Product
Company memory for agents and developers.
Anuvrit keeps your team's fixes, decisions and lessons with their sources, and hands them to coding agents and developers before work starts. Anuvrit doesn't write code itself. It gives your agents your team's history, runs them on the right model, and keeps work moving when a tool hits its limit.
Your agents write the code. Anuvrit equips them.
Anuvrit is not a coding agent and does not write code itself. You connect it over MCP to the agents you already use, such as Claude Code, Codex and Cursor. Memory gives them your team's history, and the Specialised Router runs them on the right model.
Memory
Before an agent acts, it receives the team's past decisions, fixes and reviewed lessons, each with its source. Afterwards it records the outcome.
Writeback
Accepted decisions and outcomes are saved back with a receipt, so the next task can find them.
Specialised Router
Plans each task, runs every step on the right model through your own tools, checks it, and hands the work to another tool when one hits its limit. Every task ends with a cost line.
One memory for agents and people.
Agents read it through MCP tools. Developers ask questions in plain language. Both read the same sources under the same access checks.
For agents
Claude Code, Codex and Cursor can call engineering_preflight before substantive work and knowledge_decision to record what was accepted.
For developers
Ask why something was decided or whether it was fixed before. The answer cites the pull request, task or document behind it.
How an answer is built.
Sources are captured with their versions and indexed. Each question is authorised first, then answered from the evidence it is allowed to read. Citations are checked before the answer is shown, and those that fail are rejected.
Diagram: sources are captured, versioned, extracted, normalised, chunked, embedded and indexed. A query is authorised, candidates are retrieved and packed into bounded context, an answer is drafted, citations are validated and support is verified, and only then is the answer returned with citations. Citations that fail validation are rejected.
Go deeper. Each part in detail.
Each page explains one part of Anuvrit and how it works.
- How it worksCapture, the versioned archive, search and cited answers.
- Engineering memoryPreflight, writeback receipts, trust records and reuse ranking.
- TimelineDecision, project and person histories over time.
- ConnectorsCode, docs, tickets, chat, files and voice, plus an API and webhooks.
- TeamsPersonal and team workspaces, roles and sharing levels.
- Specialised RouterPlans, runs and hands off work across your coding tools, with the right model for each step.
- DevelopersMCP setup for Claude Code, Codex and Cursor.
- SecurityAccess checks, isolation and data controls.
Give your agents your team’s history.
Anuvrit is in beta and opening to engineering teams in stages. Join the waitlist and we will contact you when there is a place for your team.