Findings, fragments, tasks, truths, skills, and sessions across machines. Private markdown. Git-backed. No database.
~/.phren, wires MCP, installs hooks.
phren add from any repo.phren init picks it up.Phren catches findings, links fragments across projects, and figures out what matters. Gets sharper over time.
Git syncs after every response and pulls on open. Optional periodic checks keep running MCP servers updated at a configurable interval.
Terminal shell and graph, web UI, VS Code sidebar, iOS app. Browse findings, fragments, and the full knowledge graph. Search across projects.
AGENTS.md dumps everything every time. Phren searches, ranks, and surfaces only what matches. About 550 tokens instead of the whole file. Less noise, lower cost.
Switch agents, switch machines. Phren keeps up. What you figured out in Claude this morning is already there when you open Cursor tonight. Good stuff sticks. Bad stuff fades.
Claude asks a question. Phren already looked it up. The findings, the gotchas, the thing that broke last time. All there before the first token.
Terminal, browser, editor, phone. Same markdown, same findings, same tasks. Pick a way in.
Browse what phren knows. Check on tasks. Review the queue. Type : for the command palette. Deep-link a keybinding straight to a view.
Opens with a splash: the mascot beside the wordmark, revealed with a decrypt effect and a light beam.
Shell guide →
The same graph as the web viewer, drawn on a braille canvas. Projects cluster, findings take their topic color, fragments bridge projects. The layout is deterministic, so the same store always draws the same map. Press v and it becomes a sphere you can drag around; in watch mode, every recall lights its node and phren walks over to it.
Run phren web-ui for the 3D map: projects as containment fields, findings, tasks and fragments inside them, quality scores, and what needs your attention. Review, edit, merge and prune from the contents pane.
Projects, findings, tasks, fragments, skills, hooks and sessions in the activity bar, and Ctrl+Shift+K to search. Add, review and push without leaving the editor.
The graph panel, web UI and iPhone Memory map share one renderer.
Steer agents on your computers, read their changes, and browse indexed code from your phone. Remember a note about a function and send it to its session, or ask your conductor through Siri and the Action button. GitHub sync keeps findings, notes and tasks in Memory, with map and list views. Local schedule and approval notifications need no relay.
Memory syncs directly with GitHub. Phren Hook connects your running agents over SSH: native chat, Herdr terminals, project context, and web previews. Your keys stay in the iPhone Keychain.
Get the app and set it up → See the apps → iPhone → Android →
The Herdr plugin binds the shell to a keybinding: tasks, findings and the review queue for whatever project the pane is sitting in, popped over your layout and gone again when you close it.
phren agent is a coding agent with phren as its memory: it opens with the project's gotchas, active tasks and past decisions loaded, writes new findings as it works, and triages the review queue with you. Run it on your ChatGPT/Codex subscription, OpenAI, OpenRouter, Anthropic or local Ollama models.
A team store is a git repo full of phren knowledge. Share it with your teammates and everyone's findings flow through the same place. Keep as many as you like, one per team, company, or client.
Any empty git repo will do. Run phren store add and point it at the remote. Phren fills in the structure.
They run the same phren store add against the repo URL. That's it. Phren pulls on open and pushes after each response.
No cap on stores. One for your team, one for the company, one for a client. Search runs across every store at once.
MCP tools handle search, capture, review, and sync. Skills are shortcuts for larger jobs. Global skills live in ~/.phren/global/skills/, project skills in ~/.phren/<project>/skills/. Project skills override global ones with the same name.
~/.phren/global/skills/~/.phren/<project>/skills/MCP and hooks. Same knowledge, same findings, no matter which tool you're in.
The store is the point. These are three more places it shows up.
Your findings, tasks and review queue synced from GitHub, with Siri capture and widgets. Chat with the agents on your computers, open their terminals, read their diffs.
Phren Hook runs on each computer beside Herdr. Steer Codex, Claude Code, Copilot and OpenCode sessions, answer their questions, and read their code and changes from the phone over your own SSH.
A coding agent that starts every session already knowing the project's memory. Install @phren/agent and run phren agent.
No API calls. No third party.
Findings tagged with sessions, linked to tasks, surfaced before every prompt.