Naseem 1.3.3

Naseem 1.3.3

• Token-aware agent — the agent now does exactly what you asked and nothing extra. Instead of running unrequested work (like testing an app you only asked it to open), it offers in one line and lets you decide — your tokens, your call.
• Includes everything from 1.3.2: vision for every model, verified file writes with rollback, memory insights, smoother streaming.

Naseem 1.3.2

Naseem 1.3.2

• Vision for every model — when your chat model can't see images, Naseem hands screenshots to a vision model and gives your model a full description, so screenshot-driven work just works.
• Accurate vision detection — vision support now comes from each provider's own catalog (Ollama Cloud, ChatGPT subscription), not guesswork.
• Protected file writes — every file the agent writes is verified and rolled back if it arrives corrupted, with automatic retry guidance.
• Memory insights — a new Memory section in Usage & Cost shows the app's live footprint and where it goes.
• Smoother streaming — replies repaint less often while streaming, easier on the battery.
• Smarter pause guard — no more false "I've paused" interruptions during simulator debugging, and pauses no longer dump step logs into the chat.
• Fixed: Choose Folder in New Project, editable project path.

Naseem 1.3.1

Smarter on big jobs

• Huge files handled properly — design exports, logs, and data dumps are split or queried instead of read endlessly, so the agent gets to work in a handful of steps instead of burning dozens.
• Picks up where it left off — on multi-step projects the agent keeps a status ledger of what's done and what's next, so "continue" resumes immediately instead of re-checking everything it already knew.

Coming soon: Rules

Scheduled agent runs — morning summaries, daily triage, repo chores — with approvals delivered to your phone via Telegram. Sneak peek in the View menu; shipping in an upcoming update.

Naseem 1.3.0

Suggested next prompt

After each reply, Naseem suggests your likely follow-up as ghost text in the composer. Press Tab to accept, edit if you like, Return to send — or Esc to dismiss. Suggestions come from the model in the same reply (no extra calls, no extra cost) and stick with the conversation across restarts. Turn it off in Settings → Agent.

Smoother simulator sessions

The iOS-simulator driver now recovers automatically if it goes quiet mid-session — a hiccup costs one short pause instead of failing every simulator tool until restart.

Fixes

• Switching a conversation's model while a reply is streaming now sticks — the next message always uses the model you picked.
• Hitting your subscription's usage limit now reports instantly instead of stalling ~30 seconds per message.
• Long agent runs can rediscover older tool outputs that were filed away mid-conversation.

Naseem 1.2.2

Naseem 1.2.2

Your images, right in the chat — plus code preview in the Files pane.

• Attachment thumbnails — images you attach now appear as thumbnails right in your message; tap one to open the original. No more guessing which picture a message carried.
• Code preview in the Files pane — Dart, Gradle, plists, Makefiles, and more now render inline instead of "No preview". Great for reviewing what the agent just wrote.

---

Auto-updates via Sparkle. Universal (Apple Silicon + Intel), Developer ID-signed and notarized.

Naseem 1.2.1

Naseem 1.2.1

A fast follow-up to 1.2.0, refining the new no-progress guard:

• Exploring is progress. The guard now recognizes that reading new files and running new searches is real work — long, multi-step builds proceed uninterrupted, and the agent only pauses to ask you when it's genuinely spinning (repeating earlier steps or hitting errors turn after turn).
• Tune or disable it any time in Settings → Agent → Pause if stuck after.

---

Auto-updates via Sparkle. Universal (Apple Silicon + Intel), Developer ID-signed and notarized.

Naseem 1.2.0

Naseem 1.2.0

See where your tokens and money go — and an agent that knows when to stop and ask.

Usage & Cost insights

• New Usage & Cost window (View menu) records every model call and charts your tokens and estimated cost over time — by day and by model, with a Tokens/Cost toggle.
• History is bounded: Settings → Agent → Keep usage history (default 90 days).

Stays on track — and knows when to stop

• Each multi-step task shows a goal chip in the header.
• New no-progress guard: when the agent goes several turns without real progress, it pauses and asks you — with the goal and what it tried — instead of spinning to the turn limit. Tunable/off in Settings → Agent → Pause if stuck after.

More accurate cost numbers

• Model prices are now sourced live from models.dev, covering far more models automatically (with an offline fallback). The status line labels cumulative output as "total out".

---

Auto-updates via Sparkle. Universal (Apple Silicon + Intel), Developer ID-signed and notarized.

Naseem 1.1.8

More readable, and more capable

Comfortable, adjustable text

Naseem's text is now larger and easier to read across the whole app — sidebar, settings, and conversation. Fine-tune it any time in Settings → General → Text size (Small to Largest).

See what the agent captures

• When the agent saves a screenshot to a file, it can now view that image with a vision model — so it can inspect an app window it just captured, not just save it.
• New query data (jq) tool: the agent queries a large JSONL/JSON file (logs, event histories) and gets back only the matching records — no loading the whole file into context.

Polish

• Search your settings — a pinned search box filters the General tab (type "font", "autonomy", "vision"…).
• Background commands are labeled honestly — a build shows as a background task that flips to finished, not a perpetual "running" dev server.

---

Existing users get this automatically via Naseem → Check for Updates.

Naseem 1.1.7

Bigger text, and a more accurate context readout

Adjustable chat text size

Find the conversation text too small? Settings → General → Chat text size scales it from Small to Largest. Headings, body, and code all scale together, so the layout stays clean at any size.

Accurate context on reopen

The context readout (used / total · %) now shows the model's real window the moment you reopen a chat — no longer waiting for your next message to correct itself.

---

Existing users get this automatically via Naseem → Check for Updates.

Naseem 1.1.6

Reliable delegation, accurate context sizing, project descriptions

Sub-agents that actually run

Delegating work to sub-agents now works on every provider — including ChatGPT-subscription and cloud models — where it was silently unavailable before. Ask the agent to research or build several things in parallel and it fans them out into focused sub-agents, then composes the results.

Knows each model's context window

Naseem now resolves the real context window for any model — from a live registry (models.dev), model-family knowledge, and a safe floor — so long-run behavior and sub-agents are sized correctly. New models just work, no update needed. The composer shows used / total · % for the context, colored amber then red as it fills; set a manual window per chat from ⋮ → Context window when you know better.

Project descriptions

Each project can carry a one-line description — type it via ⋮ → Description…, or let Naseem infer it from your first conversation. Shown in the sidebar and given to the agent as context.

---

Existing users get this automatically via Naseem → Check for Updates.