CodeSesh

Local / Zero config / 9 agents

Every AI coding sessionis still here.

CodeSesh scans local histories from nine AI coding agents and puts them in one index: organized by project, structurally searchable, and replayable message by message. Session content and indexes stay local. No account, cloud sync, or session telemetry.

npx codesesh
http://localhost:4521
Reads sessions from
01 / Overviewapps/web / dashboard

Open it and see where the week went

Sessions, messages, tokens, cost, and recent activity come from one local index. Change the time window and the overview recomputes.

Switch the time range and watch the whole panel recompute

CODESESH
? ShortcutsLast 7d / Apr 16-23, 2026Sample data / DEMO
Dashboard
Sessions
212
+18.4% 30.3/day
Messages
32,764
+12.1% 154.5/session
Tokens
1.58B
+9.7% Cache hit 61%
Cost
$84.20
+7.2% $18.40 estimated
Last activity
2h
webapp / Claude Code
Daily Activity212 sessions / last 7 days
  1. Apr 17
  2. Apr 18
  3. Apr 19
  4. Apr 20
  5. Apr 21
  6. Apr 22
  7. Apr 23

Agent Distribution

5 agents / 212 sessions
  • Claude Code67%142 / 67.0%
  • Codex18.4%39 / 18.4%
  • OpenCode8.5%18 / 8.5%
  • Kimi5.2%11 / 5.2%
  • Cursor0.9%2 / 0.9%
02 / Project treeapps/web / project timeline

Sessions go back where they belong

Group history by repository and project, keep subagent sessions under their parent, and aggregate messages, tokens, and cost by hierarchy.

Expand a session marked sub to inspect its child agent work

CODESESH
Dashboard / webappLast 7d / Apr 16-23Sample data / DEMO

webapp

96 sessions / 4 agents / /projects/webapp

APR 23, 2026 / THURSDAY

Implement onboarding wizard flowClaude Code / 84 msgs / 412K tok / 3 sub-sessions3 sub$1.86 incl. sub
  1. Generate step validation schema22 msgs / 96K tok
  2. Write WelcomeStep component tests17 msgs / 71K tok
  3. Audit a11y on the wizard stepper9 msgs / 38K tok
Refactor query building for readabilityCodex / 41 msgs / 188K tok$0.74
Fix webhook signature verification bugOpenCode / 29 msgs / 104K tok / 1 sub-session1 sub$0.41 incl. sub
  1. Reproduce the HMAC mismatch locally11 msgs / 34K tok
Tune session list virtualizationCursor / 18 msgs / 63K tok$0.22
03 / Replayapps/web / session detail

Replay the complete path of a task

Read messages, tool calls, and file changes in sequence, then use type filters and file tracking to find the context that matters.

Expand tool calls for output, or use the TOC to hide message types

Dashboard / Codex / Refactor query building for readability

SessionRefactor query building for readability#9c7a2d1b / updated 7h agoSample data / DEMO
Showing 3 messages
  1. USER

    The current query builder is hard to follow. Refactor it into composable helpers and document the intent of each step.

  2. Codex
    AGENTgpt-5.4

    I will split the builder into focused functions and keep its behavior intact.

    edit_filesrc/db/queryBuilder.tsEDITED+128 -46
    @@ buildQuery()
- const parts = [];
- return appendFilters(parts, filters);
+ const clauses = buildClauses(filters);
+ return clauses.filter(Boolean).join(" AND ");
    bashpnpm test:unitSUCCESS

    148 passed0 failed / 2.34s

  3. Codex
    AGENT

    Updated the builder, added types and focused comments, and kept behavior intact.

    writesrc/db/queryBuilder.spec.ts+86 lines

Discover, organize, recover, replay.History becomes useful after all four.

CodeSesh follows the real loop of AI-assisted engineering work.

Discover

Bring local sessions from different agents into one index.

  • Zero configuration

    Run one command and scan supported agent sessions on your filesystem.

  • Unified timeline

    Browse histories from nine AI coding agents in one interface.

  • Live refresh

    Add local session changes to the index without restarting the UI.

Organize

Put sessions back into project, task, and engineering context.

  • Project and session tree

    Group by repository and keep subagent sessions under their parent.

  • Smart tags

    Label fixes, refactors, features, tests, docs, planning, and more.

  • Session aliases

    Give key sessions memorable names that persist in search and bookmarks.

Recover

Bring old decisions, paths, and context back into the current task.

  • Structured global search

    Search titles, messages, tool output, and paths, then filter the results.

  • File activity index

    Start with a file and find sessions that read or changed it.

  • Keyboard navigation

    Switch views, focus search, and move through groups from the keyboard.

Replay

Reconstruct the full path from problem to result.

  • Full conversation replay

    Keep messages, tool calls, and reasoning steps in sequence.

  • Cost and token visibility

    Compare tokens, cache hits, recorded cost, and model estimates.

  • Local SQLite index

    Use one local database for fast restore, search, and schema migrations.

Built for the local AI coding stack

Each agent connects through a core adapter and contributes to one index of sessions, projects, search, and file activity.

  • Claude Code
  • Cursor
  • Kimi
  • Kimi-Code
  • Codex
  • Grok
  • Pi
  • OpenCode
  • ZCode

Frequently asked questions

Answers about CodeSesh, installation, data boundaries, and large histories.

What is CodeSesh?

CodeSesh is a local developer tool for discovering, aggregating, searching, and replaying AI coding session history. It turns local records from Claude Code, Cursor, Kimi, Kimi-Code, Codex, Grok, Pi, OpenCode, and ZCode into a project-aware engineering memory layer.

Does CodeSesh upload local AI session data?

No session data is uploaded. CodeSesh uses a local SQLite index and a Web UI served on localhost. Session content, file paths, token statistics, and recorded costs remain on the computer. The product requires no account, cloud sync, or session telemetry.

How do I install and start CodeSesh?

Run npx codesesh in a terminal. CodeSesh scans supported local AI coding sessions and opens its Web UI at http://localhost:4521. If the default port is busy, it tries the next available port. The published CLI requires Node.js 22 or later.

How does CodeSesh stay responsive with a large history?

The first scan persists backfill progress and can resume after interruption. Later starts restore from the local SQLite cache, while file watchers incrementally update changed sessions. Long conversation timelines use viewport virtualization instead of rendering every message at once.

Bring your coding history back

Open source and free, with session data kept locally. One command starts a searchable engineering memory for your AI coding work.

npx codesesh
View on GitHub