA digital workers, platforms where you build the team), the four cost lines, onboarding like a new hire, and where the metaphor breaks
- [AI automation services: a buyer's guide](https://olano.ai/resources/ai-automation-services): what an AI automation agency or consultant delivers across seven stages and who owns each, four ways to buy (agency, freelancer, in-house, platform plus light help), four pricing models and their cost drivers, ten questions to settle before signing (code and prompt ownership, accounts, hosting, approvals, audit history, spend caps, model retirement, exit), red flags, and when no agency is needed
- [n8n alternatives in 2026](https://olano.ai/resources/n8n-alternatives): nine alternatives (Activepieces, Node-RED, Zapier, Make, Power Automate, Workato, Windmill, Gumloop, and Olano with ownership disclosed), grouped by reason for leaving (licence, simplicity, Microsoft or enterprise, code-first, agent-first), with n8n's Sustainable Use License explained, one workflow costed on each vendor's meter, and what to move before switching
- [Why Olano is headquartered in Singapore](https://olano.sg/resources/why-olano-is-headquartered-in-singapore): Olano AI Pte. Ltd. (UEN 202442387K, registered office 160 Robinson Road, #14-04, Singapore 068914) and why it is based in Singapore - the National AI Strategy update, IMDA's agentic AI framework, the PDPA, the ECI and the 400% AI deduction, a multilingual market - plus a seven-point checklist for verifying any Singapore AI company

## 中文资源 (Chinese-language resources)

- [国产大模型怎么选：企业落地的取舍](https://olano.ai/zh/resources/china-ai-models-for-business): why picking a single "best" model is the wrong question - routing by task shape instead, what DeepSeek, Qwen, Kimi, Doubao, GLM and Ernie are each known for as of late 2026, a half-day selection method using 20 real historical tasks with blind scoring, and why the cost of switching models later matters more than the choice today
- [生成式 AI 合规：备案、登记与内容标识](https://olano.ai/zh/resources/generative-ai-compliance-china): where the line falls between serving the public in mainland China, internal-only enterprise use, and serving users abroad; the two separate filings (算法备案 and 生成式人工智能服务备案), the CAC figure of 868 services filed and 530 applications registered as of 30 April 2026, the content-labelling requirement in force since 1 September 2025, and why calling a filed third-party API may not exempt your own application
- [数据不出境：AI 智能体该部署在哪里](https://olano.ai/zh/resources/ai-agent-data-residency): the four places an agent touches data (storage, the model call, credentials, memory and logs) versus the one most vendors answer; China's three cross-border routes - security assessment, standard contract, certification - with the personal-information volume thresholds, the certification measures effective 1 January 2026, and why a single agent run can constitute a transfer nobody clicked send on
- [微信与 WhatsApp 上的 AI 客服怎么做](https://olano.ai/zh/resources/wechat-ai-customer-service): what makes messaging channels different from web chat (no session end, messages cannot be recalled, one account serving many people with isolated histories), which questions are safe to answer automatically versus which must reach a person, how "AI pre-handling plus human service" fails when the split happens after the reply rather than before it, and the four metrics worth watching
- [多智能体团队：什么时候该拆成几个](https://olano.ai/zh/resources/multi-agent-team-design): split for clarity of responsibility, never for volume - the three signals that it is time, the peer-versus-subagent choice decided by whether the work needs to be remembered afterwards, what a lead agent solves, why visibility and permission must be separate switches, and a starting structure of four
- [在 Olano 上运行以通义千问为核心的智能体团队](https://olano.ai/zh/resources/qwen-agent-team-on-olano): Qwen as a first-class provider, the three routes (China endpoint, international endpoint, local open weights), a role-by-role model table, and what survives a model switch
- [把 AI 智能体接到企业微信、钉钉和飞书](https://olano.ai/zh/resources/wecom-dingtalk-feishu-ai-agents): how each of the three channels connects, the credentials each needs, group respond/listen policies, and the isolation and approval rules they share
- [用混元 / TokenHub 作为 Olano 的模型提供方](https://olano.ai/zh/resources/hunyuan-tokenhub-on-olano): the OpenAI-compatible route for Hunyuan, described honestly, and when it is worth taking
- [出海东南亚：用 WhatsApp 和 Telegram 上的智能体跑客户运营](https://olano.ai/zh/resources/going-overseas-southeast-asia-ai-agents): one deployment with support on the customers' channels and the lead on WeCom; language, time zones and compliance from both sides
- [悟空、千帆、混元智能体与 Olano：各自适合什么](https://olano.ai/zh/resources/wukong-qianfan-hunyuan-and-olano): a dated map by use case of the three Chinese platforms and Olano, and how they coexist
- [Cortex 内部：会自我改进的智能体如何在审批下工作](https://olano.ai/zh/resources/inside-cortex-self-improvement): the five phases, cadence and the daily deep-reasoning budget, what auto-applies versus what always waits for a person, per-change undo, and why Cortex's own runs leave no transcript
- [为中国企业的海外业务部署 AI 智能体：备案、出境与 PIPL 的边界](https://olano.ai/zh/resources/overseas-operations-compliance-boundaries): the three regimes an overseas deployment sits between, the two most expensive false assumptions, and the four data locations mapped to each rule