VaultBase's context graph captures the full trace behind every decision — what was read, which policies applied, the reasoning, and the outcome. Search precedent before deciding. Get warned when you contradict yourself.
Every decision is a markdown file with YAML frontmatter. The frontmatter captures the inputs (documents read), policies (rules applied), tags, status, and outcome. The body holds the reasoning in plain text with wiki-links back to source documents.
Traces are first-class objects in VaultBase. They show up in search, get indexed for full-text retrieval, and participate in the permission system like any other document.
Before making a new decision, call trace_find_precedent() to search how similar decisions were handled. Filter by type, tags, or free text. Results include the full trace with edges and outcomes, so you can see exactly what was considered and why.
This is especially powerful for AI agents. Instead of hallucinating policy, the agent finds real precedent from your organization's history and follows established patterns.
When you create a new trace via trace_create(), VaultBase searches for prior decisions with overlapping tags and types. If the new decision contradicts established precedent, you get a warning with links to the conflicting traces.
You can still proceed — sometimes policy changes. But you won't accidentally contradict a prior decision because nobody remembered it existed.
Every document in VaultBase has a revision level. When a policy or source document is updated, any decision trace that cited an older revision gets flagged as potentially stale. You see which decisions may need revisiting because the underlying facts changed.
This closes a gap that every wiki has: decisions made based on information that was later corrected sit unnoticed forever. VaultBase surfaces them.
When agents or users search for information that doesn't exist, those failed searches are aggregated. VaultBase shows you what documentation is missing based on real demand — not guesswork about what someone might need.
Combined with stale reference detection, this gives you a living map of your knowledge base's health: what's outdated, what's missing, and what needs attention.
trace_create() Create a new decision trace with inputs, policies, reasoning, and outcome. Runs consistency check against prior decisions.
trace_find_precedent() Search existing traces by type, tags, and text. Returns full traces with edges and outcomes for context.
Institutional memory. People leave, context gets lost. Decision traces preserve the reasoning, not just the outcome. New team members can understand why things are the way they are.
Consistent decisions at scale. When 50 people are making vendor, hiring, and architecture decisions, contradictions are inevitable — unless you can search precedent first. VaultBase makes precedent searchable and warns on conflicts.
AI agents you can trust. Agents with access to decision traces don't guess at your organization's standards. They find real precedent, follow established patterns, and cite their sources. That's the difference between a useful agent and a liability.
Audit readiness. Every decision has a paper trail: what was considered, which policies applied, who approved it. When compliance asks "why did you choose this vendor?", the answer is already written down.
Free. Works offline. No account needed.
Open source. AGPL-3.0. Plain markdown files. Export anytime.