KodaCode
Coming soon
Docs/Working with the agent

Tool calls in the transcript

How the transcript shows each call the agent makes, and the three ways to draw it.

Everything the agent does is a tool call: reading a file, querying the code graph, running a command, editing, delegating. The transcript shows every one as it happens. Choose how in Preferences → Appearance → Tool calls.

Compact (the default)

Compact draws the session as plain lines:

  • One line per call, with its time.
  • Diffs under their edit.
  • Your message and each answer between bars.

Click a call to see its output.

──────────────────────────────
❯ The chunked upload test fails on the last chunk. Fix it.
──────────────────────────────
  $ cargo test chunked           2.3s [exit 101]
  ⌬ Graph read_chunk             0.2s
  ▤ Read src/parser.rs
  ✎ Edit src/parser.rs           awaiting approval
  $ cargo test chunked           2.1s
──────────────────────────────
✦ Koda
  The last chunk was dropped when it was shorter than the buffer…
──────────────────────────────
─ 41s · 5 calls · read 1 file · edited 1 file +2 −1

Each row starts with a mark for its kind:

MarkCalls
Document, search, folderRead, Search, List
HubGraph: a code graph query
History, bookRecall and memory: memory
Terminal$ shell commands
PencilEdits, opened to show their diff
HandDelegate: work handed to an agent
PlugAn MCP tool
GlobeWeb search and web fetch
ShieldA call awaiting approval
CrossA failed call, with [failed], [exit N], [interrupted] or [invalid]

A running call pulses and says running. The line under the last answer summarizes the turn:

  • How long it took.
  • The number of calls.
  • The files read.
  • The files edited, with lines added and removed.

Thinking is hidden in Compact.

Live step

Shows only the running call, and a failure in the latest turn, as cards. Finished calls fold into a step timeline.

Grouped

Folds each run of quiet calls into one line with a summary and a count, which you can open.

Output and long results

Long command output keeps its head and tail. Show all N lines loads the rest from disk. Saved results are kept for 30 days by default (Tool Result Retention in General settings).

Show thinking

Show thinking in Appearance shows the model's thinking open beside its answer. Off, it folds into one line you can click to read. Compact always hides it.