Sessions

Surfaces: web app · desktop app (rail and chat tabs) · CLI (kcap sessions, kcap recap) · your agent (search and recap tools)

The web app is at https://<your-workspace>.kcap.ai for a cloud workspace; on the local edition the same pages are served from your machine. The desktop app opens it from Help and support → Open dashboard, and kcap local open opens the local one from a terminal

A session is one conversation between a person and a coding agent, stored as it happens: prompts, replies, every tool call with its result, token usage, the commits it made and the pull requests it opened. Sessions are the raw material for everything else in kcap

What you can do

Find a session. Sessions in the sidebar is a list of cards, newest first, for whatever Scope is set to: one repository, a project, or everything. Above the list, All, Active and Ended filter by state, and the owner selector switches between your sessions and another person’s visible ones. Each card shows the title, the owner, the coding agent, when it was last active, and the token usage and evaluation score where known; a running session carries a live indicator. Search in the sidebar jumps to a session by title when you know it

Read what happened. Opening a card lands on the Transcript tab: a turn index down the left, one entry per prompt, and a reading pane on the right. Select a turn to read the prompt, the reply and the tool calls under it; expand a tool result to see what came back. Full shows every tool call; Summary shows a shorter reading where a summary exists. A subagent is a row under the turn that spawned it, readable in place. Shell commands keep their output in one card with the command. The reader remembers your place across reloads

Watch it live. A running session’s reader follows the newest turn until you select an older one, so you can inspect an earlier exchange while the agent keeps working. The Events tab shows the underlying stream: tool calls, results, errors and lifecycle events, for when the conversation alone does not explain something

See what it accomplished. The Overview tab holds the outcome (a generated account of what was done and what is left), the task list and any plan documents the agent declared, the pull requests it opened with their state, the work item it is filed on, and Tokens: usage per model, fresh versus cached, and estimated cost. Outcomes are generated after a session ends; if one is missing, owners and admins can choose Generate outcome

Follow a session across resumes, compaction and /clear. A long piece of work often spans several agent sessions: a resume after a break, a compaction, a /clear. kcap links them into a chain and shows the chain as one card with its segments, so the work stays together. Clearing a chat, though, starts a session that inherits nothing: it is unfiled until its own work files it. In the terminal, --chain on recap, errors and eval covers the whole chain

Share a session. The Share control on a session sets its audience: My default, Org members, Project members, Teams & people, or Private. Where the detail-level selector is offered, Full, Activity and Overview decide how much a reader sees; only Full includes raw transcripts and tool output. A link alone never grants access. A session link pasted into Slack unfurls with its summary when the session is organisation-visible, and as a lock placeholder when it is not. Change who can read your sessions has the detail

Fork a session from a turn. In the transcript, choose the turn where you want things to go differently and fork it: pick a machine, say what should change, and a new hosted agent picks up from that point while the original keeps running. The Compare view later sets the two runs side by side: changes, cost, evaluation. Forks need a daemon; see Hosted agents

From your agent

Everything above is also a question you can ask the coding agent you are working in:

What have we done on rate limiting in this repository? Find the sessions and summarise them

Recap session acme-billing-rate-limit and tell me what is left

The agent searches the sessions you can see, reads what it needs, and cites them. What your agent can do with kcap has more

From the terminal

kcap sessions                       # active sessions in this repository
kcap sessions --all --mine
kcap sessions --touching src/app.ts # sessions that edited a file
kcap recap <session>                # plan, summary and a turn outline
kcap recap --repo                   # recent summaries in this repository
kcap recap --chain --full <session>
kcap recap --per-turn <session>     # then --get-turn N for one turn in full
kcap errors <session>               # the failed tool calls

A session id or its readable slug both work for recap and errors

Help me get unstuck

A session I know exists is not in the list. Either you cannot see it (ask its owner to share it) or the server is catching up; a notice above the list says so, and the session appears when it is done

Two cards for one piece of work. The second is a continuation that kcap could not link, most often because the chat was cleared. Both are complete on their own

The outcome is missing. Outcomes need a configured summary model and are generated after the session ends. Admins set the model under Administration → AI & models → Summaries

The transcript stops partway. A record too large to process is replaced by a capture-loss marker and the session shows a warning. For a Claude Code or Codex session you own, kcap import --session <id> --repair-capture restores what the local transcript still has

Token counts look wrong. They are what the agent reported. Cache reads and writes are counted separately where the agent provides them, and cost depends on a known price for the model

Older links to Trace or Details. They open the Transcript and Overview tabs