Decispher

System of record for engineering decisions.

Productivity

Developer Tools

Artificial Intelligence

Visit WebsiteSee on Product HuntTwitter ⧉

Hey Product Hunt 👋 I built Decispher because I was tired of one specific thing: my AI agent writing clean code that was confidently wrong. It used floats for money. It brought back a bug we had fixed months ago. It picked an approach we already tried and dropped. Every time, the rule existed somewhere. It just lived in a Slack thread, a PR comment, or someone's head, never anywhere the agent could read. So it guessed, and I paid for that guess in tokens, review cycles, and trust. Here is the realization that started it. Every team has two kinds of knowledge. The explicit kind lives in your repo: code, READMEs, tickets, and AI tools see that fine. The tacit kind lives in people's heads: why the architecture is that way, what got tried and failed, which invariant must never break. Your agent is blind to the second kind, and that is the part that actually matters. Decispher is the layer that fixes this. 1) It captures the "why" automatically. It reads the conversations you are already having in Slack, GitHub, and Jira, and pulls out the decisions, conventions, and constraints. No new tool to babysit. No writing docs by hand, which nobody keeps up anyway. 2) It serves context on demand, not in bulk. Instead of dumping your whole codebase into every agent call and hoping it finds the right thing, the agent pulls only the decision it needs, when it needs it. In my testing this cut token use by about 65%, and the answers got better because they came with the reasoning, not just the code. 3) One shared place for humans and agents. Your teammates and your AI tools (Claude Code, Cursor, Copilot, over MCP) read and write to the same context. A decision made once is there for the next person and the next agent, in whatever tool they use. Knowledge stops disappearing into DMs. 4) Branch Story. Agents forget everything between sessions. The reasoning, the approach already abandoned, the constraint learned the hard way, all gone when the session ends. Branch Story keeps that per branch, so you never start a session cold and the agent stops repeating mistakes. A few principles I held to while building it: a) Capture has to be passive. Any system that asks engineers to write decisions down dies in week two. The signal already exists in your conversations, so we pull it from there. b) "Why" beats "what." Code already tells you what. The expensive missing piece is the reasoning and the rejected alternatives, and none of that is in the repo. c) Context is the real moat, not the agent. Models are getting cheap and interchangeable. Knowing why your system is the way it is is the hard part. d) It is early, and that is exactly why I am here. Try it on something real and tell me what value it brought, or where it fell flat. I want honest feedback over polite feedback. New makers get 50 free credits, no card needed. Would love your thoughts. AMA below 🙏