Compare
How is Inhold different from a vector database like pgvector or Pinecone?
Inhold is a policy and audit layer for agent context; pgvector, Pinecone and Weaviate store and search vectors at far larger scale.
Inhold is being built. No customers yet. Repo goes public 9 Nov 2026.
Comparison
| Product | Where it runs | Policy per read, by purpose | Audit of reads, incl. denials | Tamper evidence | PHI labelling | Approval of learned behaviour | HIPAA / BAA (vendor's words) | Licence | Better than Inhold at |
|---|---|---|---|---|---|---|---|---|---|
| Inhold | In the end customer's account; v0 single machine, AWS kit in build | Yes: actor, purpose, data label; deny by default | Yes, one row per decision | Hash-chained audit rows | Local patterns, plus an optional local name model | Yes, named approver for each skill | No certification; runs in your account | Apache-2.0 | Not applicable |
| pgvector | Anywhere Postgres runs, including the customer's own account | Postgres roles and grants; purpose not stated | Not stated by pgvector; depends on your Postgres setup | Not stated | Not stated | Not applicable | Not applicable (extension) | PostgreSQL License | ACID, point-in-time recovery, joins with your other data, mature tooling |
| Pinecone | Managed service; BYOC data plane in your AWS, GCP or Azure account | RBAC for users, service accounts and API keys. Purpose not stated | Audit logs cover the management and control plane, on Enterprise or the HIPAA add-on. Data-plane reads not stated | Not stated | Not stated | Not applicable | HIPAA claim across AWS, Azure and GCP in an Oct 2023 blog post; BAA on request | Proprietary service | Managed scale, BYOC, CMEK, private endpoints |
| Weaviate | Serverless, Dedicated Cloud, BYOC in your VPC, or self-managed | RBAC at collection and tenant level. Purpose not stated | With RBAC on, logs all authorisation decisions; denials logged at error level | Not stated | Not stated | Not applicable | BAA available; contact sales before storing PHI. SOC 2 Type II | BSD-3-Clause | Hybrid search, multi-tenancy, BYOC, logs allowed and denied decisions |
Where they are the better choice
- You need fast similarity search over millions of vectors. Any of these will outperform Inhold v0, which runs on SQLite.
- Your data team already runs Postgres. pgvector keeps vectors next to your other tables with joins and transactions.
- You want a managed service with BYOC. Pinecone and Weaviate both run a data plane inside your cloud account.
- You want authorisation decisions logged out of the box in a database. Weaviate logs each allowed and denied decision when RBAC is on.
- You need a BAA from the database vendor. Pinecone and Weaviate both say a BAA is available.
Questions
Questions
Is Inhold a vector database?
Inhold is a context layer that stores agent memory, approved skills and corrections, with a policy check and audit row on every read and write. It uses a database underneath; the AWS version is planned on Postgres 16 with pgvector.
Why not just add access rules to my vector database?
Inhold checks purpose and data label on each read, which database roles do not express on their own. You can build that yourself on any database; Inhold packages it with a hash-chained audit and an evidence pack.
Can I use Inhold with Pinecone or Weaviate?
Inhold has no connector to Pinecone or Weaviate today. Its planned store in AWS is Postgres 16 with pgvector.
Sources (all checked 31 Oct 2026)
- pgvector repository: https://github.com/pgvector/pgvector
- pgvector licence: https://github.com/pgvector/pgvector/blob/master/LICENSE
- Pinecone security reference: https://docs.pinecone.io/reference/security
- Pinecone audit logs: https://docs.pinecone.io/guides/production/configure-audit-logs.md
- Pinecone BYOC overview: https://docs.pinecone.io/guides/production/bring-your-own-cloud
- Pinecone HIPAA post (11 Oct 2023): https://www.pinecone.io/blog/hipaa/
- Weaviate security: https://weaviate.io/security.md
- Weaviate RBAC and authorisation logging: https://docs.weaviate.io/deploy/configuration/configuring-rbac
- Weaviate repository and licence: https://github.com/weaviate/weaviate
Facts about other products come from each vendor's own pages. "Not stated" means we did not find it there, not that the product lacks it. Tell us if a row is wrong and we will fix it.