DeviDevs, internal system
AI agents that search the company's notes and decisions before they act
DeviDevs agents search what the company already knows before they take a step: notes, rules, decisions and past work sessions. The search runs by keyword and by meaning at the same time, and the answer comes with its source. A fixed error goes into a register so it is not debugged from scratch again.
Updated
The challenge
An agent without memory repeats mistakes that were already solved and decides without knowing what was agreed before.
What we built
We connected the agents to a search over the company's notes and to a second search that goes through email, documents, Drive and tickets and reports which source did not respond. We build the same thing on your contracts and procedures, as an assistant that answers from your company's documents.
How it works
- Before a decision, the agent searches notes, rules and earlier decisions.
- Keyword search and meaning search run together, so it also finds things worded differently.
- The answer comes with its source, so a person can check it.
- If a source did not respond, the agent names it, and an empty result does not count as proof.
What changed
- An error fixed once is not debugged from scratch again.
- A "we don't have that" only comes after the company's sources were searched.
- Agents look up what was already decided, then decide.
Stack
- RAG
- Hybrid search
- Embeddings
- MCP
Related services
Related projects
Want an assistant that answers from your company's documents?
Your contracts, procedures and archive can be searched the same way, with the source next to the answer.
See the document chatbot