Your agent forgets you every session. I built the memory layer instead of another wrapper

Description and use

Pomnia builds a local memory layer that stores and indexes agent conversations, eliminating the need to re‑explain context and saving tokens.

Problem

Coding agents retain context only for a session, forcing users to repeatedly re‑explain repo layout, constraints, and past decisions, which wastes tokens and time.

Outcome

Pomnia eliminates the need to re‑explain context, cuts token costs, and gives users a portable, copy‑able memory vault that can be backed up, shared, or deleted at will.

Workflow

• 1. Install Pomnia Desktop on the local machine.
• 2. Pomnia Desktop reads chat logs from installed agents and external sources, distills them into structured notes using a local LLM via Ollama.
• 3. Notes are written to a plain‑text vault folder (encrypted blobs for sidecars, plaintext for notes and index).
• 4. Run brain‑core on a server to expose an HTTP API that serves the indexed vault to agents using bearer tokens.
• 5. Agents query the API with MCP; brain‑core performs hybrid search and returns chunks with a prose verdict (strong, lexical, none).
• 6. Agents use the returned chunks to answer queries without re‑explaining context.

Evidence

Pomnia is a memory layer that sits underneath whatever agent you happen to be using, speaks MCP, and stores its corpus on hardware you control.

Source

I stopped re‑explaining myself. This is the boring one and it is the one that matters.

Source

Pomnia’s vault blobs are encrypted with AES‑256‑GCM, random IV per blob, integrity tag, key derived by scrypt at N=2¹⁷ from a passphrase that is never stored.

Source