Persistent Memory
Hybrid BM25 and vector search, fused with Reciprocal Rank Fusion, recalls decisions, bug fixes, and project context across every session — auto-curated, no manual tagging.
Persistent · Multi-Agent · Always On
Your AI employee on the desktop — persistent memory of your codebase, skills that compound with every task, and up to nine agents shipping in parallel while you're away. Bring any model.
An engineer who shows up on day one with no context ignores your architecture — not out of malice, but because nobody told them the rules. Most AI coding tools put your agent in exactly that position, every single session. They don't know your patterns. They don't remember yesterday's decision. They start from zero, every time you open a new chat.
Ptah onboards its agents the way you'd onboard an engineer: it studies the codebase before the first message, keeps what it learns, and gets better the longer it works with you.
$ new-agent-session --tool cursor
> no context loaded. starting cold.
$ ptah session --resume
> loaded 1,425 memories.
> continuing from yesterday's decision: "memory format → JSONB"
No slideware — this is Ptah running against real projects: memory recall, sub-agent fan-out, and scheduled runs, captured as they happened.
Ptah desktop — memory recall, sub-agent fan-out, scheduled runs.
Most agents start cold. Ptah indexes your project before the first message and keeps what it learns after the last one.
Hybrid BM25 and vector search, fused with Reciprocal Rank Fusion, recalls decisions, bug fixes, and project context across every session — auto-curated, no manual tagging.
Structural AST parsing across JavaScript, TypeScript, Python, and Go indexes every function, class, and import with exact file positions — not regex guesses.
Ask 'where do we validate auth tokens' in plain English and get ranked, cited results injected straight into agent context.
Hybrid BM25 + vector memory search, fused with Reciprocal Rank Fusion.
Repeat a workflow and Ptah turns it into a reusable skill. Then it can run that skill nine times over, in parallel.
When a workflow repeats successfully, Ptah extracts the trajectory, judges its quality, and promotes it to a permanent, shareable SKILL.md file — no manual authoring required.
A main agent fans work out to parallel sub-agents across a three-tier hierarchy, each with its own provider, model, and context window.
Run up to nine concurrent agent sessions in one drag-and-resize grid. Background agents keep working while you focus on a single tile.
Ship with pre-built skills for common stacks and browse more from the skills registry — install with one click.
Up to 9 concurrent agent tiles in one gridstack view, each with an independent provider and model.
Schedule agents like cron jobs. Approve their work from your phone.
SQLite-backed, slot-claimed scheduled runs. Nightly code reviews, Sunday dependency scans, daily standup summaries — no server to babysit.
Trigger and approve agent work from Telegram, Discord, or Slack, including voice input. Discord supports per-thread multi-session conversations, so each thread keeps its own agent context.
Review and approve tool calls before they execute, from any connected gateway — nothing runs unattended that you haven't authorized.
Trigger and approve agent runs from Telegram, Discord, or Slack — including per-thread sessions on Discord.
Claude, GitHub Copilot, OpenAI Codex, 200+ OpenRouter models, local Ollama, Kimi K2, and GLM — switch providers mid-session without losing context.
Secure per-provider API key management. Real-time cost and token tracking per session.
Cursor and Copilot are excellent at finishing your sentence. They don't remember yesterday's decision, they don't work while you're away from the keyboard, and they don't take a message from your phone. That's a different job — the one Ptah does.
Persistence
No cross-session memory — each chat starts cold.
Multi-agent
One inline suggestion stream per editor tab.
Schedulability
Runs only while your editor is open and you're typing.
Reachability
Desktop editor only.
Persistence
Hybrid BM25 + vector memory, auto-curated, recalled across every session.
Multi-agent
Up to 9 concurrent agents in one grid, each with independent provider, model, and context.
Schedulability
SQLite-backed cron scheduler runs agents unattended on any cron expression.
Reachability
Telegram, Discord, and Slack — approve or trigger work from your phone, including voice.
Cursor and Copilot remain the better choice if all you want is inline completion inside an editor you already love — Ptah doesn't compete on autocomplete latency inside a text buffer, and it doesn't pretend to. Raw CLI agents remain the right call for a single scripted task in CI. Ptah is for the fourth axis nobody else covers: an agent that persists, works in parallel, runs unattended, and is reachable outside the IDE. That's a different job description — "employee," not "autocomplete."
Ptah also ships as a VS Code extension and a headless CLI for CI/CD and scripted workflows. Same license, same seven providers — without the desktop-only Memory, Skills, Cron, and Gateway suite, which requires the desktop app.
100 days free. No credit card. Windows, macOS, and Linux.