01 · Lost session
Your laptop restarted. Your session didn't die.
Every session is recorded as it runs. Reopen it, read the recap, continue where the agent stopped — on any machine, even the one at home.
session recording + recapFor everyone who codes with agents
Your laptop restarted mid-session.
kcap records every coding-agent session — so you can continue it, hand it to another agent, search it, and learn from it.
Recording is opt-in, and your first import gives you a searchable history in minutes.
Try the dashboard in the browser first. Install the CLI when you want sessions recorded.
Works with Claude Code Codex Cursor Agent Gemini CLI Copilot CLI Kiro OpenCode Pi Antigravity
❯ kcap setup
✓ Connected · Claude Code, Codex detected
❯ kcap import
✓ Imported 214 sessions (3 months of history)
Your past work is now searchable.
Session recap · yesterday 16:42
Fix flaky auth-token refresh test
Tried mocking the clock first — abandoned after the race persisted. Root cause: token cache TTL rounding. Patched in auth/cache.ts, added regression test. Two approaches rejected, one shipped.
What a session leaves behindWhen the agent stops, the session becomes a record: what it tried, what it changed, what it cost, and how it scored.
One recorded session. Five problems gone.
02 · Credit wall
Out of Claude credits at 4pm? Continue in Codex at 4:02.
kcap is vendor-neutral. The session's context follows you across nine harnesses, so switching tools doesn't mean starting over.
cross-agent handover
03 · Monday morning
“What was I working on?” — answered, with receipts.
Ask what you did yesterday and get the actual sessions back. Work items track what's in flight and what's next.
recall + work items
04 · Groundhog day
Your agent made that same mistake last Tuesday. Make it the last time.
Evals score sessions against your rubric, and what they learn is injected into future runs. Mistakes become guardrails.
evals + injection
05 · The bill
Where do my tokens actually go?
Auditable charts of usage and spend — per repo, per agent, per person. Know before the invoice does.
analytics
The whole team, one dashboard
Everything your agents did this week, on one screen.
Live sessions, work in flight, what it cost and what was learned.
the honest timeline
Value on day one. Compounding after.
Minute 5
Start with history
kcap setup, then kcap import backfills the sessions you've already run. A searchable history, not an empty tool.
Day 1
See the work
Recap any session: what the agent actually did, in reviewable form. Share a link instead of pasting diffs.
Day 10
Start warm
“Why did we do this?” has an answer. New sessions begin from memory — your agents query it themselves.
Day 30
Compound
Decisions and dead ends retrievable by anyone — and any agent. Evals encode your rubric. Spend is visible.
two ways in
Built for how you work
Flying solo
You and your agents have already solved this — find it again. Your past sessions are your documentation: every approach you tried, every fix that worked, across every agent you use.
Start freeOn a team
Ship agent-built code you can explain. Review PRs with the reasoning attached, stop re-explaining context between teammates, and see quality and spend across the team.
Start a team trialRecording is opt-in per agent. Exclude any project and nothing from it is ever sent. Bring your own keys on the free plan, or self-manage the deployment.
fair questions
Before you ask
What is kcap?
Shared memory for coding agents. kcap records every Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, Kiro, OpenCode, Pi, and Antigravity session, then makes that record searchable, shareable, and usable by the next agent — so you can continue lost work, switch tools without starting over, recall prior decisions, and review PRs with the reasoning.
Which coding agents does it work with?
Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot CLI, Kiro, OpenCode, Pi, and Antigravity. Recording is opt-in per agent. All nine support hosted launches from the dashboard when installed and authenticated on a connected daemon; controls depend on the vendor and platform.
My agent already has session history.
Per tool, per laptop. Claude Code can resume Claude Code — on that machine. kcap is one record across nine harnesses and your whole team, and it's served back to every agent, not just the one that wrote it.
How long until this is actually useful?
About five minutes. kcap import backfills your existing sessions, so search and recap work before you’ve recorded anything new.
You're storing my transcripts?
Only if you turn recording on, and never for projects you exclude. Sessions are stored on your organization's kcap server. Free includes managed server summaries and evals.
We already have observability tools.
Those measure your app in production. kcap records how your software gets built — the agent sessions behind the commits.
The next session you lose is the one you didn't record.
Install now, import your history, and the next crash, credit wall, or Monday morning costs you nothing.