iGPT
iGPT is a context engine from Spikenow Ltd. that turns email threads, attachments, documents and live web data into cited, reasoning-ready answers, delivered either through a chat workspace or through a single usage-based API endpoint.
What is iGPT?
iGPT calls itself the context engine for work. Published by Spikenow Ltd., the Israeli company behind the Spike mail client, it starts from the premise that the richest and least exploited context in a company already sits in its mailboxes, and sets out to turn that material into something an AI system can genuinely reason over.
It comes in two surfaces. iGPT Chat is a workspace on web, desktop and mobile where you connect your mailbox and ask in plain language what needs attention today, what changed since yesterday, what is blocking a project, or what to prepare before the next call. It keeps a personal context memory across sessions, drafts and rewrites using your real material, generates images, and links every answer back to the message or attachment it came from.
The Context Engine API is the developer-facing half. One authenticated request to the recall endpoint carries an input, a quality tier and an optional JSON output schema, and runs retrieval, context shaping and reasoning in a single pipeline. A Python SDK, a live playground, MCP and LangChain are the documented ways in. What the vendor claims to remove is the plumbing: no parsing, chunking, vector store, indexing or prompt tuning left to maintain. Underneath sit real-time ingestion, attachment processing, hybrid retrieval scored and reranked in one call, automatic thread reconstruction, and permission-aware retrieval that never surfaces what a user cannot already read. Stated performance is roughly 200 ms to retrieve and about 3 seconds to first token over large, messy mailboxes.
Security is argued rather than assumed: SOC 2 Type II, AES-256 at rest through a FIPS 140-2 validated module, TLS 1.3 or better in transit, per-user encryption keys, zero data retention at inference and no training on customer data. Three deployment models are offered, from fully hosted cloud to a hybrid setup keeping data in the customer's VPC, to a fully private on-premises install. An August 2026 benchmark, published and replayable on the site, pits iGPT against a Gmail agent on the same underlying model across eight real-inbox questions and reports eight wins, up to 5.5 times lower cost and 1.2 times faster aggregate time.
What it does
- Answer natural-language questions from your email, attachments, documents and connected tools, citing the source of every claim
- Rebuild entire email threads across time, participants and attachments to reconstruct what was actually decided
- Return structured JSON matching a schema you supply, such as action items with owners and due dates
- Extract text, data and structure from documents, PDFs and spreadsheets attached to conversations
- Combine internal context with live web search inside a single request
- Produce a daily briefing of what changed, what is blocked and what needs attention
- Build and publish no-code agents to specific groups from a central admin console
When to use iGPT / When not to
A quick filter to help you decide if iGPT is the right fit.
When to use iGPT
- Developers building AI agents that must reason over email threads and attachments through a single API call
- Sales and revenue teams tracking deal risks, stalled threads and CRM updates buried in correspondence
- Customer success and support teams rebuilding a full customer story across long chains and attachments
- Legal, compliance and finance functions that must trace decisions, approvals and budget commitments back to the source
- Executives, founders and executive assistants who need a daily briefing and meeting preparation drawn from real work context
When not to use iGPT
- Anyone unwilling to connect a work mailbox: without connected sources, iGPT has nothing to reason over
- Users under 18, whom the privacy policy explicitly excludes from the service
- Teams that need a non-English interface, since the site declares no supported languages and is written entirely in English
- People looking for a full email client: iGPT reads and reasons, and its own menu points to Spike Email for actual mail handling
- Small groups hoping to run the Team plan on one or two seats, as it starts at a three-seat minimum
How to use iGPT
A typical end-to-end flow, from setup to results.
- Decide your entry point: the iGPT Chat workspace for everyday work, or the Context Engine API to build on top of it
- Sign in to the chat workspace, or open the developer playground to obtain an API key
- Connect your sources through OAuth, email first, then calendar, documents and other tools, much as you would connect an app like Zapier
- Let the background indexing run over recent emails, files and messages; it then keeps running so the context stays current
- Start asking questions in plain language straight away, with no upload and no prompt engineering
- Check each answer through its citations, which link back to the original message or attachment
- For API work, send an authenticated POST to the recall endpoint with an input, a user identifier, a quality tier and an output_format schema
- Read the structured JSON back, or stream it, and feed it into your own agent or workflow
- On a Team plan, use the admin console to manage members, groups, data sources and agents
- On Enterprise, build no-code agents in Agent Studio by picking data sources, capabilities and the groups to publish to
Pros & Cons
Pros
- A single API call replaces the whole retrieval stack: no parsing, chunking, vector store, indexing or prompt tuning left to maintain
- Every answer is cited and traceable back to the message or attachment it came from
- Permissions are inherited from the source systems and enforced end to end, so answers never exceed a user's own access
- Unusually explicit privacy posture: no training on customer data, zero data retention at inference, per-user encryption keys
- Serious security credentials for a product this young, including SOC 2 Type II, AES-256 at rest via a FIPS 140-2 validated module and TLS 1.3 or better
- Three deployment models, up to a fully private on-premises install for regulated or air-gapped environments
- Public, itemised pricing on both sides, with a permanent free plan on the chat, per-million-token rates on the API and 10 USD of free credits
Cons
- The value depends entirely on connecting your mailbox, a high bar of trust to clear before you can judge the product
- No named subprocessor list is published; the privacy policy lists categories of vendors only
- No Article 27 EU representative and no data protection officer are named, although the publisher sits outside the European Union
- Governing law is England and Wales with exclusive jurisdiction in London, a remote forum for most users
- The documentation and the detailed pricing hub render in JavaScript and are unreadable without a browser
- No interface language is declared anywhere, and the Android app announced in the pricing table has no findable Play Store listing
- The benchmark carrying much of the marketing argument was designed, run and published by the vendor itself
Pricing & Plans
A permanent free plan is available at no cost. The cheapest paid tier is the Pro plan at USD 20 per month on an annual subscription, or USD 25 if billed monthly. The Context Engine API is charged on usage, starting at USD 1.10 per million input tokens.
- CEF-Normal context reasoning
- one connected email datasource
- real-time web search with reasoning
- and the ability to ask
- draft
- brainstorm and analyse
- subject to usage limits
- everything in Free plus the CEF-High engine
- the full datasource stack
- higher usage limits
- faster responses at peak times and priority access to new capabilities
- everything in Pro plus role-aware assistants and agents
- enterprise search across the organisation
- an admin console with central billing
- role-based permissions
- cloud
- hybrid or fully private deployment
- SOC 2 Type II and full data encryption
- organisation-wide deployment
- fine-grained permissions
- RBAC and audit controls
- a unified workspace across email
- documents and tools
- bespoke security and compliance review
- and no-code agents built on company knowledge
- USD 1.10 per million input tokens and USD 6.50 per million output tokens
- USD 3.50 per million input tokens and USD 21 per million output tokens
- USD 7 per million input tokens and USD 42 per million output tokens
Data, GDPR & hosting
A consolidated view of how iGPT handles your data.
GDPR overview
The security page states plainly that customer data is maintained and secured in accordance with the GDPR, and the privacy policy is built around it: a purpose-by-purpose table of legal bases for EEA users, a clear split between controller duties (developer accounts, billing, authentication and security logs) and processor duties (end-user data handled on the customer's instructions under a data processing addendum), and the full list of EEA rights, from access, copy, rectification, restriction and withdrawal of consent to portability, erasure, freedom from solely automated decisions and complaint to a national authority. Transfers out of the EEA rely on an adequacy decision or on standard contractual clauses, and rights are exercised at support@spikenow.com. Two gaps remain: no Article 27 EU representative is named and no data protection officer is designated, although the publisher is established in Israel. The policy was last modified in December 2025.
Who owns the data?
Under the Terms, users keep ownership of their User Data; Spikenow Ltd. acquires no ownership and receives only the limited licence needed to run the service. Users also retain ownership of the output iGPT generates, and the company assigns them any right it might hold in it, although outputs are not unique and other users may receive similar ones. Data may be shared with trusted third-party providers, including AI service providers, as needed to deliver the product. Legally, Spikenow acts as controller for developer accounts, billing, authentication and security logs, and as processor for end-user data handled on a customer's instructions under a data processing addendum.
Reuse rights
Users may reuse their own data and the outputs iGPT generates without asking permission: the Terms assign them whatever interest the company might hold in that output. The freedom stops at four explicit limits. Outputs may not be used to develop, train or improve AI models or services that compete with iGPT. They may not drive decisions carrying legal, financial, medical or material consequences for a person, including credit, education, employment, housing, insurance, legal matters or medical treatment, without human review and independent verification. AI-generated content may not be passed off as human-written. And outputs may not be extracted programmatically through scrapers or bots outside the authorised API. Separately, any feedback sent voluntarily to the company grants it a perpetual, worldwide, royalty-free licence.
Data retention & training
Hosting summary
For the hosted service, iGPT states that data sits in US and EU regions on SOC 2 certified infrastructure. Two alternatives cover stricter requirements: a hybrid model in which data stays inside the customer's own VPC while inference runs in the cloud, and a fully private deployment where the whole platform runs inside the customer's infrastructure, on premises or in a VPC. Encryption is documented in some detail: AES-256 at rest through a FIPS 140-2 validated cryptographic module, TLS 1.3 or better in transit, and an individual encryption key per user rather than tenant-level isolation alone. Each inference is processed in memory, with inputs, prompts and outputs never stored. The publisher is established in Israel and anticipates international transfers, out of the EEA under an adequacy decision or standard contractual clauses and out of other jurisdictions, including Israel, under equivalent mechanisms. What is left unsaid matters too: no cloud provider, data centre or subprocessor is named anywhere on the site.
Things to keep in mind
Risks and trade-offs to weigh before adopting iGPT.
- Connecting a work mailbox hands over the whole history of your correspondence at once: the exposure is maximal from the first OAuth click, before you know whether the tool is worth it
- Answers are cited but not guaranteed. The Terms disclaim any warranty that content is accurate, reliable or correct, and citations can make a wrong answer look better sourced than it is
- The publisher's own Terms forbid using outputs for decisions with legal, financial, medical or material impact on a person without human review, a risk it names because it is real
- A generated daily briefing quietly replaces reading the sources; over months, the habit of forming your own read of a thread can atrophy
- The Terms warn that some processed data may be stored only temporarily and deleted after a set period, with no liability for the loss and no published retention duration
- In enterprise deployments, audit logging can be switched on, so individual privacy in the chat depends on an administrator's setting rather than on the product's design
- Data may be shared with third-party providers, including AI vendors, none of whom are named, and transfers to countries with weaker protection are accepted by using the service
Setup & Integrations
Technical difficulty
Three very different levels. Using iGPT Chat requires no technical skill: sign in, authorise your sources through OAuth, let the background indexing run and start asking, in a process the publisher compares to connecting an app like Zapier. The API needs a developer, but only briefly: an API key, one authenticated request, an optional JSON output schema, with a Python SDK and a live playground to test against. An enterprise rollout is a genuine infrastructure project, involving hybrid or on-premises deployment, an admin console, RBAC and a security review. In all three cases, no retrieval stack has to be maintained.
Deployment
Apps stores
Integrations
Behind iGPT
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement iGPT.
Frequently asked questions
What exactly does iGPT do?
How is this different from enterprise search or chatting over documents?
Is my data used to train models?
How are permissions handled?
What does it cost?
Can it be deployed inside our own infrastructure?
Do I need developers to use it?
Is there a mobile app?
Who is behind iGPT, and where is the data hosted?
Is there an age limit?
Should you pick iGPT?
iGPT is a narrow product argued well. Rather than adding another general assistant, Spikenow Ltd. bets that the useful and under-exploited context of a company already sits in its mailboxes, then builds the retrieval, thread reconstruction and attachment parsing needed to make it usable. The technical case fits in one sentence: a single API call replaces parsing, chunking, vector stores, indexing and prompt tuning, and every answer comes back cited.
The security posture is more explicit than most products this young bother with, from SOC 2 Type II and AES-256 at rest to per-user encryption keys, no training on customer data, zero retention at inference and three deployment models running all the way to a private on-premises install. Pricing is public and legible on both sides, with a genuinely usable free plan and per-million-token rates on the API.
Two things deserve caution. The benchmark carrying much of the marketing weight was designed, run and published by the vendor; it is replayable on the site, which is more than most offer, but it is not independent. And European buyers will find gaps in the paperwork: no named Article 27 representative, no designated data protection officer, no named subprocessor list, a publisher established in Israel and exclusive jurisdiction in London.
Behind the product sits a company that has been shipping mail software for over a decade and raised around USD 13 million for it, which is reassuring on the dimension that matters most here: knowing what email data actually looks like at scale. The entry price is not money but trust, since you have to connect your mailbox before you can judge anything. For teams willing to do that, the free plan makes the evaluation cheap and the enterprise options keep the eventual answer negotiable.
- Choosing a selection results in a full page refresh.
- Opens in a new window.