Memory systems
Hindsight
Biomimetic memory networks (facts/experiences/beliefs) with Retain-Recall-Reflect
Open source + paid cloudself-hostable
★ 60,234 on GitHub · +90 today · last commit Nov 10, 2026
What it is
Our note
Fastest riser in the space (19k->60k stars in ~10 weeks 2026). LongMemEval 91.4 — the most independently reproduced memory benchmark claim (Virginia Tech, Washington Post). Publishes a live per-model benchmark site.
At a glance
- Vendor
- Vectorize
- Business model
- Open source + paid cloud
- Pricing
- Self-host free; Cloud usage-based (Retain $10/M tok, Recall $0.75/M, Reflect $0.05/call) pricing page ↗
- License
- MIT
- Self-hosting
- Yes, can be self-hosted
- Access
- REST, Python SDK, MCP (default-on), Docker/Helm
- Architecture
- four-network biomimetic
- Funding
- $3.6M seed (True Ventures, Dig Ventures)
GitHub activity
This product has a public repository, so we read it nightly from the GitHub API. Stars measure developer attention, not quality or market share.
“Hindsight: Agent Memory That Learns” (repository description)
- Stars
- 60,234
- Change since last reading
- +90
- Forks
- 7,355
- Open issues
- 199
- Last commit
- Nov 10, 2026 (0 days ago)
- License (GitHub)
- MIT
Links and sources
- Official sitehindsight.vectorize.io
- Pricingvectorize.io/pricing
- GitHub repositorygithub.com/vectorize-io/hindsight