AgentAnywhere Blog
Notes on sovereign AI, training-data provenance and model governance, from the team building AgentAnywhere in India. Numbers here are measured, and each one says what it covers.
One disputed card transaction followed through the AgentAnywhere platform at a bank: what agents do, what policy decides, who approves and the audit trail.
AgentAnywhere Research12 min read
One service call followed through the AgentAnywhere platform at a contact centre: live assist, a refund a person approves, wrap-up, review and the audit.
AgentAnywhere Research13 min read
One tasking request followed through the AgentAnywhere platform in an operations room, for satellites, launch and drones: who proposes, who approves.
AgentAnywhere Research12 min read
What prompt injection means for an AI agent that reads customer text and calls tools, what a guardrail can and cannot promise, and what to ask a vendor.
AgentAnywhere Research16 min read
An AI guardrail that blocks ordinary customers is a service and fairness problem. What it costs, what to measure, and how to pilot on your own traffic.
AgentAnywhere Research14 min read
What RBI, SEBI, IRDAI and the card standard expect when a bank or insurer sends customer data to an AI model, and what masking before the model changes.
AgentAnywhere Research18 min read
How an Indian government department can use AI on grievances and records while citizens' identifiers stay inside its perimeter, and which rules apply.
AgentAnywhere Research19 min read
How a contact centre or BPO can use AI on transcripts and live assist while card and ID numbers stay out of the model, for Indian and EU clients.
AgentAnywhere Research18 min read
Step by step: clean nine public-domain story books with Shuddhi, train a 2.6M-parameter model from scratch on a CPU, and seal what it read in a manifest.
AgentAnywhere Research15 min read
Tatva Edge is an on-device AI model that turns a plain-language instruction into exactly one command a machine understands, or refuses. Offline, in India's languages, with a person in command. Talk to us about your machine.
Siddhartha Chandurkar5 min read
Six questions that separate AI you are renting from AI you control: where the data runs, which model decided, what it saw, what stops harm, who approved, and what you will show the regulator.
Siddhartha Chandurkar7 min read
India's DPDP Rules apply to every AI agent that touches personal data. What the Rules require, where agents leak data, and ten fixes to make before the main obligations take effect in May 2027.
AgentAnywhere Research11 min read
The RBI's FREE-AI framework explained: the seven sutras, six pillars and key recommendations, the 2026 draft model-risk guidance that builds on it, and a practical checklist for banks and NBFCs.
AgentAnywhere Research10 min read
Hosting a model in India does not make it sovereign. Five common ways sovereign AI is overstated, and the one-line check that exposes each, from a founder who builds it.
Siddhartha Chandurkar7 min read
India doesn't speak one language per sentence. Why voice AI breaks on Hinglish and code-mixed speech, what building for it properly looks like, and how to test a vendor on it.
Siddhartha Chandurkar7 min read
Ten practical AI agent use cases for Indian banks, NBFCs and insurers, from KYC exceptions to complaints and fraud triage, tiered by risk, with the human decision and the controls each one needs.
AgentAnywhere Research9 min read
Practical AI agent use cases for defence and national security in India, from maintenance and logistics to intelligence summarisation and cyber triage, run air-gapped, with a human deciding and a red line on targeting.
AgentAnywhere Research9 min read
Where AI agents help across a space programme, from requirements and design reviews through test campaigns and launch to mission operations, with the flight software untouched and every step on record.
AgentAnywhere Research8 min read
Multi-agent AI fails in the gaps between agents. How AgentAnywhere Orchestrator runs agents as services with contracts, typed handoffs, retries, human approval steps and a signed record of every run.
AgentAnywhere Research8 min read
Regulators are asking which AI systems you run, who owns them and what changed. How a registry gives every agent, model, prompt, tool and policy an owner, a version history, approvals and rollback.
AgentAnywhere Research7 min read
Tatva is AgentAnywhere's on-device model family, built from scratch in India. Tatva Edge turns plain language into one exact command, or none, on satellites, drones, robots, vehicles, wearables and defence systems, with no network.
AgentAnywhere Research8 min read
Tatva Edge lets satellite operators type or say what they want, in English or an Indian language, and get exactly one valid telecommand, checked against flight rules, approved by a named person and queued for the pass.
AgentAnywhere Research8 min read
Tatva Edge gives launch and test-stand teams a plain-language console for telemetry, procedures and checkout, offline and inside the range, while the flight computer and flight termination stay model-free by design.
AgentAnywhere Research7 min read
Tatva Edge runs on a drone's companion computer with no network, turning an operator's plain-language instruction into one exact mission command, with geofences enforced outside the model and a named human approving.
AgentAnywhere Research8 min read
Tatva Edge lets operators give robots and machines task-level instructions in Hindi, Hinglish or English, on the device and offline, with limits enforced outside the model and the certified safety controller untouched.
AgentAnywhere Research8 min read
Tatva Edge gives humanoid and mobile-manipulator robots plain-language task commands in English and Hinglish, on the robot and offline, with people-safety rules outside the model and a human approving.
AgentAnywhere Research8 min read
Tatva Edge is an offline, in-cabin voice command model for cars, two-wheelers and commercial vehicles: Indian languages and Hinglish, on the vehicle, with no cloud. Comfort, navigation and media only, never driving.
AgentAnywhere Research7 min read
Tatva Edge is a small, offline command model for smartwatches, earbuds, fitness bands and rugged field wearables: Indian-language voice commands understood on the device, with health data kept on the wrist.
AgentAnywhere Research7 min read
Tatva Edge brings plain-language commanding to sensors, drones and logistics at forward posts and air-gapped sites: built from scratch in India, runs with no link out, and cannot target or release a weapon.
AgentAnywhere Research8 min read
The AgentAnywhere platform lets banks, insurers, contact centres, defence and space programmes build, run and govern AI agents on one sovereign runtime — inside their own perimeter — with a named human approving consequential actions and a signed Trust Receipt on every call.
AgentAnywhere Research10 min read
AgentAnywhere Aavaran classifies PII, financial and regulated data on every call and masks it — by reversible tokenization or irreversible redaction — before any model, tool or embedding sees the raw value, with every mask logged as a signed Trust Receipt. Built for contact centres, PSU banks, insurers, defence and space.
AgentAnywhere Research9 min read
AgentAnywhere Model Hub is a governance-aware model registry: every model carries provenance, evaluation history, policy bindings and a signed audit trail, so a bank, a defence programme or a contact centre can answer which model produced an output, what its limits were and who approved it.
AgentAnywhere Research9 min read
Shuddhi is an Apache-2.0 data factory that cleans LLM training corpora and seals them with a verifiable build hash — including your EU AI Act Article 53(1)(d) summary.
AgentAnywhere Research8 min read
Journalists & analysts
Want a briefing on anything here, on or off the record, with the engineers who built it?
