# Perch AI > Perch AI is a verifiable AI coworker for legal, finance, research, and coding. It reads the files in your folder, does the work, and attaches the source behind every claim, so each statement can be checked. Unsupported lines are flagged or removed before they reach the final draft. Perch runs across three surfaces that share one account, memory, and sourced answers: Perch AI Web (browser chat), Perch Terminal Desktop (macOS and Windows, with local folder access), and the Perch CLI (agentic work from any directory). ## What makes Perch different - Every claim points back to the material Perch actually read; citations are verified against the source rather than generated. - Perch works inside the apps you are already signed into (Google Docs, Gmail, Calendar, Outlook, Word, OneDrive, Slack, Notion) and leaves a receipt for every action. - Two working styles share one workspace: Saffron (operator, for folder work that needs judgment) and Quill (writer, for structured, sourced drafts). - Flock mode plans a custom team of agents for a hard task, runs them in parallel, and has checker agents review the work before it answers. - Starter tier is free with 20,000 PT per month. No credit card, no waitlist; pick a plan on sign-in. ## Core pages - [Home](https://www.perchai.app): Product overview and the case for verifiable, citable AI. - [Pricing](https://www.perchai.app/pricing): Starter, Pro, and Enterprise plans with Roost tiers up to 1M context. Bring-your-own-key is available on every plan. - [Demo](https://www.perchai.app/demo): Screen recordings of Perch working a task end to end, with a written walkthrough of each step. The current one audits a folder of 100 invoices, purchase orders, payment logs, and vendor records from a single instruction, then reports 93 findings ranked by risk with the IDs and dollar amounts behind each. - [Perch CLI](https://www.perchai.app/cli): Install with `npm install -g perchai-cli`, sign in once, and work from any folder. - [Perch Terminal Desktop](https://www.perchai.app/desktop): Download for macOS and Windows with local folder access. - [Financial Forensic Intelligence](https://www.perchai.app/financial-forensics): Deterministic checks on invoices, payments, and ledgers — AP audit, duplicate detection, three-way match, and reconciliation. - [Writing Studio](https://www.perchai.app/writing-studio): Drafts that cite their sources, verified claim by claim. - [Flock](https://www.perchai.app/flock): Multi-agent mode that plans a custom team of agents for a hard task, runs them together, and has them check each other's work before it answers. - [Docs](https://www.perchai.app/docs): Setup, usage, and reference documentation. - [MCP servers](https://www.perchai.app/docs/concepts/mcp-servers): How Perch connects to external data sources through the Model Context Protocol, with CourtListener and Playwright included by default. - [About](https://www.perchai.app/about): Why Perch exists and what verifiable AI means. ## Research - [Research index](https://www.perchai.app/research): Perch's writing on verifiable AI, citation verification, accounts-payable audit, and model evaluation. - [Why legal AI hallucinates citations](https://www.perchai.app/research/why-legal-ai-hallucinates-citations): Why legal AI tools like Harvey and CoCounsel invent case citations, why a generator cannot check its own citations, and what real verification requires. For anyone choosing a tool to catch fake or hallucinated citations in a brief. - [Accounts payable and reconciliation software, compared](https://www.perchai.app/research/accounts-payable-software-compared): A buyer's guide to AP automation, audit, and reconciliation tools for mid-size finance teams — where duplicate-payment detection, three-way match, and deterministic checking fit. - [AI for accounts payable](https://www.perchai.app/research/ai-for-accounts-payable): Why general assistants like Copilot and Cowork fall short on invoice and audit work, and what deterministic checking does differently for duplicate detection and AP audit. - [The best AI models for coding](https://www.perchai.app/research/best-ai-models-for-coding): Nine frontier and open-weight models scored on one real build task, with the leaderboard. - [How we evaluate models for verifiable work](https://www.perchai.app/research/evaluating-models-for-verifiable-work): Perch's evaluation harness of realistic professional tasks with planted issues, scored on grounding, tool use, and honesty. - [Fine-tuning open models for financial analysis](https://www.perchai.app/research/fine-tuning-open-models-for-finance): Parameter-efficient fine-tuning and domain-adaptive pretraining on open base models for financial work. - [Why most AI agents fail in production](https://www.perchai.app/research/why-ai-agents-fail-in-production): The technical failure modes behind high AI-agent abandonment rates, and what reliable systems do differently. ## For teams - [Team inquiries](https://www.perchai.app/access): Business pilots, partnerships, and larger deployments. ## Legal - [Privacy Policy](https://www.perchai.app/privacy) - [Terms of Service](https://www.perchai.app/terms)