Share context between your AI agents.

Claude Code keeps its history in ~/.claude, Cursor keeps its own per workspace, Codex somewhere else again. None of them read each other. So the moment you move a task from one agent to another — because one is better at refactors, another at long CLI runs — the receiving agent starts blind. It re-reads the repo, re-asks the conventions, and re-proposes ideas the previous agent already rejected.

The usual workaround is to paste a paragraph of context into the new tool, by hand, every time. It is slow, it is lossy, and it forgets the decisions that never made it into a message. Context isn't shared between agents by default — it evaporates at the boundary.

Tramya indexes the local history of every agent you use and, when you switch, generates a context capsule: the goal, the durable decisions (with the alternatives that were rejected), the files touched, the current Git state, the tests, and the next verifiable steps. The receiving agent reads that capsule over MCP through prepare_handoff and get_project_context, and resumes exactly where the previous one stopped — in a different terminal, a different worktree, or a different tool.

Because the capsule is structured rather than a raw dump, the new agent gets signal, not noise: it inherits the “why”, not ten thousand tokens of scrollback.

Search across tools

Ask “what did we decide about auth?” and Tramya surfaces the relevant Claude Code discussion and the Codex task and the saved note — each with its provenance, so you know which agent and which session it came from.

Scoped by repo & worktree

Context is indexed by repo, branch and Git worktree, so a handoff lands in the right place instead of mixing two features or two branches.

Decisions that persist

An agent calls save_memory on a structuring choice; the next agent — whichever tool it is — finds it again. The shared memory outlives any single session.

Sharing context between agents on one machine is local by default — nothing leaves your disk. When a team needs the same project context, Tramya's optional encrypted vault syncs the decisions, conventions and structure of a repo between teammates, so anyone who joins the codebase starts with the architecture the team already settled. Prompts and raw conversations stay private; only the durable project context is shared, and the vault is encrypted device-side (AES-256-GCM) before it is uploaded.

The result: a new teammate — or the same person on a second machine — opens their agent of choice and inherits the shared thread, without a single manual copy-paste.