- Category
- Developer Tools
- Rank
- No. 1941Tools index
Previous survey · No. 1857 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Data, Retrieval & Knowledge
- GitHub
- 5 stars
- Date
About
Ingot is an open-source structured memory layer for AI agents that stores tool results and ingested documents as SQL-queryable rows in Postgres and as Parquet files. It chunks each document type on its own boundaries (PDFs by page, slides by slide, Markdown/HTML by heading, CSVs into typed rows) and runs without any model calls or API key by default, with embeddings, summaries, and OCR available as opt-in features.
At a glance
Ingot stores agent tool results and documents as structured records that can be queried across runs. It is worth evaluating if useful evidence is disappearing into transcripts and you need to inspect it with ordinary data tools.
Good fit
Teams that already operate PostgreSQL and object storage and want a searchable agent evidence store.
Consider something else when
Use a simpler local record if you only need to retain a handful of results.
Start here
Ingest one synthetic document and one fake tool result. Retrieve both, check their source boundaries and inspect what was written to storage.
Cost and requirements
The project uses PostgreSQL and object storage with Parquet. Basic storage and querying do not require a model key according to the docs.
Limits to know
The store can retain sensitive tool results. Embeddings, OCR and summaries are optional and may add separate processors. Check access and retention before enabling ingestion on real workloads.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
A memory layer that stores agent context as plain SQL-queryable Parquet rather than a vector store or proprietary format is a genuinely different, more inspectable approach to agent memory.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.