ReasonTrail for Teams is now available

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 full transcript pages and marked-private turns stay owner-only.

Explicitly captured project sources from Claude Code, Codex and the ReasonTrail Companion pass through classification and review, where an authorized reviewer decides what becomes approved project memory for the next person or agent. Private turns stay owner-only; non-private project turns can be retrieved by authorized project agents.
PRIVATE SOURCEClaude CodesessionCodexsessionCompanionmacOSCLASSIFY · REVIEWManageror delegateapproves · editsor rejectsAPPROVED MEMORYCompleted workDecisionsTasks · derived tasksIdeasBlockersRelevant knowledgeNEXTNext devor agent
  • Claude Code
  • Codex
  • MCP-compatible clients
  • Companion for macOS

Watch the overview

How ReasonTrail works.

An illustrated workflow showing how project work becomes reviewed knowledge for the next person or AI session. Example panels are illustrative; the current screen guides are linked below.

69 seconds · Synthetic English narration and captions

Developer and sysops examples · What your team can access

Read the transcript

A new AI session can mean explaining your project all over again. The decisions, the trade-offs, and what still needs checking.

ReasonTrail gives that context a home. It keeps reviewed project knowledge available across people, tools, and work sessions.

First, configure capture for your project. Work from tools like Codex and Claude Code becomes source material and review proposals.

Review the evidence behind a proposed decision. Approve, edit, or reject it before it becomes trusted project memory.

When a teammate or another AI tool picks up the work, request the relevant context. See the approved decision, its source, and the open blocker.

Access stays scoped to the project. Transcript pages and marked-private turns stay owner-only. Authorized project agents can search non-private turns.

Give your project context that survives the session. ReasonTrail. Every decision has a trail. Explore more at reasontrail.ai.

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

Awaiting review
  • Task RT-112 completed
  • 1 decision recorded
  • 2 derived tasks
  • 1 new idea
  • 1 blocker
ApproveInspectEditReject

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.

Governance you can inspect

Choose EU, German or Swiss control profiles for developer and ops project memory. Restrict ReasonTrail's hosted AI routes, require a signed review of the saved configuration, and download the control assessment. Changes require a new review.

Each assessment shows operational gaps, including retention, data-rights procedures and processing destinations. Country profiles support your review; they are not legal certification or a guarantee of worldwide compliance.

See the controls and their scope

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
reviewed into project memory

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.

  • Full transcript pages and private turnsOwner-only
  • Non-private linked project turnsAuthorized project agents
  • Rejected review proposalNot project memory
  • Approved project knowledgeShared within project access
  • Bounded repository evidenceShared within project access

Full transcript pages and marked-private turns stay owner-only. Authorized project agents can retrieve non-private turns from linked conversations within their project and token permissions. 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

Source and revision references let you recheck the evidence. Generated answers can vary.

In controlled gpt-5.6-sol xhigh benchmarks, applicable tasks used 31.20% fewer provider-native worker tokens with no quality regression. The evaluation page explains when that claim does and does not apply.

Give your project context that survives the session.

Connect a project, keep working across tools, and decide what becomes trusted shared knowledge.