Switch Multiple Claude Code Accounts, Track Token Usage, and Manage AI Memory in One Workspace

Most developers running Claude Code, Codex, or Gemini CLI eventually hit the same three walls. The first is account juggling: your personal Claude Max login is not the one your employer pays for, but the CLI only holds one set of credentials at a time, so switching means logging out, logging in again, and re-authenticating in the browser every single day. The second is invisible cost: tokens and model spend stack up inside session files you never open, and by the time a rate limit hits you have no view of where the budget went. The third is forgotten memory: every agent stores its long-term context separately, scattered across ~/.claude, ~/.codex, ~/.gemini, and project-specific subdirectories, with no central way to search, edit, or share them.

1DevTool v1.16.0 turns all three into first-class UI. The new Accounts tab holds multiple OAuth logins per agent behind the OS keychain. The new Usage dashboard breaks down tokens and cost per agent per model, reading directly from the session JSONL files on disk. The new Memory Manager reads every agent memory in every project and lets you browse them as a list or an interactive graph, with one-click export to an Obsidian vault.

Switch AI Accounts Without Logging Out

If you have ever run claude /logout followed by claude /login just to copy a conversation from your personal account to your team account, you know how much friction a single browser round-trip can add to an AI coding workflow. v1.16.0 replaces that loop with a dedicated account registry under Settings → AI → Accounts. Each agent card — Claude Code, Codex, Gemini, Qwen — holds a list of saved logins that you can switch between with one click.

There are two paths into the feature. Add account opens a guided terminal pre-running the selected agent's login command; you finish the OAuth flow in the browser, and 1DevTool writes the new credentials into an encrypted snapshot as soon as the CLI reports success. Save current account snapshots whatever credentials are live right now, so if you were already logged in when you installed the release, one click turns that login into a reusable, labelled entry ("Personal", "Acme Team", "Client X") without another browser trip.

Encrypted by Your OS Keychain

The account vault lives under ~/.1devtool/ai-accounts/ with per-agent subdirectories (claude/, codex/, gemini/, qwen/). Each saved login is stored as a {id}.enc file encrypted by Electron safeStorage — macOS Keychain on Mac, Credential Vault on Windows, and libsecret or kwallet on Linux. A registry JSON keeps a backup of the previously active credentials under __previous__.enc, so a failed switch can roll back without leaving the CLI stranded between two accounts.

The practical upshot: tokens never leave your machine, and you do not have to invent a secret-management story on top of your AI subscriptions. The credentials still load from the same paths each CLI expects on disk. The vault is just a safe staging area.

Drift Detection When a CLI Login Changes Behind Your Back

If you log in from a terminal outside 1DevTool — say you run codex login in a plain shell — the app notices the live credentials no longer match the saved snapshot and surfaces a Signed in · not saved hint next to the agent. One click labels and saves the new login so it joins the vault. Without this, credential drift is invisible until you break something; with it, the UI tells you where the ground truth diverged.

The Accounts tab also exposes a Switch Account entry in the status bar AI dropdown that jumps straight to Settings → AI → Accounts, because the common case is not "open settings and navigate three clicks deep" — it is "I want a different login right now."

Claude Code Token Usage and Cost Tracking

The second headline feature is the Usage sub-tab under Settings → AI. Until now, if you wanted to know how many tokens your last week of AI coding actually consumed, you had two options: paste a transcript into an external counter, or trust the CLI's own reporting and hope it covered everything. v1.16.0 replaces that with a dashboard that reads the session JSONL files each agent writes locally — the same files the agent itself uses as the source of truth.

The result is a per-agent breakdown of sessions, input tokens, output tokens, cache-read tokens, and total cost in USD, further decomposed by model. Cache hit percentage is computed as cacheReadTokens / (inputTokens + cacheReadTokens), so you can see how much of your spend Claude's prompt cache is saving. All of it is real usage, not estimates.

Codex 5-Hour and Weekly Rate-Limit Windows

Codex users get an extra panel. The Usage tab renders both the 5-hour window and weekly window Codex enforces on its cloud plan, showing current utilization and reset times side by side. When you hit a rate limit at 3pm mid-sprint, this is the difference between "I have 90 minutes of headroom" and "I should switch to a different agent for the rest of the afternoon."

Performance-wise, the dashboard keeps a per-file mtime cache, so even if you have months of session history across multiple agents, reopening the tab loads in milliseconds. Filter presets cover All time, Last 7 days, Last 30 days, Last 90 days, and a custom date range, and the current filter persists when you close and reopen the sub-tab.

AI Memory Manager with a Graph View

The third new feature is the most unusual. Every AI agent now keeps long-term context in Markdown memory files — Claude's CLAUDE.md, Codex's notes, Gemini's, and project-scoped memories under ~/.claude/projects/<encoded-project>/memory/. Over a year of real use, those memories grow into the developer equivalent of a second brain: past decisions, style rules, incident postmortems, one-off preferences. But they are scattered across four CLIs, dozens of projects, and hundreds of files you rarely open directly.

1DevTool v1.16.0 adds a Memory Manager panel that reads all of them in one view. You can filter by agent and by project, search across contents, and open any entry in a Markdown editor with autosave — no hand-editing files on disk, no worrying about frontmatter syntax. The manager parses YAML frontmatter (name, description, type) using a lightweight, dependency-free parser and indexes cross-references — [text](file.md) links between memory files — into a relationship graph.

List or Graph, Your Choice

The List view is what most developers will live in: a searchable, filterable table with a 240-character preview per entry. The Graph view is the interesting one. It uses Cytoscape with multiple layouts — concentric (hub-and-spoke memories that reference one central doc), cose (dense cross-reference graphs), breadthfirst (deep chains), and grid (flat catalogs). The app auto-selects a layout based on entry count and edge density, and you can zoom and pan freely without lag even on hundreds of nodes because the Cytoscape bundle is lazy-loaded only when you open the tab.

If you use Obsidian, the panel has a first-class export. Pick an Obsidian vault folder and 1DevTool writes every memory entry as a Markdown file, preserving per-agent and per-project folders so your AI memories become browsable, linkable, searchable notes alongside the rest of your second brain. For entries already inside a vault, an Open in Obsidian deep link jumps straight to the matching note.

Agent Input Personas and Other Polish

The agent input overlay (⌘I) now has nine writing personas: Lowkey, Coder, Blogger, Novelist, Journalist, Poet, Vibe Coder, Researcher, and Marketer. Each ships with its own font stack, color palette, and chrome accents — scanlines for Vibe Coder, halftone for Poet, dotgrid for Researcher, conic-gradient stripes for Marketer — implemented as CSS overlays with mix-blend-mode so there is no runtime cost. Pick your favorite once and it persists across restarts.

The rest of the release cleans up three smaller things. Settings → AI is restructured into Storage, Usage, and Accounts sub-tabs instead of a single long scroll. The HTTP Client now remembers request/response split and sidebar widths between sessions, stored per project in the layout store. And Vercel deploy's Node-version fallback searches more install locations and logs the exact Node binary it picked and why, which turns a silent fallback into an auditable one.

Before vs After

Who Benefits Most

Try It Today

v1.16.0 is a release about taking things that were previously invisible — extra accounts, real token spend, long-term memory — and making them directly browseable. If you are on Claude Code or Codex today, the first five minutes after installing are worth a small experiment: open Settings → AI → Usage to see the last 30 days of your actual agent spend, then open the Memory Manager to see how many memory files you had no idea existed.

Download the latest build from 1devtool.com, save your current logins as named accounts, and find out how many tokens you burned while you were not looking.