Resources
Where do I start with Inhold?
Start with the quickstart if you build agents, or the evidence pack if you review them. Every page here says what is built and what is not.
Inhold is being built. No customers yet. Repo goes public 9 Nov 2026.
Proof you can check
Proof you can check
Red-team results
30 attack cases against the policy, with the result of the run on main.
Download an evidence pack
A real pack from the synthetic demo, and 12 lines of Python that check its chain.
Reviewer packet
The 20 answers, the known gaps and the 164.312 map on one printable page.
We run our own agents on Inhold
Our company is tenant zero. What that looks like in code.
Try it
Try it
Quickstart
Write context, watch a second agent get denied, export the pack. About 15 minutes.
Connect an OpenAI Agents SDK agent
The MCP server entry, the run and the two audit rows it leaves.
Connect a LangGraph agent
Same two reads through langchain-mcp-adapters.
Connect a Claude Agent SDK agent
Same two reads, and the sign-in gap on AWS that is still open.
Keep your own agent sessions
Personal mode: your Claude Code and Codex history on your disk, secrets out of the agent's reach. Free.
Glossary
Principal, purpose, bundle, anchor and the other words the docs use.
Compare
Compare
Inhold vs hosted agent memory
AgentCore Memory, Foundry memory and Memory Bank.
Inhold vs agent memory libraries
mem0, Zep, Letta, LangMem and Cognee.
Inhold vs a vector database
pgvector, Pinecone and Weaviate.
Build it yourself or use Inhold
What your team would own.
Writing