Memory you can shape.
Your agent can revisit, rewrite, and forget memories. Its knowledge graph follows the retained facts, keeping useful connections grounded in what actually belongs.
Explore memoryA RESIDENT AGENT FOR YOUR MAC
Remembers what matters. Works alongside you.
Grows into apps never made for it.
One Swift-native home for your agent’s memory, tools, and ongoing work. On your Mac, with you on iPhone.
Apple silicon · macOS 26+ · AI provider account required

MEET IT WHERE IT LIVES
Your agent keeps the thread between conversations, so you start where you left off.
Can we fit the website redesign in this week?
Recalled“I do my best thinking before the day gets busy.”
Mornings, then. You told me you think best before the day gets busy. I’ll put it first on tomorrow’s Desk.
ILLUSTRATIVE · NOT A SCREENSHOT
MEANING THAT STAYS WITH YOU
The bundled large embedding model gives semantic recall a richer representation of what you mean. Your agent can connect a differently phrased question to a relevant memory—and bring that context back into the conversation.
Search earlier conversations by date, revisit decisions in order, and keep track of who said what. The memory model runs locally on your Mac, ready from the first launch.
Explore memory and continuity“I do my best thinking before the day gets busy.”
“When should I set aside time for focused work?”
bge-large-en-v1.5 1,024-dimensional embeddings
MORE THAN THE NEXT REPLY
Give an idea somewhere to go. NativeAgent connects what you say to what it remembers, what it can do, and the work you’re already carrying.
REMEMBER
The bundled bge-large embedding model brings meaning-based recall into the conversation. Fluid Context adds a selective working set instead of loading everything you’ve ever said.
Inside the resident mind“What did we decide last time?”
CONTINUITY THAT BECOMES USEFUL
Memory, background research, and ongoing work give the next conversation something real to build on.
Your agent can revisit, rewrite, and forget memories. Its knowledge graph follows the retained facts, keeping useful connections grounded in what actually belongs.
Explore memoryStanding bots have their own brief, conversation, and model. Run them yourself, set a schedule, or wake them on supported GitHub and Slack events. Their cards show results, missed runs, and what needs attention.
Meet standing botsIts own turn is a part. If a version of it breaks mid-turn, the one before is swapped back in, the turn replays, and the reply still lands. A relaunch replays the turn from its log, and a Steer lands mid-reply.
See how it worksTHE SAME AGENT. WITHIN REACH.
Continue a conversation, watch a reply arrive live, decide an approval, change a Mac setting, or send a file from iPhone and iPad. The Mac keeps the runtime, memory, and credentials.
Telegram, Slack, and authenticated local bridges connect to that same agent through shared policies.
Explore Apple continuity
Your MacAgent · Memory · WorkMac running · Devices paired · Permissions shared
YOUR AGENT. YOUR CHOICES.
HOW FAR IT CAN REACH
Safe. Reads files. Changes nothing on your side; shell and git are refused.
Your saved entries and integration switches still bind. Apple privacy controls and each connected agent’s permissions stay separate boundaries.
Start with one connected account, then pick a model for Chat, Work, and Memory and mind if you want. Bots keep their own choices, on Mac and iPhone.
Ask it to connect a supported local agent, or let it find the ones already installed. It shows what will change, backs up the settings, and can repair a link that breaks. Agents you explicitly trust act as you; a turn steered by one you haven’t marked trusted is judged as Work mode at most.
Explore agent connectionsMemory and runtime live on your Mac; relevant context goes to the provider you choose. The context ring shows what a turn used, and errors say what ran, what failed, and what remains unknown.
Read the trust model
NATIVEAGENT 0.6.6
Signed and notarized · Signed in-app updates · macOS 26+
Release details and development notes