AdroitPadNEW IN ADROITPAD 2.0
The project's memory, kept true
Decisions, conventions, gotchas and findings — with why they're true, who vouched for them, and what replaced them. Your IDE, agent and model can change. This stays.
MEMORIES
A memory is a fact and its why.
Redis holds session state
The app runs on several ECS instances, so sessions need one shared store.
- Link it to the issue, TODO, note, design doc, meeting or diagram it explains.
- Watch the files it describes and ask when they change.
- Every memory belongs to the Project and Module selected at the top.
TRUST
Trust you control.
Anything the assistant or an agent writes starts as a Hypothesis. Only you make it trusted.
Observation
Hypothesis
Reviewed
Human verified
Project canonical
Confirm, Still true, Supersede and Retract move a memory. Nothing changes status by itself.
FRESHNESS
Knowledge that notices when it goes stale.
Trusted memories can be re-checked every 30, 60, 90, 180 days or never. After twice the interval they become “probably out of date” and count less in search and answers. Watched files ask “changed since you last confirmed this — still true?” Duplicates and contradictions are flagged so you pick which one stands; the other moves to history.
ASK
Ask the project.
Ask a question and get Trusted knowledge, Unconfirmed, No longer true and Related work. Every line is numbered and clickable. Copy for AI makes a brief to paste anywhere; the on-device assistant answers from the same sources with [1], [2] citations and says when it relies on something unconfirmed.
PROJECT STATE
Project state and handover.
One tab answers “what is true here right now?” with Needs attention, Trusted knowledge, Unconfirmed, Changed in the last 7 days, open issues, active TODOs and meetings. Counts come from your records, no AI involved. Copy state for AI and Copy handover for AI make briefs for a coding agent.
CONTEXT
Why? on every record.
Link knowledge to the work it explains. On an issue or file, Why? shows Trusted, Check first and Probably related — with the memory behind the record.
ADRS
Decisions become ADRs.
Export a Decision memory as an Architecture Decision Record (ADR-0001 …): a Design Doc with Status, Context, Decision and History filled in, refreshable from the memory. Signed design docs are recorded as Decisions automatically.
EXTERNAL AGENTS
Your coding agent can use it.
Agents ask with adroitpad_get_project_context and get the same cited sections. Suggestions arrive as Hypotheses for you to confirm.
FAQ
A local source of truth, with honest boundaries.
Is this a cloud knowledge base?
No. Memories live in your encrypted vault on your machine. Nothing is uploaded unless you connect an external agent, in which case the agent sends the memories you allow to its own model provider.
Can the AI mark things true?
No. Anything written by the assistant or an agent starts as a Hypothesis. Only a person can confirm it.
What if a decision changes?
Supersede the old memory. History keeps the original and links it to its replacement.
Does it need the AI?
No. Memories, trust, project state and Why? work without it. The local AI adds semantic search and cited answers.
Can my coding agent read it?
Only if you connect one. It then sends what it reads to its own provider, under that provider’s terms.
ADROITPAD 2.0