Shared, verifiable project memory for humans and AI
Shared project memory
for you and your AI.
Keep decisions, reviewed knowledge and repository context available across AI tools and team members. ReasonTrail preserves provenance and project boundaries while each contributor's raw conversations stay private.
- Claude Code
- Codex
- MCP-compatible clients
- Companion for macOS
The problem
Your project is bigger than one AI session.
Developers move between Claude Code, Codex and their own sessions all day. Each one starts without knowing what the last decided, so the same context is rebuilt from scratch and the reasoning behind a change lives in a transcript nobody reopens.
Context restarts
Every new session begins from nothing and rebuilds what the project already knew.
Decisions scatter
Why a change was made stays in the session that made it.
The project forgets
Work is remembered by whoever happened to run it, not by the project.
Review and control
Captured does not mean trusted.
A captured source becomes a classification and review proposal, not a fact. An authorized reviewer decides what becomes trusted project memory and can inspect the available evidence first.
Session review
Codex · Project Phoenix
- Task RT-112 completed
- 1 decision recorded
- 2 derived tasks
- 1 new idea
- 1 blocker
Illustrative review summary.
Capture → classify → review
Source material stays distinct from the proposal derived from it.
Delegate the decision
Review can sit with a manager or with someone they nominate.
Approve → reuse
Accepted or corrected knowledge becomes available to later authorized work.
Continuity
One project. Multiple AI tools. Continuous context.
One agent leaves reviewed knowledge and evidence. Another retrieves the bounded project context it needs and continues the work.
Agent A
Developer · Codex
- Completed the authentication middleware
- Recorded the decision behind the approach
- Flagged an unresolved blocker
Agent B
Reviewer · another AI tool
- Picks up the related task in the same project
- Asks ReasonTrail for the relevant project context
- Sees the decision and the open blocker before starting
Capture requires explicit setup. Continuity does not mean ReasonTrail automatically starts agents or resumes every native session.
Privacy by boundary
Share the knowledge, not the private transcript.
A team can reuse approved project knowledge without opening every contributor's raw AI conversation. Organization, project and role checks apply before context is retrieved.
- Contributor's raw transcriptPrivate to its owner
- Rejected review proposalNot project memory
- Approved project knowledgeShared within project access
- Bounded repository evidenceShared within project access
Project association never grants access to another contributor's raw transcript. How data is handled.
Context
Retrieve what the current task needs.
A client asking about one part of a project does not need the entire project history. ReasonTrail assembles a bounded, authorized slice and preserves its sources and revision context.
Relevant
Scoped to the task at hand rather than the whole history.
Bounded
Assembled to a budget, with truncation and coverage made visible.
Traceable
Every item cites the source it came from.
Versioned
Tied to a specific revision, so an answer is reproducible.
Bounded retrieval is verified; token savings are not established. The method is on the evaluation page.
Give your project context that survives the session.
Connect a project, keep working across tools, and decide what becomes trusted shared knowledge.