Every agent session is recorded and fed back into future sessions on the same files — no cold starts, no repeated mistakes. Sessions are linked to the PRs they produce, so every line of merged code is traceable back to the agent decisions behind it.
Session timeline
Files indexed (5)
Context available
Agent decisions
MCP query result
get_waypoint_context(“src/auth/middleware.ts”)
4 sessions · JWT validation was refactored in session #1841 after discovering the previous approach didn’t handle token rotation correctly. The sliding-window rate limiter was chosen after global mutex caused queuing under load.
The problem
AI agents have access to the current state of a codebase. What they don't have is the history of decisions that produced it — or any record of their own activity.
A developer picks up a task and uses Claude Code to implement it. The agent has no idea a previous session last Tuesday already tried this approach and abandoned it for a specific reason. The mistake gets repeated.
Why does this function look overcomplicated? Why was this architecture chosen? The answers lived in an agent session that is now gone. Git shows the diff. It cannot show why the diff looks the way it does.
When multiple developers use agents on the same codebase, each session starts cold. Approaches get re-explored. Constraints get re-discovered. Decisions made in one session invisibly shape the code another agent misunderstands weeks later.
Compliance teams, security reviewers, and regulators are starting to ask: what did your AI systems change, and why? Today there is no answer. Agent sessions leave no verifiable record — just a diff in git and a timestamp.
Waypoint makes the decision record persistent, queryable, and auditable — automatically, as a side effect of the agent session.
How it works
A hook fires after every agent turn, parsing prompts, plans, file edits, and commands into a structured event log stored on your infrastructure.
During a session, the agent calls get_waypoint_context() via MCP for any file it's about to work on, pulling in the decision history from previous sessions as live context.
When a PR is detected, Waypoint associates the sessions that contributed to it — making every merged diff traceable back to the agent decisions behind it.
What changes
Every session inherits the decision history of previous sessions on the same files. Approaches that were tried and abandoned stay dead — the next agent knows why before it starts.
Every reasoning step, rejected path, and architectural decision is captured as it happens. Git shows what changed. Waypoint shows why the diff looks the way it does.
Session history is shared across every developer. Anyone picking up a task sees what every previous session tried — regardless of who ran it or when.
Every session produces a structured log of what the agent touched, decided, and rejected. Ready for internal review, security audit, or regulatory inquiry — no extra work required.
What gets captured
Raw events, not summaries. The reading agent synthesises what is relevant to the current task — because it knows what it is trying to do.
Every prompt the developer sent to the agent — the full request, not a summary.
Todo lists and thinking sequences the agent produced before taking action.
Every file touched during the session, with the context in which it was accessed.
Every shell command run, with exit codes and whether it succeeded or failed.
Every tool invocation — filesystem, search, browser, custom MCP servers.
Approaches the agent tried and discarded — including the reasoning for rejection.
Branch, commit SHA, changed files, and uncommitted work at time of capture.
Sessions associated with the branch that produced a PR — traceable from merged diff back to agent decision.
Which model was used — records are model-neutral and readable across Claude, Copilot, and others.
Agent compatibility
Stop hook — native
Stop hooks are a first-class feature of Claude Code. After every agent turn the hook fires automatically, capturing the full session record with no additional configuration.
Agent mode — hook-compatible
Copilot's agent mode in VS Code supports post-turn hooks. Waypoint's hook registration works without additional setup — the same capture capability across the most widely deployed AI coding tool in enterprise.
Also works with
Get started
Enterprise pilot · self-hosted
LaserOwl deploys Waypoint on your own infrastructure. No data leaves your environment. Fixed fee, clear scope.
£15k UK · $20k USA · fixed fee
20-minute intro call · fixed scope · no commitment
Cloud · hosted
Waypoint hosted on our infrastructure. Free trial, then a flat monthly fee. Point your CLI at our server and you’re done.
