A learning agent that keeps a dossier on you — and a wiki

What it does

Inno Agent is a single-user learning companion built atop the unmodified Pi coding-agent SDK. It wraps an LLM loop in three explicit memory layers — a learner profile, a native wiki knowledge base, and cross-session recall — then adds a cron scheduler, personal chat channels, and a browser-based practice terminal. The whole thing runs in one serialized process aimed at pedagogy, not open-ended software engineering.

The interesting bit

The architecture deliberately rejects scale: one process, one queue, and if a session is busy the agent returns a 409 session_busy instead of silently queueing. That constraint keeps the memory layers inspectable and the learner model editable by the human, with unevidenced labels explicitly forbidden.

Key highlights

• Three-layer memory: an L1 learner profile, an L2 wiki with hybrid BM25/graph retrieval, and L3 session history in SQLite with threshold-gated cross-conversation recall.
• Proactive scheduler that turns natural-language instructions into cron jobs, plus Feishu and WeChat channels for reminders.
• Ships as an Electron desktop app, a React 19 web UI with SSE streaming, or a pure terminal CLI — all sharing the same runtime state.
• Pluggable LLM providers (OpenAI, Anthropic, DeepSeek, Ollama, local) switchable live in the UI.
• Optional OS-level sandbox via pi-sandbox and a session replay showcase that exports real sessions to a standalone site.

Caveats

• Explicitly not a team tool: no auth, no tenant isolation, no horizontal scaling, and no concurrent sessions. If you want multi-user, run one instance per person.
• The macOS desktop build is unsigned, so first launch requires a right-click open.
• When the prompt queue is occupied, cross-session operations surface a 409 session_busy error rather than hiding the delay.

Verdict

Grab it if you want a private, offline-capable tutor that remembers your weak spots and schedules review drills. Skip it if you are building a classroom platform or need concurrent multi-user sessions.