Skip to Content
Release notesRelease notes

Release notes

AgentFoundry (formerly CodeNow) ships in batched signed/notarized desktop builds plus continuous Vercel deploys for the website + admin.

VersionDateHighlights
v0.1.702026-08-01Agents you build actually run — and actually deploy. Three lifecycle dead-ends fixed. Start agent now launches the same interactive login shell as the embedded terminal, so it loads your shell profile — your PATH tools (bun, pyenv/python, Homebrew) and exported API keys are finally present; an agent that ran in the terminal or Cursor no longer dies under Start agent (the runner used a bare non-login shell). Graduate is real: an agent you build+save in Business mode → Build can now be promoted into your open project as a first-class Studio agent, unlocking the full Developer mode → wrap in TrustModel AGP → deploy → TrustScore path (Graduate was a “coming later” stub, so built agents had no route to deploy). And TrustModel’s Publish a policy button no longer 404s — it deep-links to that agent’s Control Plane. Verified by a new full-lifecycle E2E (build a social-media poster → Graduate → wrap AGP → deploy → govern). Mac signed + notarized; Windows + Linux published, Windows unsigned pending Azure. Gated on full smoke + all three E2E journeys.
v0.1.692026-07-28Guided agents actually answer your messages. When you build an agent by describing it, the card’s Send used to replay the build-time plan and ignore what you typed — so it felt like the agent “didn’t work.” Send now routes your message through keyless served chat with the agent’s persona, holds a real multi-turn conversation inline, and answers what you actually asked. Mac signed + notarized; Windows + Linux published. Gated on full smoke + both E2E journeys.
v0.1.682026-07-28Guided agents actually run — and leave real files behind. When you build an agent by describing it in plain language, two things are now true: (1) it scaffolds a real agent.py alongside the plan, so your project folder isn’t empty and the agent is real code you can open, fork, and graduate (before, it wrote only a hidden manifest — ls showed nothing); and (2) testing it runs keyless through the AgentFoundry engine instead of failing with “No agent.ts or agent.py found” — a described agent is a plan, not a key-carrying process, so Send now routes it to our served inference (no key to paste) and shows the result in the card. Bring-your-own-key “build from a kit” agents still run your agent.py directly. Mac signed + notarized; Windows + Linux published, Windows unsigned pending Azure. Gated on full smoke + both E2E journeys.
v0.1.672026-07-27The full IDE is now free for everyone. No trial clock, no expiry — multi-project, multi-model (BYOK), terminals, git, the agent Studio, and Live Replay are all free. The old “Day N of 30 · Upgrade” trial pill is now a calm “Free” badge, and the trial-expiry paywall is retired (a signed-in user always lands straight in the editor; lapsed trials on older builds are restored). Paid tiers return later only for served SMART-Code (BYOK stays free) and Business mode. Signing up is open — no invite or coupon needed (promo/invite links still apply automatically). Building an agent is smoother: the Test tab auto-detects your start command (agent.py, agent.sh, npm run agent:dev) and reconciles the framework picker instead of leaving an empty box; and SMART-Code no longer asks you to sign in when you already are (a stale token now refreshes silently). Mac signed + notarized (Apple Silicon + Intel); Windows + Linux published, Windows unsigned pending Azure. Gated on full smoke + both E2E journeys.
v0.1.632026-07-19Simpler agent-building · first-class Memory tab · Agent21 publish fixed. The agent-building surface is redesigned into a focused five-section Build panel — Anatomy · Prompts · Memory · Tools · Skills — one clear home per part of an agent, in the order you think about it; the Codebase-map card moved to the bottom of the Build nav. Memory is now first-class — its own tab in the Build panel (short-term budget, episodic→semantic consolidation, governed writes), front-and-center. Agent21 publishing fixed end-to-end: the app now requests the full agents:publish scope on connect, and refreshes your session first so an expired login no longer fails with “Invalid or expired token” (reconnect once to pick up the scope). The CodeNow → AgentFoundry rebrand is now complete across every remaining string. Mac + Linux signed + notarized; Windows unsigned while Azure validates. Gated on full smoke + E2E.
v0.1.622026-07-15Windows terminal fixed · Agent21 publish works · newest Gemini. The embedded terminal was blank for every Windows user (it launched a Unix shell that doesn’t exist there) — it now runs PowerShell via ConPTY; CLI/served agent runs on Windows are fixed too. Agent21 publishing now completes end-to-end — a dedicated first-party agentfoundry OAuth client + corrected authorize route + an Agent21-side root-cause fix, verified authorize→token mint (still one click, no secret). Google retired gemini-2.5-flash for new keys, so BYOK Gemini targets the current GA models — Gemini 3.5 Flash + Gemini 3.1 Pro. The CodeNow → AgentFoundry rename is now complete in-app (menu, dialogs, copy; bundle identity stays stable). Plus first-class TrustModel AGP + AgentCert (partner oversight + cert). Mac + Linux signed + notarized; Windows unsigned this release while Azure validates. Gated on full smoke + E2E.
v0.1.612026-07-14Simpler agent-building · first-class Agent21 · keyless studio. Building an agent is now one path — Describe → Build → Run: a single “what should your agent do?” box + one button lands you on a running agent; the 9-step wizard, SDK choice, and jargon move behind Advanced. The studio rail is decluttered (overlapping run tabs collapsed, groups hidden until you have an agent, “Build” → “Define”). Agent21 publishing needs no client secret — click Connect Agent21, approve, done (the credential lives on our backend). And the studio Run/Test default to CodeNow-served SMART-Code — zero keys to try your agent (BYOK still free for side-by-side). Mac + Linux signed + notarized; Windows ships unsigned this release while Azure Trusted Signing validates — a signed .exe replaces it on this same release shortly. Gated on full smoke + E2E.
v0.1.592026-06-25Mission Control artifacts + in-browser proof-of-work. Every finished mission now opens a Review drawer of typed, verifiable artifacts derived from the real run — Plan, Walkthrough (what actually happened, step by step), Files touched, Output, and Error if it stopped. And ▶ Verify in browser: when a mission builds something viewable, AgentFoundry drives the embedded browser to actually run the agent’s page and captures a screenshot as proof, attached as a verification artifact — the agent shows you it works instead of claiming it. With Mission Control (v0.1.58), this puts AgentFoundry at or above agent-first IDEs like Google Antigravity on the trust surfaces — plus multi-model BYOK, an own-your-harness runtime, and a Business mode a single-vendor IDE can’t match. All on real run records + real browser tools; no mocks. Signed + notarized, all platforms; gated on full smoke + E2E.
v0.1.582026-06-22Meet AgentFoundry · Async Mission Control. CodeNow is now AgentFoundry — the name caught up to an IDE built around agents, not the editor (in-app branding flips now; the macOS bundle + auto-update identity stay “CodeNow” until a later cutover so nothing migrates). The headline: a new Mission Control tab — dispatch tasks to agents and supervise them as they run, concurrently and asynchronously, with live Queued → Running → Needs review → Done columns, inline Approve / Deny on any externally-writing step, and real step results + TrustScore on completion. Our answer to Google Antigravity — except every model is yours (BYOK) and the harness is code you own. Plus fixes that bite daily: the constant “Session expiring soon” nag (two processes fighting one refresh token) is gone, building an agent no longer dead-ends on session_expired, the white theme is readable (user menu, GitHub icon, autonomy toggle), and embedded-browser navigation is hardened. Signed + notarized, all platforms; gated on full smoke + E2E.
v0.1.572026-06-20Agent memory substrate — the memory frontier, made real. The agents you build now get a managed memory substrate, not a junk-drawer log: short-term context budgeted like RAM (distilled facts first) with a memory.ts/.py adapter scaffolded into every agent; a background episodic→semantic consolidation loop that distills runs into durable insights (the engineering analogue of sleep, CodeNow-served — no key); and memory governed by default — provenance on every write, poisoning defense (agent-written text can’t smuggle instructions into a later run), and a TTL forgetting sweep that’s still auditable. All visible in the Foundry’s Knowledge & Memory step (live stats + one-click Consolidate/Sweep). Plus run-it-now (land on running your agent after create, not a file tree) and a zero-BYOK on-ramp (the describe path runs on CodeNow-served inference — no key to paste). Signed + notarized, all platforms; gated on full smoke + E2E.
v0.1.562026-06-20Agent Foundry, rebuilt as a guided step-by-step builder. The Foundry empty state is now a 9-step guided build — at every step it tells you what you’re building, which step is next, and lets you jump back to edit via a clickable progress rail. It quietly decodes the buzzwords you hit daily (harness engineering, context engineering, skills, knowledge/RAG, MCP, sub-agents) on the exact step where each lives. Step 1 makes CodeNow’s stance explicit — build and own your own multi-model harness instead of renting a one-click “managed agent” that locks you into a vendor’s sandbox and model — with honest Google ADK/Antigravity vs Anthropic Agent SDK guidance. The final step covers deploy & daily use: one-click Agent21 publish plus real, copy-to-terminal deploys to Google Cloud Run / AWS App Runner / Azure Container Apps, each with its actual free-credit program. No mocks: steps 1–2 create real code; steps 3–9 explain each block and route into the already-shipped surfaces. Renderer-only; signed + notarized, all platforms; gated on full smoke + E2E.
v0.1.552026-06-15One-click light/dark · no more stuck “session expiring” · clickable invite links. The Theme button now flips the whole app dark↔light in one click (live 🌙/☀️ label); editor color schemes moved to ⚙ Editor → “Editor theme…”, and that picker can finally be closed. The stuck “Session expiring soon” loop — looking signed-in while every server action fails — is fixed at the root: a genuinely dead refresh chain now resolves to a clean sign-out (one sign-in restores you), while a network blip never signs you out. Project-invite Accept invitation emails (and in-app Copy invite link) now use a clickable https /join link instead of a codenow:// deep link that webmail silently strips. Plus an honest Windows install card + “Run anyway” guide while code-signing is set up. Signed + notarized, all platforms; release gated on full smoke + E2E.
v0.1.542026-06-10Kimi & Qwen models · drag-portable Foundry agents · screenshots · cross-agent context. Open-weight coders Kimi K2 and Qwen (via Fireworks) are now first-class across SMART-Code, Cmd+K, autocomplete, and the multi-model Studio — CodeNow-served + metered, or BYOK with FIREWORKS_API_KEY. The Agent Foundry learned to drag an agent into another project and to auto-generate a run walkthrough (✨ Summary card). SMART-Code now reads screenshots (paste/drop an image into chat) and picks up where your other AI agents left off via cross-agent context continuity (~/.claude memory + transcripts + CLAUDE.md + claude-mem). New light theme. Fixes: the random sign-outs every few hours are gone, the SMART-Code chat no longer wipes your selection/scroll mid-render, file drops land in the right project, and in-app signup honors invites + promo codes. Release builds now gate on the full smoke + E2E suite before publishing. Signed + notarized, all platforms.
v0.1.532026-06-08Agents lifecycle rail · drag-a-file context · phone write-approval. The Agents panel’s 14 horizontal tabs become an ordered lifecycle rail — Anatomy → Run → Test → Runs → Trace → Evals → Tools → Prompts → Skills → Publish — so you always know where you are; a stale version string and duplicate nav are gone. Drag any file into the SMART-Code chat and it becomes a context chip that grounds the turn before the agent answers or edits — no more pasting whole files. From v0.1.52: approve the agent’s writes from your phone (race-both with the desktop modal — whoever responds first wins), a Desktop Relay keepalive fix so phone↔desktop sessions stop dropping mid-run, and viral promo codes (FOUNDERS10K = 4 months free). Signed + notarized, all platforms.
v0.1.522026-06-02Mobile loop closes — remote write-approval. The v0.1.51 “waiting for desktop approval” dead-end is gone: the agent’s write approve/reject prompt now reaches the mobile Session, race-both with the desktop modal. Desktop Relay gains a keepalive + concurrency guard so long phone-driven sessions stay online. Self-serve referral promo codes land at signup — FOUNDERS10K grants 4 months free. Signed + notarized, all platforms.
v0.1.512026-06-01Reliability — cross-lab failover · mobile Desktop Relay · grep_code · Anatomy. SMART-Code now fails over to the other lab when the chosen one is down or over-quota (a rate-limited or unfunded OpenAI transparently falls back to Haiku) — no more dead turns. The Cloud LAB pills finally pin the model: your Smart / Haiku / GPT-5 pick now reaches the server instead of being ignored. grep_code works again — it was reading the wrong field of the search response and always returned “No matches,” so the agent was editing blind; the advertised file-glob filter is honored too. The visual Anatomy card promotes to the full editor area instead of the squished ~300px sidebar. Scrollbars are visible app-wide (they were near-invisible dark-on-black) and the SMART-Code chat scrolls back to older messages. Under the hood: usage metering records correctly, BYOK is hardened to real sk- keys, and the mobile relay never strands a prompt (it shows a “waiting for desktop approval” notice when a write needs sign-off). Signed + notarized, all platforms. Full smoke green: static 0/0, runtime 229 pass / 0 fail / 1 skip.
v0.1.502026-05-31Mobile M1 — Desktop Relay. Drive your desktop’s live SMART-Code loop from your phone: send a prompt in the mobile app (codenow.pro/app → Session), your desktop claims it, runs the real agent loop, and mirrors the stream — tokens, tool calls, diffs — straight back to your phone. Write-approval stays on the desktop for now (remote approval is queued). Plus iOS Session composer fixes — the soft keyboard no longer covers the input, and the textarea + Send button are usable. Built on the v0.1.27 mobile PWA + the Mode-C long-poll relay; real Supabase realtime, no mocks. Signed + notarized, all platforms.
v0.1.492026-05-31Real MCP + CLI agent imports · Agents-panel workspace hardening. Business mode can now import two new kinds of agent for real — no stubs: an MCP server (CodeNow launches it over stdio and calls one of its tools as the agent) and a CLI / stdin program (your command receives the prompt on stdin or as an argument; its stdout is the response). Both run a real verification before they save. The whole Agents panel got a layout sweep: the Run / Test / Tools / Evals / Trace / Prompts / Publish panes and the Threads workbench now open in the full-width main editor area instead of the cramped sidebar where buttons were getting clipped; the agent chat scrolls properly through long conversations; the Import Agent wizard no longer dead-ends (its steps scroll and Next stays pinned, so you can reach Provenance + Verify); eight modals that could clip their own footers got scroll bodies; and 15 ambiguous dropdowns/radios gained clarifying tooltips. Also: Share → Save GitHub URL now persists (a column-name bug had every save failing). Signed + notarized, all platforms. Full smoke green: static 0/0, runtime 222 pass / 0 fail / 1 skip.
v0.1.482026-05-31Credit-based pricing · honest Fleet status · Agents-panel UX sweep. SMART-Code now meters in credits (1 credit = 1,000 tokens) with monthly pools — Developer Pro 2,000/mo · Business 6,000/mo · 500 free for 30 days. Credits only meter CodeNow-served inference; BYOK is always free and you’re never locked out (switch to your own key or top up). Agent Fleet drops the false time-based “stale” badge for honest lifecycle status (Draft / Tested / Published / Failing). The Agents panel is friendlier for newcomers — the empty state offers three clear paths (Build · Import · Fork, with Fork now first-class instead of buried in the marketplace), the Ask codebase-Q&A box is roomier with its controls regrouped and a fixed engine hint, and the Import wizard’s source-type layout is fixed. The codebase map (Understand-Anything) now opens full-screen in your system browser and self-heals a stale local server instead of refusing the connection. Guardrails persist for locally-built agents. TrustModel Edge Govern hardened — tracks real container health (no fabricated TrustScore) and re-hydrates governance state across reloads. Full smoke green: static 0/0, runtime 217 pass / 0 fail / 1 skip.
v0.1.382026-05-25SMART-Code grows up. Five items in one cycle: (1) write tools — edit_file / create_file / delete_file with an Approve/Reject diff modal (Monaco-style red/green); (2) the Confirm / Safe / Full topbar pills are finally wired — one pill click drives both per-project AI-CLI autonomy AND SMART-Code’s permission mode; (3) GPT-5 with tools — DEEP_INTENT prompts no longer silently fall back to Anthropic when tools are attached; (4) codebase semantic index — Voyage voyage-code-3 + Supabase pgvector for cloud, Ollama nomic-embed-text + per-project SQLite for local mode, file-watcher incremental re-index, new semantic_search tool replaces regex grep for natural-language queries; (5) routing transparency drawer — click the model pill on any answer to see which of the 6 signals matched, what evidence won, and why this turn routed to GPT vs Haiku. Speculative dual-execution deferred to v0.1.39. 165/166 runtime smoke pass.
v0.1.372026-05-25Focused patch. Trial pill no longer paywalls comped/lifetime users (renderer was computing trialDaysLeft from createdAt regardless of plan). Yellow “Native installation exists but ~/.local/bin is not in your PATH” warning gone — zsh shim now writes .zshenv so non-interactive sub-shells (which Claude probes) see ~/.local/bin too. App icon redrawn black-bg + white-shape so the Dock + menu bar stop hiding it inside macOS’s default grey rounded square. codenow.pro/launch/ 404 fixed with a 15-reasons landing page. Wiki: SMART Code Assistant promoted to top-of-left-nav, homepage sidebar restored, stale “CodeNow channels vs Slack channels” doc deleted (channels were ripped out in v0.1.28), Local-with-Ollama install flow documented end-to-end. 156/157 runtime smoke pass.
v0.1.292026-05-24Visible 7-day trial pill · Stripe wired end-to-end · every smoke green. Topbar pill shows “Day N of 7 · Upgrade” with green→amber→red escalation; click opens the live $9.99 Stripe Payment Link. The Vercel STRIPE_PAYMENT_LINK env var (empty for 10 days, breaking the paywall) now points at the real Stripe URL. All four previously-failing smoke cases (shareViewLink Refresh-session dead-click, browser.bottom-tab-renders, jira.studio-context-bridge, deploy.k8s-target-selector-renders) traced to one root cause + fixed: __cnSessionRefresh no longer auto-signs-out on failure, leaves the user on the page. 141 pass / 0 fail. Sidecar Docker :0.1.29 + :latest publish automatically alongside this release.
v0.1.282026-05-24CodeNow Sidecar (Mode C) ships · pair-programming + team-invite consolidation · Workforce humans (no channels) · auth silent-retry restored. Customer-cloud agents now receive Skillify + Guardrail updates via a tiny in-VPC daemon that long-polls codenow.pro and serves the agent over loopback. Docker image (karlmehta/codenow-sidecar:0.1.28), Kubernetes manifest, hardened systemd, Prometheus /metrics, k8s /livez + /readyz probes, audit-log forwarding. Drop-in sidecar.ts / sidecar.py shims for all four harness scaffolds. Team-invite stack collapsed into a single lifecycle-rail Share modal — old “Team members” modal + 5 endpoints + 4 Supabase tables deleted; net −742 LOC. New wiki pages: pair-programming + deployment-modes. 136/140 smoke pass.
v0.1.272026-05-19Browser control panel · Mobile PWA · JIRA→Studio context · K8s adapter · Phase 2 trust runtime · GH Action wrapper · GSD Phase 1 · Enterprise compliance. 139/139 smoke cases pass.
v0.1.222026-05-13Agent-card lifecycle bridges · Actionable fork errors · Skillify reads local manifests. Five new pill buttons on the Anatomy card pre-load manifest data into the Studio sub-tabs — Run in Studio, Test, Skillify, Edit manifest, View log (scoped to the agent’s runs). When an agent run fails, the conversation panel now shows a classified error inline (missing API key / dep / runtime / timeout) with a copy-paste fix command instead of a 3-second toast that vanishes. Skillify in Business mode now surfaces local filesystem-manifest agents alongside HTTP-imported ones — local skills land in <project>/.codenow/skills.json, tagged per slug. Scaffolder ships a zero-dep .env / .env.local loader so LLM-rewritten agent.ts files keep working. Born from dogfooding v0.1.21 by building a real multi-tenant SDR agent (Sophia) inside the IDE.
v0.1.212026-05-10Auth self-heal · Read-only share links · Mac Intel restored. tryRefreshToken is single-flight + recovers from terminal failures (refresh_token_already_used etc) by auto-clearing local auth and bouncing to sign-in — no more dead-end “Sign in again” panel. New Share view-only status-bar button mints anonymous URLs anyone can open in viewer mode; codenow://view/<token> deep link clones the repo and opens read-only. Mac Intel restored via single-runner cross-compile (--mac --arm64 --x64 on macos-latest); both DMGs ship from one job. Model Arena now shows an inline empty state instead of an invisible toast when no provider keys are configured. New popup dead-click smoke case + overlay-aware snapshots in the existing sweeps.
v0.1.202026-05-09Lab-pitch surfaces — Model Arena, provider transparency drawer, capability tuning, partner-publisher slot in Catalog, per-lab “Powered by” lockups in agent kits. Read-only viewer role plumbing (DB CHECK + Monaco/CRDT enforcement; surfaced as a one-click button in v0.1.21). Auto-update structured event hooks + per-event JSONL log. Voice cleanup. Intel Mac DMG did not ship — see v0.1.21 for restoration.
v0.1.192026-05-09Critical bugfix. Terminal lockup under heavy PTY output — frame-batched IPC drain (~16ms or 64KB), write-buffer backpressure, focus recovery on mousedown, Cmd+Shift+T reset hatch. CI hardening unblocked alongside (smoke lockfiles committed, artifact-quota dance replaced with direct release attach, codesign identity-count parser fix).
v0.1.182026-05-07JIRA full integration — connect, panel, transitions, comments, branch auto-pin, Send to AI, agent-draftable Create-ticket modal. Slack post-only via webhook URL with auto-detect format. Auto-update reliability — persistent log, boot-time install verification, recovery banner. Productivity bundle.
v0.1.172026-05-06Integrations rail foundation — AI / GitHub / Slack / JIRA tiles bottom-left, real IPCs (no mocks), two-pane panels for GitHub PRs/checks/issues + Slack channels/DMs/transcripts, beautiful integration modal.
v0.1.162026-05-05Onboarding tour + welcome screen + Create New Project + IDLE pill fix + comprehensive E2E integrations rail test suite with fixture isolation.
v0.1.152026-05-05Beta-feedback sweep. Editor essentials (auto-indent · JSX auto-close · autocomplete · TS language service · bracket guides). File tree right-click menu + + New file / + New folder buttons + .env* visible by default + Show hidden toggle. Agent scaffolder writes into a slug subfolder, auto-loads .env.local, sets per-kit preferred_model. Build New Agent picker recommends a kit based on detected API keys. Migration Assistant rebuilt with progress strip + skip-for-now + expandable “needs check” + collapsed already-configured. Claude Code tab shows install card when CLI missing + Pro/Max requirement. macOS application menu now has File + Terminal submenus. Sidebar tooltip polish. Terminal panel drag-resize + lazy spawn (no more empty terminal on boot). Live collab pill hidden when alone.
v0.1.142026-05-04Team feature, hardened end-to-end. Personalized invitations with First/Last name + greeting (“Hi Sarah,”). Members modal consolidates Members + Channels + Live coding session + Recent activity audit log into one surface. Server-side team_audit_log records every membership event; system messages post into channels on join/leave/remove. Trial users can invite (capped at 5 seats per project). Email FROM unified to no-reply@codenow.pro. Stale-session banner removed.
v0.1.132026-05-04Internal — built but never published. Fixes rolled into v0.1.14.
v0.1.122026-05-04Sign-in hotfix — keychain dual-storage fallback (kills the persistent “Your sign-in needs a refresh” banner) + banner-overlay fix (banners no longer paint over the topbar). Supersedes v0.1.11. Manual install required for v0.1.10/v0.1.11 users.
v0.1.112026-05-04Partial hotfix — included the keychain fix but missed the banner-overlay fix. Superseded by v0.1.12; do not install.
v0.1.102026-05-04Connections catalog restored (52 MCPs, was 2) · + Invite team in topbar · login screen show/hide password + Remember me · stale-session banner (universal “asking to login when logged in” fix) · voice MVP pulled, replaced by Slack/Discord/Zoom integration · promises-driven E2E smoke rig (92 promises tracked, 85 covered, 0 gaps)
v0.1.92026-05-03Threads sidebar (Zed parity) · per-project Autonomy toggle · embeddings codebase indexer · Tab autocomplete v2 · DefenseClaw scaffold-time policy · footer/topbar bug sweep
v0.1.82026-05-03Zed parity (Vim, real-time CRDT collab, voice, channels) · production primitives (prompt caching, channel-separated injection defense, Merkle audit trail, DefenseClaw runtime) · onboarding prereqs panel · prominent in-app update banner
v0.1.72026-05-03IDE essentials cycle — Cmd+P, Cmd+Shift+F, Cmd+Shift+P, @ mentions, Apply button, codebase Ask, tab autocomplete, source-control GUI + 4 beta-user blockers
v0.1.62026-05-02Cmd+K, Composer, Team Invites, file associations, signup credits, auto-update via Vercel proxy
v0.1.52026-05-02Smoke rig, prompt() fixes, HubSpot/Salesforce, Card/Code toggle
v0.1.42026-05-01Live Agent Replay, harness kits, 2026-Agent Anatomy card, signed DMG

The IDE auto-updates on launch (every install in v0.1.6+ checks GitHub Releases via the Vercel proxy). For v0.1.5 → v0.1.8 you have to install once manually from codenow.pro/download ; subsequent versions auto-update.

Last updated on
Install issuesv0.1.68 — Guided agents actually run (and leave real files behind)