Rapid context enrichment for artificial intelligence (AI) agents
Give your agents the right context, instantly.
Index your repository and your team's knowledge once, then enrich every prompt with exactly the right context, within a token budget, over a simple application programming interface (API) and the Model Context Protocol (MCP). Local-first, and free to start.
Open source core · Local-first · Works with any agent
Local-first
Indexing, memory and retrieval run on your machine. Your data stays yours.
Open-source core
The engine and its APIs are free and open, under the Massachusetts Institute of Technology (MIT) licence.
Any agent, over MCP
Serve budgeted context to any agent over the Model Context Protocol and the web.
Works with your tools
Claude Code
Cursor
The Model Context Protocol (MCP)
The command line
Your editor
Any agent over Hypertext Transfer Protocol (HTTP) or WebSocket
The problem
Agents start every session from nothing.
Two things leak away the knowledge that should make each session smarter than the last.
Agents forget
A coding agent only sees what fits in its context window, and loses its reasoning the moment that window scrolls. Next session, it relearns your codebase from nothing.
Knowledge is scattered
Your team's decisions, conventions and gotchas sit spread across chat histories you do not control — effectively rented, never compounding into something you own.
How it works
Index once. Enrich every prompt.
Trove does the retrieval, so an agent gets exactly what a task needs — no more, no less.
-
01
Index once
Point Trove at a repository. It parses the code into layered memory — symbols, modules, subsystems and architecture — and captures the decisions and conventions your team records.
-
02
Enrich every prompt
For any task, Trove retrieves the context that matters, widest layer first, and stops the moment it has enough — all inside a token budget you set.
-
03
Keep it current
As files change and your team learns, the index and memory stay in sync, so every agent starts from what you already know.
Context, assembled to a budget
A query comes in. The right context goes out.
Trove walks its layers for the task, ranks what fits, and assembles a budgeted context block — with the provenance of every piece.
Query
how do we verify a vendor before payout?
Context budget1,180 of 2,000 tokens
Context card
verify_documents()
Checks a vendor's uploaded documents against the onboarding policy before a payout is allowed. Writes a vendor_events row on success.
- Module
- onboarding.rs
- Links
- decision: one writer
- Retrieved
- widest layer first
Illustrative interface with example data. The query, nodes, numbers and activity feed are made up for the demo.
Two layers, one system
An engine that maps your code, and a memory that keeps what you learn.
The engine answers “what is in this codebase, and where?” The memory answers “what did we decide, and why?” Together they give an agent both structural knowledge and durable memory, as plain files you can read, review and commit.
The engine
Parses your repository and writes a layered map of it: a symbol index, module, subsystem and architecture summaries, and a dependency graph. Retrieval walks from the widest layer down and stops as soon as it has enough.
- Symbols, modules, subsystems, architecture
- Dependency graph across the codebase
- Progressive retrieval inside a token budget
Lives in .trove/ in any repository
The memory
Durable entries for the decisions, gotchas and conventions your team learns, each with the reasoning behind it. They reload at the start of every session, so an agent begins already aware of what you have figured out.
- Decisions and their rationale
- Gotchas and hard-won conventions
- Reloaded every session, shared with the team
Lives in ~/.claude/trove or ./.claude/trove
Open source, free
Start with the open-source core.
The Trove engine and its context-enrichment APIs are released under the Massachusetts Institute of Technology (MIT) licence. It is plain files in your repository, runs locally and works offline. No account needed. Build the engine, or install the Claude Code plugin in a minute.
- Version
- v0.2, v0.3 in progress
- Licence
- MIT licence, free forever
- Engine
- Rust, built with Cargo
- Storage
- Plain Markdown and JavaScript Object Notation (JSON)
- Interface
- Command-line interface (CLI), HTTP, WebSocket, MCP
- Network
- Runs locally, works offline
Requires Rust. From any repository:
git clone https://github.com/anishfyi/trove
cd trove && cargo build -p trove-cli
# Index the current repo (writes .trove/, gitignored)
cargo run -p trove-cli -- index
# Enrich a task, widest layer first, to a budget
cargo run -p trove-cli -- query "modify vendor onboarding"Other commands include status, ask, remember and tools. Add --json for machine-readable output.
Run these one at a time inside Claude Code.
- Add the marketplace
/plugin marketplace add anishfyi/trove - Install the plugin
/plugin install trove@velofy-trove - Create and map your trove
/trove:init && /trove:index
Then /trove:remember to capture, /trove:recall to search with citations and /trove:query for layered context.
Trove Business
Safe, secure knowledge deployment for teams.
Deploy Trove self-hosted or in your private cloud, with the controls an enterprise needs. Your knowledge is governed and auditable, stays yours, and never trains anyone else's model.
Self-hosted or private cloud
Run Trove inside your own infrastructure or private cloud. Your code and knowledge never leave your perimeter.
Access controls & single sign-on (SSO)
Role-based access and single sign-on, so the right people reach the right knowledge and no one else does.
Audit log
Every query and change is recorded, so you can answer who saw what, and when, for review and compliance.
Shared team memory
One governed library your whole team and all your agents pull from, instead of knowledge trapped in private chats.
Priority support & onboarding
We set up your deployment with you and stay on call, with a direct line for your team.
Managed app & sync Coming soon
A managed macOS app and hosted sync across machines, so your memory follows your team everywhere.
Pricing
Start free. Deploy securely when you're ready.
Two tiers, nothing hidden. Build on the open-source core for free, forever, and move to Trove Business when your team needs governed, secure deployment.
Trove
$0Free forever
For individuals and developers.
- Open-source engine & context-enrichment APIs (MIT)
- Local-first and offline
- MCP and command-line interface (CLI)
- Progressive, budgeted retrieval
Trove BusinessRecommended
CustomPer deployment
For teams and enterprises.
- Everything in Trove, plus…
- Self-hosted or private-cloud deployment
- Access controls, SSO and audit log
- Shared team memory & priority support
Own your intelligence
Your knowledge is an asset you own — never rented, never trained on.
Most teams rent their intelligence back from a chat history they cannot see into. Trove keeps it local-first, self-hostable and yours, so every session adds to something you keep.
Local-first
Indexing, memory and retrieval run on your machine and work offline. Your code and decisions never have to leave it.
Self-hostable
Run the whole service yourself, or in your private cloud. Hosted sync is an option you opt into, never a lock-in.
Yours, never trained on
Your knowledge is governed and stays yours. It is never used to train anyone else's model, and you can export it as JSON or Markdown any time.
From Velofy
Built by an applied AI and security lab.
Trove is made by Velofy, an applied artificial intelligence and security lab. It is also how Dexter, Velofy's multi-agent harness, remembers across sessions — so agents pick up a project already knowing what was decided and why.
Explore Velofy