changelog

What changed, and when.

Release notes, newest first.

v0.21.0

feature

Grok Build capture, a live Session Graph, and docs that match the CLI

Hook Grok Build the same way you hook Claude Code or Cursor. Watch a session as a live graph in the browser. The public site now lists all six tools, the init wizard, and how to install the editor extension.

  • new Grok Build hooks: `promptconduit install grok` captures prompts, tools, sessions, and subagents in real time
  • new `promptconduit graph` opens a live Session Graph in any browser — 100% local, no editor required
  • new Docs for the editor extension: search PromptConduit on Open VSX, or let `promptconduit install cursor` sideload it
  • improved Public docs, pricing, and llms.txt list all six hook targets — Claude Code, Cursor, Gemini CLI, Codex, Copilot, and Grok Build
  • improved Onboarding copy points at `promptconduit init` instead of copying an API key per tool
  • improved Weekly spend digest leads with API-equivalent cost so the email matches what the dashboard already shows
  • improved Homepage editor screenshots match the current extension — cost ledger with coaching tip, Session Graph, Stream, and Agent Coaching

v0.20.0

feature

Pro tier, GitHub PR cost comments, and OTLP persistence

Personal Pro subscriptions ($10/mo) for cloud history and digests, GitHub connect UI for PR comments with estimated spend, OTLP traces/logs persisted to the platform, and buyer-intent blog posts for cost tracking.

  • fixed Session Graph follows the repo you have open instead of another live session on the machine
  • fixed Cursor sessions are tagged with the workspace folder, not whichever repo Cursor last launched from
  • improved Editor extension cost panel is simpler — one status bar item, prompt ledger with coaching tip, and Advanced drawer for raw JSON and what-if
  • fixed Cursor agent turns now show real prompt text in the cost ledger instead of "(uncaptured turn)"
  • new Team analytics break down AI spend by git branch and pull request, not just by repo
  • new Pro plan ($10/mo): personal cloud sync, 90-day history, weekly digest, cross-machine skill library — 14-day trial, no card required
  • new GitHub integration page at /github — connect repos and post AI session links plus estimated branch spend on pull requests
  • new OTLP receiver persists traces and logs to raw_events + R2 (set OTEL_EXPORTER_OTLP_ENDPOINT=https://api.promptconduit.dev/v1/otlp)
  • new Blog: track Claude Code token usage and Cursor spend by repo — SEO pages for buyer-intent queries
  • improved Editor extension links include UTM params for funnel attribution (no activation telemetry)

v0.19.1

feature

More accurate cost estimates for latest models

Cost estimates now use the latest Cursor and Anthropic published rates, including corrections for Claude Sonnet 5, GPT-5.6 Sol, and Grok 4.5 Fast, plus pricing for Opus fast mode and additional third-party Cursor models.

  • improved Claude Sonnet 5, GPT-5.6 Sol, and Grok 4.5 Fast sessions show corrected per-token pricing instead of stale or generic rates
  • improved Opus 5 and Opus 4.8 fast mode, GPT-5.4/5.5/5.6 variants, Gemini 3 Pro, and Kimi K2.7 Code are priced when selected in Cursor

v0.19.0

feature

Updated cost estimates for new Cursor models

The CLI and editor extension now price Grok 4.5/4.6, Claude Opus 5, Gemini 3.x Flash, and other recently added Cursor models instead of showing them as unpriced.

  • improved Cost estimates now include Cursor Grok 4.5/4.6, Claude Opus 5, Gemini 3.x Flash, GLM 5.2, GPT-5.6 Sol, and Kimi K3
  • improved Composer 2.5 sessions include cache-read token pricing from Cursor’s published rates

v0.18.0

feature

The editor extension follows the tab you select

Run several AI sessions at once and the extension keeps up: select a terminal running Claude Code or a Cursor agent tab, and the live Stream and status-bar cost switch to that session together.

  • new Selecting a Cursor agent tab switches the Stream panel and status-bar cost to that conversation — including in the Agents Window
  • new The Stream panel shows why it is on a session: focused terminal, selected agent tab, or your own drill-down — and "All activity" is one click away
  • improved The status-bar cost reflects the session you selected, not just the one you last prompted
  • improved The unified "All activity" stream and drill-down shipped in 0.17.2 stay put: event volume never moves the view — only your own selections do

v0.14.0

feature

Sign in from the CLI

Install, log in, and wire up your AI tools in one flow — no more copying API keys or running install commands per tool.

  • new `promptconduit login` signs you in via browser and saves your API key automatically
  • new `promptconduit init` walks you through tool selection, hook install, and a connectivity check
  • improved Getting started and overview pages now point to login + init instead of manual per-tool setup

v0.13.2

feature

One install command — curl, not Homebrew

We dropped the Homebrew tap to reduce install friction. Every doc, page, and CTA now points to a single curl command, and the CLI auto-updates itself in the background.

  • improved Install with one command everywhere: `curl -fsSL https://promptconduit.dev/install | bash` — no tap, no brew upgrade, auto-update handles the rest
  • removed Homebrew tap distribution — existing brew installs should reinstall via the curl script

v0.13.1

feature

Cursor sessions labeled correctly in the event log

A hook-detection bug caused Cursor Composer and Agent sessions to appear as Claude Code in the local event log and Stream panel. New events now carry the correct tool name.

  • fixed Cursor sessions no longer show as "claude-code" in the Stream panel and event log — tool detection now checks for Cursor before the shared hook_event_name field

v0.13.0

feature

See exactly what every prompt cost — and what it would have cost

The editor extension's AI Cost Breakdown is now a per-prompt detail report: every prompt shows its spend, tool calls, subagents, permission mode, and full raw event JSON, plus a "what if" comparison against other models. The live Stream panel gained the same raw-JSON superpowers.

  • new Per-prompt cost ledger: each prompt expands to its requests (token splits with plain-language tooltips), tool calls (MCP server, Skill, and failure markers with durations), subagents with per-agent model/cost/duration, permission events, and the complete raw event JSON — syntax-highlighted, copyable, with Expand all / Collapse all
  • new "What if" model comparison per prompt and per session: what the same token counts would have cost on other models, in both directions ("sonnet would have saved $0.54 (−80%)", "fable-5 would have cost +233%"), with an honest caveat that it compares rates, not capability
  • new Glossary tooltips on every metric term — cache read/write multipliers, fresh input, model tiers, permission modes (plan, accept edits, …), subagents, MCP — each linking to the relevant docs
  • new Session context line: repo @ branch, linked PR number/title/state, and a worktree badge, straight from the events your CLI already captures
  • new Stream panel: every live event row now expands to its human-readable raw JSON with a copy button, and the panel states plainly which session/conversation id it is following (copyable)
  • fixed Terminal focus could briefly show the wrong session after fast terminal switching; log rotation could double-count subagent stats; sessions with unparseable timestamps never sorted as active; unpriced models showed "$0.0000" instead of "unpriced"

v0.12.1

feature

Honest cost counts and history you control

Two fixes to the local cost tooling. The per-session request count in the AI Cost Breakdown is now the real number instead of a silently capped one, and the CLI keeps your full hook history for a window you choose rather than dropping it at a fixed size.

  • fixed The AI Cost Breakdown showed exactly "50 requests" for any busy session because the panel capped its history at 50. It now shows the true per-session request count and never drops a prompt from the data (the newest 100 rows render, with a "+N more" note for the rest)
  • new Configurable local history retention: keep every captured event for a window you set (default 30 days) with `promptconduit config set --event-log-retention-days=N` (-1 keeps forever), plus a new `promptconduit prune` command to reclaim space on demand
  • improved The CLI no longer discards recent events at a blind 50MB rotation — anything inside the retention window is always kept, and older records are trimmed oldest-first only past a generous size ceiling

v0.12.0

feature

App redesign: the paper theme comes to app.promptconduit.dev

The app now matches the marketing site: light paper page, ink text, IBM Plex type, and dark pixels reserved for your own data. Alongside the restyle, the app got dramatically simpler — one Overview page puts sessions, tokens, and coaching front and center right after login.

  • new Overview — a single post-login landing page with period stats (sessions, prompts, tokens, active days), coaching highlights, and recent sessions with one-click transcript drill-in
  • improved Every page restyled to the engineering-manual aesthetic: paper surfaces, hairline rules, teal accents, and terminal blocks for transcripts, commands, and API keys
  • improved Navigation simplified to Overview · Coaching · Skills — no more hunting for the useful pages
  • improved Transcripts support ?session= deep links from the Overview recent-sessions list
  • removed Dashboard and Insights merged into Overview (old URLs redirect); decorative system-status footers, gradient orbs, and ~3,000 lines of dead marketing code removed from the app bundle
  • fixed Getting Started page error/success notifications were silently broken; they now display correctly

v0.11.0

feature

Session restore, envelope v2, and a rebuilt site

Your AI sessions now survive an editor restart, capture moves to a single forward-compatible event log, and a rebuilt marketing site leads at promptconduit.dev.

  • new Session restore: the CLI reconstructs your recent Claude Code sessions (promptconduit sessions and resume), and the editor extension auto-reopens interrupted ones after a restart
  • new Live Stream panel in the editor follows the active AI session and shows its events as they happen
  • improved Envelope v2 — one on-disk event log per machine with slug-keyed enrichments: simpler capture that stays forward-compatible as new signals are added
  • improved One-click "Reload Window" when a CLI update refreshes the extension, so you reload without killing running terminal sessions
  • improved Rebuilt marketing site with refreshed docs, pricing, and this changelog

v0.10.0

feature

Cost Breakdown, Agent Coaching, and our first marketplace release

The editor extension learns to explain costs and coach your workflow, and PromptConduit ships to its first extension marketplace.

  • new AI Cost Breakdown panel shows what each session would cost at API rates — a teaching tool for understanding token spend, not a bill
  • new Offline-first Agent Coaching tab suggests ways to use your assistant more effectively, entirely on your machine
  • new Agent Coaching trends on the web: a /trends report drawn from your own sessions, with linked coaching guides
  • new First marketplace release — the extension is live on the Open VSX registry for Cursor
  • improved brew-aware self-update keeps the CLI and its bundled extension current
  • improved Captured git context now detects worktrees, so branch work is attributed correctly

v0.9.0

feature

A live cost meter and the editor extension

See what your AI sessions cost in real time — from the terminal or right inside your editor — with everything computed locally.

  • new promptconduit cost — a real-time, local-only token-cost meter for your AI coding sessions
  • new Free local-only tier: a full-fidelity event log written to ~/.promptconduit/ with nothing leaving your machine
  • new The PromptConduit editor extension — session cost in the status bar and a live telemetry panel — bundled with the CLI and auto-installed into Cursor
  • improved Per-conversation cost that follows whichever editor tab is in focus
  • fixed Multi-environment configs now honor top-level opt-out flags

v0.8.0

feature

A blog, cleaner skills, and the plugin marketplace

Field guides for Claude Code go live, the Skills page stops repeating itself, and you can install PromptConduit as a Claude Code plugin.

  • new The blog launches with practical Claude Code guides: customization, where transcripts live, and writing a skill from scratch
  • new Install PromptConduit as a Claude Code plugin straight from the marketplace
  • improved Skill and pattern de-duplication — the Skills page no longer surfaces near-identical suggestions, and a cleanup panel merges the rest
  • improved Recent Sessions now sorts and labels by last activity instead of start time

v0.7.0

feature

Codex, Copilot, and local-first capture

PromptConduit reaches two more assistants and gains a fully local capture path with its own dashboard.

  • new Hook support for OpenAI Codex CLI and GitHub Copilot CLI — capture sessions from more of the tools you use
  • new promptconduit collect runs a local-first receiver with its own dashboard, so you can review sessions without sending anything to the cloud
  • new Per-skill install and uninstall with a safe manifest — pick exactly which skills land in your editor
  • improved Spec-compliant SKILL.md generation with a parallelism classifier and validation
  • improved Persistent error logging for hook events makes capture problems easy to diagnose

v0.6.0

feature

Discoverability and complete event capture

PromptConduit gets easier to find — for search engines and AI answer engines alike — and captures the full shape of a Claude Code session.

  • new Full Claude Code hook coverage: 14 canonical event types capture the complete picture of a session
  • improved The site is now built for search and AI answer engines — robots.txt, llms.txt, per-page metadata, and FAQ/Breadcrumb structured data
  • improved Auto-generated sitemap and crawlable navigation so pages actually get indexed
  • fixed Corrected the skills path to ~/.claude/skills/

v0.5.0

feature

Skills library vocabulary and UX cleanup

The Skills page now speaks the same language you do. Internal approval workflow jargon replaced with clear, action-oriented labels.

  • improved Status labels rewritten: "Pending" → "New", "Approved" → "Ready", "Rejected" → "Removed"
  • improved Action buttons renamed to match intent: "Approve" → "Save", "Reject" → "Remove", "Re-approve" → "Restore"
  • improved Confidence percentage bar removed — replaced with session count (e.g. "12 sessions") so you know how battle-tested a skill is
  • fixed Onboarding banner no longer tells you to set PROMPTCONDUIT_API_KEY as an env var (wrong approach — links to Profile instead)
  • removed Stale "View all →" link to the Transcripts page removed from dashboard Recent Sessions header

v0.4.0

feature

Dashboard metrics that mean something

The three stat cards on the dashboard used to show raw hook event counts — a number that meant nothing to anyone. They now show metrics that reflect how you're actually using AI in your workflow.

  • improved "Total Events" → "Sessions This Week": how many coding sessions you've had in the past 7 days
  • improved "Today" → "Day Streak": consecutive days with at least one session — a signal for consistent AI-augmented work
  • improved "Tools" → "Active Days": unique days with sessions in the past 7 days
  • improved Skills preview on dashboard filters out removed skills and uses New/Ready labels instead of Pending/Approved

v0.3.0

feature

Clickable session cards with tool breakdown

Session cards on the dashboard are now interactive. Click any session to see which tools ran, how often they succeeded, and a preview of what happened — without leaving the dashboard.

  • new Click any session card to expand it inline — no page navigation required
  • new Expanded view shows top tools used (e.g. "Bash 87 · Read 43 · Edit 31")
  • new Tool success rate shown as a color-coded badge (green ≥90%, yellow ≥70%, red below)
  • new Activity timeline preview shows the first 4 events from the session
  • improved Session cards now show tool invocation count ("· 187 tools") so agentic sessions are immediately visible
  • improved Non-main branches shown on session cards — feature branch work is surfaced at a glance
  • improved "View all →" link in Recent Sessions now goes to Insights instead of the raw Transcripts dump

v0.2.0

feature

Enterprise tier and team billing

Teams larger than 25 now have a path to PromptConduit with SSO, centralized billing, and audit logs. The pricing page reflects the full tier structure.

  • new Enterprise tier: custom pricing for 25+ seat teams with SSO and audit log requirements
  • improved Pricing page updated with a clear Local / Team / Enterprise breakdown
  • improved Team→Enterprise threshold set at 25 seats — below that, one manager can approve on a credit card

v0.1.0

launch

Public launch: capture, review, ship skills

PromptConduit is live. Install the CLI hook in under 2 minutes and your Claude Code sessions start flowing in automatically.

  • new CLI hook for Claude Code — installs in one command, zero config
  • new Dashboard with Recent Sessions and Skills preview
  • new Skills page: review AI-detected patterns from your sessions and save the ones worth keeping
  • new Insights page: tool usage breakdown, session frequency, and error patterns
  • new OAuth login with GitHub and Google
  • new Free local tier plus paid team plans at launch