From the State Government edition of November 2, 2026
Maharashtra signs an enterprise agreement with Sarvam, putting its Indus Cowork AI workspace live for 2,500 state officials with in-country hosting and a no-training commitment
Government of Maharashtra, Commissionerate of Electronics, Information Technology and Artificial Intelligence; Sarvam · State government internal administration: drafting, resolution search, correspondence and tender evaluation · Maharashtra, India
- Publisher
- Elets eGov
- Original publication
- 2026-11-01T18:17:02Z (Elets eGov)
- Source retrieved
- 2026-11-03
- Event date
- 2026-10-28
What happened
Elets reports that Sarvam signed an Enterprise Agreement with Maharashtra's Commissionerate of Electronics, IT and AI, making its Indus Cowork platform (branded MahaINDUS) live for 2,500 officials. Officials can draft documents, summarise files, translate scheme documents in 22 languages and digitise scanned records; three purpose-built tools are GR Expert (searches Government Resolutions and flags superseded ones), an Email Summariser and an RFP Bid Evaluator. Sarvam says hosting is entirely in India on audited infrastructure; the state says official data will not leave the country or train models. Citing Analytics India Magazine, the package was approved in August at ₹11.26 crore over two years and also includes a voice/WhatsApp citizen outreach system. The Chief Minister told departments they must actively use the platforms so shortcomings surface early.
Why it matters
Why now: the archived Oct 28 launch report did not cover the enterprise agreement, the two-year ₹11.26 crore outlay, the no-training and data-residency commitments, or the bid-evaluation tool. It is a sovereign-hosted, internal-operations deployment at a defined scale, a useful comparison for U.S. states procuring enterprise GenAI (e.g., North Carolina's open RFP). Transferability is limited by Indian procurement and data law.
Evidence and measured results
Elets eGov article opened and read in full. Vendor and state statements are attributed; no usage, accuracy or time-savings data were published. Budget figure is secondhand via Analytics India Magazine.
Limitations and uncertainty
Secondary trade-press report relying on vendor and state statements; adoption and accuracy unmeasured; Elets itself says the next test is adoption and accuracy as resolutions change. Indian context limits transferability.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-02; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- States want staff GenAI with sovereign data handling but lack contract terms and adoption measures that prove value.
- Stakeholders
- State IT and AI commissionerate
- Procurement officers
- Department secretaries
- Data protection officers
- Discovery questions
- Where must state data be hosted, and is training on it barred in contract?
- Which internal tasks (rules search, correspondence, bid review) are targeted?
- How will usage and accuracy be measured over the term?
- Who reviews AI-assisted bid evaluations?
- Value hypothesis
- Interpretation: fixed-scope workspaces with residency terms can lower data risk and focus adoption; Maharashtra has not reported outcomes.
- Potential engagement
- Bounded review of an enterprise GenAI award's data terms, domain-tool scope and adoption metrics.
- Unsupported claims
- No measured time savings
- No accuracy data for GR Expert or bid evaluator
- No evidence no-training terms are audited
Pre-sales engineering
Role takeaway
- Fit
- State administrations with large, versioned rule corpora and heavy correspondence.
- Architecture and integration
- In-country hosted workspace with retrieval over government resolutions, email summarisation and speech models integrated with a state AI platform.
- Prerequisites
- Authoritative, versioned rules corpus
- Identity integration for 2,500 users
- Contract terms on residency and training
- Audit logging
- Constraints
- Indian data and procurement law
- Vendor-reported hosting claims
- 22-language output quality varies
- Security
- Confirm residency, no-training and log retention in contract and test access controls on official files.
- Proof of value
- Test GR Expert on known superseded resolutions
- Sample bid-evaluator outputs against human scoring
- Measure active users monthly
Delivery
Role takeaway
- Work
- Operate an AI workspace for 2,500 officials with domain tools over a two-year term.
- Dependencies
- Enterprise agreement
- Rules corpus kept current
- Department adoption
- Training
- Ownership
- Commissionerate of Electronics, IT and AI with Sarvam; departments own use.
- Skills and adoption
- Officials need to verify AI drafts and resolution answers and know when bid evaluation needs human judgment.
- Governance checkpoints
- Human-review rule for AI bid evaluation
- Accuracy review after rule changes
- Usage review before renewal
- Acceptance criteria
- Proposed: monthly active share of 2,500 seats reported
- Proposed: GR Expert supersession accuracy measured on a test set
- Proposed: every AI-assisted bid score logged and reviewed
- Risks
- Low adoption
- Outdated rule answers
- Unreviewed AI bid scoring
- Unverified residency claims
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
In-country hosted AI workspace with domain tools layered on top (resolution search with supersession tracking, email summarisation, bid evaluation) and speech models integrated with the state's MahaAI platform.
Governance
Who approves, reviews and stays accountable for outcomes?
An RFP Bid Evaluator touches procurement decisions; the article describes no human-review, audit or appeal rules for AI-assisted bid scoring. Interpretation: this needs explicit controls.
Security and privacy
What data, permissions and controls need testing?
Data residency and a state statement that official data is not used for model training; the article does not describe how these are contractually enforced or audited.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
22-language support including Marathi; workforce adoption explicitly flagged by the Chief Minister.
Procurement
What should contracts, pricing and exit terms secure?
Enterprise agreement; reported ₹11.26 crore two-year package including citizen outreach and speech models.
Operating model
Which teams own the service once it runs?
Central IT commissionerate runs a shared workspace; departments are expected to use it actively.
Publication history
- 2026-11-02State Government · Issue 563 resources
Stable resource ID: maharashtra-sarvam-indus-cowork-enterprise-agreement-2500-officials-data-residency-20261102