Give an agent persistent memory
Tramya indexes the histories your agent already writes (~/.claude, ~/.codex, ~/.gemini) and hands the context back via MCP at every session.
Get started
Two minutes to give your agents memory.
Tramya is local-first: the engine runs on your machine, indexes what your agents already write to disk, and passes context from one to another via MCP. Tell us who you are and what you want to do — the rest fits in four steps.
Axis 1
Who are you?
The starting point isn't the same depending on your use. Pick the profile that fits you.
Axis 2
What do you want to do?
Every task has a short path. Choose the outcome you're after.
Store a secret used by an agent
The agent requests the use of a credential, you approve locally, it receives the result without ever seeing the value.
Resume a session
Find where an agent left off: decisions made, files touched, next actions — without re-explaining the repo.
Share team context
An end-to-end encrypted synced vault broadcasts the decisions meant for the team, without exposing your local histories.
Install
Choose your platform.
The same local memory, three ways to install it. The command updates with the tab.
Native app, signed and notarized, automatic updates via Sparkle.
curl -L -o Tramya.dmg https://tramya.com/releases/tramya-1.1.134-macos.dmg macOS 12 or newer — Apple Silicon and Intel chips.
Quickstart
Four steps, a verifiable success.
At the end, one command returns the context you just saved — the proof that memory works.
Install and launch
Install via the tab above. On first launch, the local engine starts on port 4317 and scans your agents' known directories.
Connect an agent (MCP)
Add Tramya as an MCP server to Claude Code, Codex, Cursor or Gemini CLI. Follow your agent's integration guide.
Save a decision
In an agent session, ask it to record a lasting decision via the save_memory tool (e.g. "choice: Postgres over SQLite for production").
Verify — the proof
Open a new session and ask the agent to call get_project_context. It should return the decision you just saved, indexed locally. If the context comes back, your memory is up and running.
get_project_context({ project: "my-repo", query: "database decision" })
→ returns "choice: Postgres over SQLite for production"