Changelog

The usage page now breaks down activity by agent, model, and skill so you can see exactly where your spend goes. A new date range picker, downloadable billing data, and live cost cards round out the overhaul.

Agent-level stats

A new Agent stats tab inside Settings → Usage shows runs, runtime, tokens, and cost per agent. Deleted agents appear with a short durable ID so historical spend is never orphaned. Non-agent usage is grouped under Other usage.

Better date controls

The date range picker now uses a calendar widget with presets and draft-then-apply behaviour. Pick Since or Through independently for one-sided filters, and zoom from day to month to year with continuous navigation.

Spend at a glance

The Budget card now shows real-time running cost during a session, and a spend-by-model donut chart replaces the old bar chart so you can compare providers at a glance. Skills used over time gets its own chart, and the activity log table supports Excel-style column filtering and ordering.

Download your data

A Download CSV button on the billing page exports your billing records for your own analysis or accounting.

Agent management has been refined across the board — scheduled wakeups are visible at a glance, the channel setup dialog has been redesigned, and external chat messages now look like real conversations.

Wakeups at a glance

The agent rail now shows active and paused scheduled-wakeup counts directly in the At a glance section, with clear labels for routine heartbeats and global pause state.

Redesigned channel management

The Manage dialog has been restructured into clear sections for Slack channels, general settings, and capabilities. Slack channel bindings show live status with setup checklists and sender profile photos on inbound messages. The Channels overview follows the agreed wireframes with connection status, bot channel availability, and bound-channel lists.

Who's online

A workspace presence facepile in the page header shows up to three avatars with a green dot for online members. Click it to see the full Online and Away lists, with deep links into shared sessions.

Chat surfaces that look like chat

Messages delivered to external channels (Slack, email) now render as proper markdown in the web app — bold, italics, links, and lists — so the in-app echo matches what the recipient sees. Agent sessions are visually distinguished in Recents with a distinct card style, and the conversation provenance header clarifies whether a message came from the web or an external channel.

Deep research has been redesigned from the ground up. Sources are now interactive panels, results filter in place, and a new platform skill produces publication-ready research reports.

Interactive source panels

Sources panels now render as true hypermedia — click a source to open it inline, browse its contents, and navigate back. A favicon proxy shows the real site icon next to every source so you can scan by domain. Filtering is built into the panel toolbar, letting you narrow by source type, recency, or relevance without leaving the result view.

Research-report skill

A new platform skill, research-report, ships with templates and reference docs for producing editorial-grade research reports in HTML — hand-built charts, scrollytelling, and bespoke typography. Give it a question and it plans search angles, reads the top sources, extracts quote-backed claims, and delivers a cited report.

Pick up where you left off

Stopped research sessions now recover their progress. If a run is interrupted, the agent picks up from the last completed step instead of starting over.

Agents can now reach external APIs on your behalf — scoped, auditable, and controlled from the workspace settings.

Allowlisted network access

Add external hosts to a network-access panel in workspace settings. For each scope, specify the host, port, and protocol. Agents can then make HTTP calls to those endpoints using their standard tools, and every outbound request is logged against the workspace audit trail.

Built for safety

Network access is disabled by default. You enable it per scope with explicit host and port, and the access list is visible and editable in the workspace. Agents never receive an allow-all — they see only the endpoints you have approved.

File previews in the artifact panel are now consistently styled, more faithful to the source format, and capable of rendering diffs and tabular data inline within chat messages.

Themed office previews

XLSX and DOCX previews now match your light or dark theme. DOCX tables render with the column widths the source document uses, and pages respect the sheet dimensions so you never see a table spilling off the edge. XLSX previews preserve cell styling and column proportions from the original workbook.

Diffs and tables in chat

Code diffs and tabular content now render inline inside chat messages instead of as plain text blocks. A new filterable service panel lets you browse and select MCP services directly from a searchable list without leaving the conversation.

Smarter scrolling

Long DOCX tables scroll horizontally inside the panel instead of stretching the page, and the artifact resize handle is now easier to grab with a wider hit area. Collapsed panels restore their width when you open a new file.

The agent experience has been rebuilt from the ground up — a native inspector, redesigned management, and external conversations rendered directly in your chat stream.

Agent-native inspector

Open any agent and the inspector opens alongside it — a dedicated panel for context, files, connectors, and skills that stays in sync with whatever the agent is doing. No more switching tabs to see what an agent has access to.

Redesigned management

Agent settings now use vertical tabs for clean navigation at any screen width. The management page shows connector context at a glance so you always know which services the agent can reach.

External interactions in chat

When an agent talks to someone in Slack, those conversations now render natively in the d5s chat stream — message bubbles, attribution, and deep links — so you can follow the full thread without leaving d5s.

Recents flyout

When the sidebar is collapsed to icons, a new Recents flyout shows your starred projects and recent sessions with status dots and activity badges, so you can jump back in without expanding the sidebar.

Every agent now has its own durable inbox and can hold conversations outside d5s. A shared work-cycle scheduler keeps inboxes, chats, dashboards, and automations flowing without collisions.

An inbox that persists

Agents maintain a persistent inbox across sessions. Messages, notifications, and action items queue up reliably — the agent processes them in order and never drops work between runs.

Talk to agents where you already work

Agents can now hold conversations through external chat surfaces. When someone messages an agent through a connected channel, the conversation appears in the agent's thread just like any other message — and the agent's replies show up in-channel with attribution and a deep link back to d5s.

One scheduler for everything

A new work-cycle dispatcher coordinates all agent workloads — inboxes, chats, dashboards, and automations — so concurrent runs never collide and every item gets its turn.

Three new model families are now available through the model gateway, each with documented context windows and reasoning capabilities.

GPT-5.6 joins the gateway lineup with expanded context and faster inference. Muse Spark 1.1 brings strong creative and analytical performance. Grok 4.5 rounds out the additions with deep reasoning support.

All three are available immediately in Settings → Models and the session composer — no separate API keys needed.

Give agents a presence in Slack. Connect an agent to a channel, and it responds when @mentioned, posts updates on its own, and guides you through setup if it is mentioned in a channel it has not been bound to yet.

One bot, many channels

Install the d5s Slack app once per workspace — one shared bot that every agent uses. From an agent's Channels tab, pick which Slack channel it should listen to. Each channel maps to exactly one agent, so your standup agent lives in #daily and your on-call agent lives in #incidents.

Proactive and reactive

When someone @mentions the bot in a bound channel, the agent replies in-thread with a bold attribution header and a deep link back to d5s so your team can follow the full conversation. Agents can also post proactively — a heartbeat check-in, a completed report, or an alert — without waiting for a mention.

Guided setup

Mention the bot in an unbound channel and it posts a handoff message with a link that opens the agent's Channels tab, prefilled with the right Slack team and channel IDs. A setup checklist shows your connection status, bot channel availability, and which live channels are already bound — so you know exactly what to configure next.

The skills surface is now production-ready, with editing, version history, and full file trees for every platform skill. Three new platform skills — a real browser, PDF generation, and DOCX editing — are available right now.

Browse and edit skills

Open any skill from your workspace settings to see its full file tree — every script, template, and reference document the skill ships with. Click Edit to open a versioned editor where you can modify the skill body and description. Every edit creates an immutable version, so you can always see what changed and roll back if needed.

Version history

The View History button shows every past version with change summaries. Select any older version to read it in a read-only view with a banner linking back to the latest. No more guessing what a skill used to say.

Three new platform skills

The platform skill catalog now includes a browser automation skill (use the agent to navigate the web, fill forms, and scrape pages), a PDF generation skill (create, merge, and render PDFs with full typography control), and a DOCX skill (generate and read Microsoft Word documents). Each ships with its own templates and reference docs ready to use.

You can now branch any conversation to try a different approach without losing your original thread. Think of it like a fork in the road — explore one path while keeping the other intact.

How it works

Hover over any assistant message and click the split icon. A new session is created that copies everything up to that point — messages, file uploads, and live artifacts — and a clickable divider shows exactly where the branch begins. Your original conversation continues untouched.

Why it matters

When you're working through a complex problem, you often want to try two approaches in parallel: ask a different model, give different instructions, or explore a creative tangent. Session branching lets you do that without copy-pasting context or starting from scratch.

Chat message timestamps now show the calendar day and month alongside the time. In long-running conversations that span days or weeks, you can see at a glance when each message was sent — no more mental math from relative timestamps.

The format adapts to your locale and appears on every message row in the chat stream. Pending messages that have not been sent yet continue to render without a timestamp, keeping the interface clean during active typing.

d5s.tech has a fresh new look. The landing page has been redesigned around a bento grid that puts the product front and centre, with industry-specific messaging and a smarter Ask AI section.

Bento grid layout

The new layout uses a modular bento grid — clean, modern, and responsive. Every section tells a focused story without crowding the page. The design works across desktop, tablet, and mobile with the same visual impact.

Speak to your industry

A new sector rail lets you filter the page by industry — engineering, law, medicine, research, and more. Pick your field and the messaging, examples, and case studies retune to match. It's the same product story, told the way your team needs to hear it.

Ask AI in the footer

The footer now includes an Ask AI prompt — type a question and get a direct response about what d5s can do for you, right there on the landing page.

22 new models are now available in the agent catalog, including the latest frontier releases on existing providers and first entries from ten new model families — all routable through the same gateway with no configuration.

What's new

Frontier additions include the newest Claude, Gemini, and DeepSeek variants, plus first-class entries for models from Alibaba (Qwen), ByteDance (Seed), Cohere (Command A), Meta (Llama 4), Mistral, Moonshot (Kimi), NVIDIA (Nemotron), Xiaomi (MiMo), Amazon (Nova), and MiniMax. Every model arrives with its documented context window and reasoning capabilities so the agent loop can route to the right tool for each task.

Workspaces stay curated

New models are available but not auto-enabled. Your workspace defaults to its existing curated set. Admins can add any of the 22 from the Models settings page when you're ready to try them.

Agents are now a first-class primitive in every project — autonomous, persistent, and always on. Give an agent a persona, configure how independently it should act, and it will check in on a heartbeat schedule, respond to wake triggers, and maintain a single continuous conversation that compacts instead of ending.

A home in every project

Every project now has an Agents tab. Create an agent, give it a persona and instructions, and it gets its own always-promoted home project and an eternal session. The agent's conversation lives inside the Agents view and compacts automatically when the context window fills up — it never ends, so the agent remembers everything.

Autonomy you control

Set a daily budget and a heartbeat interval, and your agent will wake on schedule to check in. Use the Wake button for an immediate check-in any time. The agent's autonomy section shows its configuration at a glance, and every heartbeat turn appears as a labeled line in the conversation so you can see exactly when and why the agent acted.

Persona-driven behaviour

Persona and autonomy settings are gated behind feature flags, visible in the right context panel. Choose from curated defaults or write your own — the agent's personality and autonomy level shape how it thinks and acts across every session.

The model catalog just got more useful. Every model family now has a distinct brand icon, models are sorted newest-first so you see what's fresh, and a NEW badge highlights recently added entries.

Brand icons you recognize

All 16 model families now show real brand marks — from OpenAI and Anthropic to newcomers like Qwen, Mistral, and Kimi. In Settings → Models, the icons render in full colour so you can scan the list at a glance. The composer picker keeps monochrome icons for a clean, focused look.

Smarter ordering

Models are now ranked by family, with the newest entries at the top of each group. A NEW badge appears on any model added in the last 30 days so you can spot what's available without checking release dates.

Change with confidence

Switching models or reasoning effort mid-conversation now shows a confirmation warning if the change would invalidate cached context, so you know the trade-off before you commit.

The composer "+" menu now lets you pick which skills and connectors your agent can use — per conversation, without changing project settings.

Skills at your fingertips

Open the Skills submenu to see every skill installed in your workspace. Toggle one on and it becomes available to your agent for the current conversation. Turn it off and the agent will not use it. Your choices appear as removable chips above the message input, so you can see at a glance what's enabled and change your mind with a click.

Connectors, for every team member

The Connectors submenu is now available to all members — not just admins. Every credential in your workspace's vault appears as an on/off toggle for the current conversation. Disconnected credentials are listed but disabled. Admins still manage which connectors are installed and where they point; everyone else gets to decide whether to use them in a given conversation.

The d5s.tech landing page has been refreshed with a new impact hero section, personalized role examples, and a connector context strip that shows real integrations flowing through the product.

What's different

The hero now leads with impact across seven professional roles — from engineering and law to medicine and research — with role-specific copy that retunes the entire page. A new connector context strip animates real connector logos in a seamless loop, showing the breadth of tools d5s connects to.

Between the Security section and the FAQ, a new "humans + agents" fanned card section shows the agent card as a peer in a team huddle, reinforcing the idea that agents work alongside people, not instead of them.

We've made connectors smarter and the project overview page more useful.

Agents can suggest connectors

Your agents can now search the connector catalog and suggest the right one for the task at hand. When an agent identifies a connector that would help, it proposes a Connect card — review it and click to connect, without ever sharing credentials with the agent.

Connectors grouped by brand

The installed connectors page now groups credentials by brand. If you have multiple connections to the same service — like two Gmail accounts or both OAuth and token-based access to Notion — they appear as a single brand card that expands to show each connection with its method label.

A redesigned project overview

The project overview page has been rebuilt with compact chat rows, frameless dashboard previews, and a "View all sessions" popup that lets you browse and search every conversation in the project at a glance.

d5s is now a desktop app. Download it for macOS, Windows, or Linux — the same workspace you know, now with a native title bar, dock presence, and a dedicated download page that always has the latest version.

Native feel, web speed

The desktop app wraps the web frontend in an Electron shell with a custom title bar that matches the d5s look. The app keeps its own menu bar, handles deep links directly, and sits in your dock or taskbar like any other app. Because the UI is the same web frontend, every feature that ships to the browser is available on desktop the same day.

Self-updating distribution

Downloads live at d5s.tech/download. The page auto-detects your platform and offers the right installer. Every release is published through an automated pipeline so the download page always serves the latest version with no manual steps.

You can now control every tool an MCP connector exposes — set each one to allow, ask for approval, or deny — directly from the connector settings panel.

How it works

When you open an MCP connector's settings, you'll see every tool the server advertises, grouped by read and write operations. Each tool has a dropdown: allow (runs without asking), ask (your agent pauses and requests your approval at runtime), or deny (the tool is unavailable). Defaults are sensible — read tools default to allow, write tools default to ask — and you can change them at any time.

Why it matters

With nearly 300 connectors in the catalog, most of them MCP, a simple on/off switch is not enough. Tool-level permissions let you give your agents exactly the access they need — reading your calendar but not creating events, searching your documents but not deleting them — without blocking entire connectors.

Every collaborator now has a face. Upload a profile photo or pick from a set of generated avatars, and your image appears everywhere your name does.

Where you'll see them

Avatars show up in the chat message stream, the members table, the share dialog, and in your own sidebar trigger — replacing the generic Settings row with your name and photo. The chat author slot now renders both your avatar and a coloured initials fallback, so every message is immediately attributable at a glance.

Easy to set up

Visit Settings → General → Profile to open the avatar editor. Choose an auto-generated avatar, upload a photo from your device, or stick with your initials. The editor validates image format and content type before saving, and your avatar updates everywhere in real time.

Connect your Outlook account to let agents read your email and calendar. The new Outlook connector uses Microsoft Graph to give agents read access to your inbox and schedule, bringing your real-world context into every project.

What agents can do

Once connected, your agents can read your Outlook mail and calendar events. They can summarise your inbox, find meeting details, prepare agendas, or cross-reference your schedule with project timelines — all without leaving the workspace.

Secure, read-only access

The connector is scoped to read-only access. Agents can see your mail and calendar to provide context and assistance, but they cannot send, delete, or modify anything in your account. You control which projects can use the connector, and you can disconnect at any time.

Agents can now run deep research — a multi-step pipeline that plans search angles, reads sources, extracts quote-backed claims, and mechanically verifies every finding against the original page before delivering a cited report.

How it works

Toggle the Research option in the composer or ask your agent to "do deep research on..." a topic. The agent runs a deterministic pipeline: it plans search angles, searches each one, reads the top sources, verifies quotes against the fetched pages, and delivers a report with inline citations. A live activity card tracks progress step by step so you can see the pipeline running.

Every claim in the final report is backed by a verbatim quote from a verified source URL. You can click through to any source to confirm it yourself — no hallucinations, no fabricated citations.

Built for trust

The verification step is mechanical, not conversational. Each quote is checked against the page it came from before being included. If the verification fails, the claim is flagged rather than silently included. This means every finding in a deep-research report has been programmatically confirmed.

Agents can now draft emails directly in chat. When you need a message written — a client follow-up, a team announcement, or a careful reply — your agent can propose one or more variants and you can edit them inline before sending.

Pick a variant, then refine

Each draft card shows up to three distinct approaches as pill tabs at the top. Switch between them with a click, then edit the subject and body directly in the card. A copy button and a revert button sit at the bottom of the card so you can capture the version you want or undo accidental edits.

Plain, portable text

Drafts are plain text only — no Markdown, no formatting — because email clients render bodies verbatim. What you see in the card is exactly what arrives in the recipient's inbox.

The Usage page has been rebuilt as a tabbed record that gives you a complete picture of how your team is using d5s — not just the current billing period.

Three tabs, one date range

A shared date-range control scopes all views. The Team tab shows your credit balance, token usage, an over-time chart, and spend broken down by model. The User stats tab gives you a per-member rollup of runs, agent runtime, tokens, and cost. The Activity log tab is the raw ledger, paginated and filterable by team member.

Export when you need it

Every tab supports CSV export scoped to your selected date range and filters. Download the data you need for billing, analysis, or your own internal reporting.

We've polished several areas of the workspace to make it more intuitive and readable. Search is more accurate, results look better, and navigation is smoother across the app.

Better Web Search Results

Web search sources now render as clean, bordered cards that are expanded by default. We've added domain-letter badges for quick identification and ensured your privacy by avoiding third-party favicon requests.

Accurate Search

Global search (Cmd+K) now correctly filters your results to show only promoted projects. This keeps your search results clean and focused on your most important work, matching the organization of the rest of the workspace.

Smooth Navigation

We've added proper scroll regions to all list pages—including Projects, Dashboards, and Automations. No matter how many items you have, you'll always be able to scroll through every card and row with ease.

The sidebar Recents list is now live and more interactive. We've added chat actions directly to the sidebar and ensured it updates instantly as you work.

Instant Updates

Starting a new chat now populates the sidebar immediately—no more waiting for a refresh. We've also added a live activity dot that lets you know if an agent run is in progress, finished but unviewed, or viewed.

Sidebar Actions

You can now rename, move, or delete conversations directly from the sidebar's action menu. This makes organizing your workspace faster and more intuitive, allowing you to manage your chats without leaving your current view.

Agents can now propose scheduled automations for you to confirm. Instead of manually setting up recurring tasks, your agents can identify opportunities for automation and offer to set them up for you.

One-click Setup

When an agent identifies a task that should be run on a schedule—like a daily summary or a weekly report—it can present a "Propose Automation" card. Review the instructions and the schedule, then click "Create" to turn that workflow into a recurring job with a single click.

Intelligent Configuration

This is part of our commitment to making agents more than just chat partners—they are active collaborators that help you configure your workspace to be as productive as possible.

Mention @d5s in any Slack channel to get a response from your agent. d5s now supports inbound Slack mentions, allowing you to bring your agents directly into your team's conversations.

How it works

Connect your Slack workspace using the Slack connector and designate a project to handle inbound mentions. When you tag @d5s in a channel, the agent resolves the thread, processes your request, and replies directly in Slack. Follow-up mentions in the same thread continue the conversation, maintaining context just like a chat in the d5s workspace.

Agent Persona

The agent uses the instructions, skills, and memory of your designated project. This means you can have a specialized agent—like a research specialist or a technical helper—ready to support your team exactly where they already work.

The Slack integration just got a lot more useful. Agent replies now render with proper Slack formatting, files and artifacts are uploaded back into your threads, and sessions get descriptive titles instead of "Untitled."

Proper formatting in Slack

Agent replies now use Slack's native mrkdwn formatting instead of raw Markdown. Headings, bold text, lists, and links all render correctly in Slack, so agent responses are readable at a glance — no more hash marks and asterisks cluttering the thread.

Files land in the thread

When an agent produces an artifact — a report, a chart, or a document — it is automatically uploaded into the Slack thread alongside the text reply. Your team gets the full output where the conversation is happening, not just a prompt to switch to another tool.

Conversations have names

Slack threads are now titled from the first message rather than showing "Untitled" everywhere. A quick scan of your Recents list tells you which thread is which.

Agents can now search your project's past conversations and read earlier discussions. Instead of starting cold on every new chat, your agent has continuity — it can recall what was decided, what was tried, and what context already exists.

Two-step recall

First, the agent searches for relevant past chats by keyword. Then it reads the transcript of the most promising one. This lets it pick up context you have already established — decisions, preferences, technical details — without you having to retell the story.

Scope and safety

The search is scoped to the current project. An agent in one project cannot see chats in another. Raw conversation history stays within the team; durable facts should still be saved to project memory for cross-session use.

A new context-window gauge sits next to the model picker in the chat composer. It shows how much of the model's context window your conversation is using, so you know when to wrap up a long thread or start fresh.

One glance is enough

The donut ring fills as your conversation grows: neutral below half, amber in the middle, red when you approach the limit. A tick marks the compaction threshold so you can see when the agent will start summarising older messages. The donut appears after your first turn and updates automatically with every exchange.

Works across every model

The gauge adapts to each model's actual context window — whether it's a model with a documented limit or one with a generous default. No configuration needed.

We've added several features to help you keep your workspace organized and distraction-free. You can now delete old chats, navigate a cleaner sidebar, and find exactly what you're looking for faster.

Delete old chats

The Chats list now features a delete action. If a conversation has served its purpose, you can remove it with two clicks. This also cleans up the associated resources, keeping your workspace lean.

A focused sidebar

The sidebar Recents list has been flattened into a single, slim list. We've removed the per-project headers that were cluttering the view, giving you more vertical space for your actual conversations. Starred projects still appear at the top for quick access.

Loose chats vs. Project chats

The Chats tab now focuses exclusively on "loose" conversations — those not explicitly promoted to a project. If a chat belongs to a project, you'll find it there, making it easier to separate quick one-offs from deep project work.

Chat messages now support beautiful LaTeX math rendering. Whether you're discussing complex algorithms, scientific research, or financial models, mathematical notation now appears exactly as it should.

Beautifully rendered

We use KaTeX to render math equations in real-time. Simply use standard LaTeX delimiters ($...$ for inline, $$...$$ for blocks), and the workspace will format them into clean, readable notation.

Technical discussions made easy

This is part of our commitment to making d5s the best place for technical teams to work with agents. From physics to finance, you can now communicate precise mathematical ideas without clunky workarounds or screenshots.

You can now connect your Discord workspace to d5s with a single click. Our new Discord connector uses OAuth, so you can sign in and authorise access without manual bot token configuration.

Agent capabilities

Once connected, your agents can interact with your Discord servers. They can read channel history to gather context, post updates to keep your team informed, and search through past discussions.

Secure and scoped

The connector supports both a one-click OAuth flow for convenience and manual bot credentials for full control. Permissions are scoped to exactly what the agent needs to be productive on your server.

Agents can now search the live web for current information. Instead of relying solely on training data, your agents can pull in the latest facts, news, and documentation to give you accurate, up-to-date answers.

Clickable sources

When an agent performs a web search, the results are rendered as a concise answer followed by a list of clickable sources. You can verify every claim by clicking through to the original page — providing transparency and trust in the agent's output.

How to use it

Web search is available as a tool in the chat composer. You can toggle it on or off per message, or let the agent decide when it needs to reach for the live web to satisfy your request.

You can now control how much reasoning a model does before responding. A new setting in the model picker lets you choose between low, medium, and high reasoning effort, and the selection persists across sessions and devices.

How it works

When you pick a model that supports extended reasoning, a new dropdown appears below the model name. Choose low for quick, straightforward answers; medium for the default balance; or high when the task demands careful, multi-step thinking. Your preference sticks — change it once and it applies to every turn until you change it again.

Where it applies

The reasoning control works in chats, agents, automations, and dashboards. It also flows through to sub-agents and respects workspace model policies, so admins can set reasonable defaults while users tune effort per task.

Connecting Slack is now a one-click flow. Instead of creating an app, copying tokens, and configuring scopes manually, just sign in with your Slack workspace and d5s handles the rest.

Sign in, done

Add a Slack connector from the catalog and choose "Sign in with Slack." You will be redirected to Slack to authorise the connection, then sent back to your workspace. The credential is stored securely and works identically to a manually configured bot token — same API surface, same agent capabilities, zero busywork.

What agents can do

Once connected, agents can read channel history, post messages, search messages, list users and channels, and more — all through the Slack Web API with scoped, read-and-write access.

A round of polish across the workspace: project context tabs now show real data instead of placeholder stubs, the workspace index has a global search bar, sidebar collapsible sections remember your preference, long file names scroll on hover, and scrollbars are thin and borderless across the app.

Real context, finally

The project Context tabs — Agents, Skills, Files, and Memory — previously showed hard-coded mock data. They now pull live data from the API, so you see your actual agents, skills, uploaded files, and saved memory entries.

Find anything, fast

The workspace home page now has a global search bar. Type to filter projects by name — no more scrolling through a long list to find the one you need.

Small touches that add up

The sidebar Recents section remembers whether you collapsed it. Long file names in the context panel slide into view when you hover over them. And scrollbars throughout the app are now thin, borderless, and unobtrusive — content gets the full width.

Workspace admins can now curate which models are available to their team. A new Models settings page lets you toggle individual models on or off, and every picker across the product — chat, agents, automations, and dashboards — respects the policy.

Set your model policy

Visit workspace Settings → Models to see every model in the catalog with a toggle. Disable the ones you do not want your team using, and the restriction applies everywhere: the chat composer, the agent editor, automation configuration, and dashboard setup. Changes take effect immediately with no deploy or restart.

Safe defaults

New workspaces start with a curated set of recommended models enabled. You can lock down further or open up as your needs evolve. The setting is enforced server-side, so a picker cannot accidentally offer a disabled model.

You can now attach files when creating a new project. Drag and drop documents, screenshots, or code samples into the New Project dialog, and they will be uploaded automatically once the project is created.

How it works

Drop files into the add-files area — or click to browse — before you confirm the project name. The files queue up and upload after the project exists. Controls are disabled while the project is being created so nothing races ahead.

Workspace admins who are not the workspace owner can now manage connectors and API keys. Previously these surfaces incorrectly required the owner role, locking out admins assigned to help run the workspace.

What was fixed

Two permission checks were gated on owner-only instead of the full admin role. The connector management panel and the API keys settings page now recognise all workspace admins, matching the backend permissions model.

The model catalog has expanded significantly. You can now choose from a wide range of frontier models across multiple families — all routable through the same infrastructure and selectable per project or automation.

What is new

Curated entries for several major model families are now available in the picker: GPT, Claude, Gemini, and Grok. Each entry includes context-window limits, environment availability, and a selection hint so you can pick the right model for each task.

Switch freely

Every model in the catalog works through the same gateway — no separate API keys, no per-model configuration. Pick a different model for your next turn and it just works.

You can now attach files directly in chat. Upload documents, screenshots, or code samples in the message composer, and they appear as cards above your message — visible to you and the agent, and tracked in the session.

Attach as you type

A paperclip button and a pending-attachment tray sit in the composer. Drop files or click to browse. A live count shows how many are queued, and you can remove any file before sending. Uploads are blocked while a message is in flight so nothing races ahead.

Files you can see

Attachments appear as cards above your message in the chat transcript. They also show up in the Context sidebar under Files, where you can click any file to preview it in the right panel — text, images, PDFs, and more. Every attachment is scoped to its session so conversations stay self-contained.

The Projects page has been redesigned for a calmer, more readable layout. Cards now show each project's instructions as a short description below the name, and the grid adapts from two to three columns depending on your screen width.

What changed

The old dense four-column grid has been replaced with a wider two-to-three column layout. Each card surfaces the project name, a three-line preview of its instructions, and the last-active timestamp. The result is a page that is easier to scan and surfaces more context at a glance.

Project memory is now visible in more places. Memory entries appear as clickable cards in the context sidebar and are listed on the project overview page — so the knowledge your agents accumulate is always within reach.

Browse memory where you work

The context sidebar now shows each memory entry as a tappable card. Tap one to read its full content without leaving the project. The project overview page also lists all memory entries, giving you a quick scan of everything your agents have saved.

Agents now have a better toolkit: a dedicated edit command for precise file changes, configurable timeouts for long-running commands, smarter output handling, and reliable memory of which files they have touched across long sessions.

Edit files with surgical precision

A new edit tool replaces the old write-and-hope workflow. Agents can now make exact string replacements in existing files — change a function signature, fix a typo, or update a config value without rewriting the whole file. Fewer mistakes, less thrashing.

Run longer commands

The execute tool now accepts an optional timeout up to 300 seconds. Builds, test suites, and dependency installs that used to get cut off can now run to completion. The output also keeps both the beginning and the end — the command result and any error message at the tail are preserved, while the noisy middle is trimmed.

Remember what they did

After a long session, when the agent's context compacts, it now retains an exact list of every file it read or modified. No more re-discovering work already done. And path errors are clear: if a file operation targets something outside the sandbox, the error message tells you exactly why.

The d5s.tech homepage has been redesigned from the ground up. A new hero, a cleaner story, and a responsive layout that works on every device.

What changed

Every section of the landing page has been rebuilt. The hero leads with a sharper value proposition. The runtime section explains how agents actually work on your infrastructure. A new how-it-works walkthrough, an FAQ that answers real questions, and a polished changelog teaser wrap the page.

Why

The old page told a broad story. The new one tells a precise one: d5s gives your team autonomous agents that run on your own infrastructure, adapt to your systems, and cost a fraction of operating an in-house AI team — all without sending your data to a third party.

Shared chats now show who said what. Every user message carries a subtle initials avatar, so you can tell at a glance which team member made which request or decision — even in a busy thread.

Multi-user attribution

With project sharing now live, multiple people can participate in the same chat. Each user message displays a coloured initials badge derived from the author's identity. Hover over it to see the full name and email. Distinct people get visually distinct colours, making it easy to follow a conversation between several collaborators.

Project-level connector management is now available. Workspace admins can add connectors directly from the project context panel, and new credentials activate for the current project immediately — no switching between screens.

GLM 5.2 is now available in the model catalog. It is a strong coding model with a 1-million-token context window — ideal for long-horizon development tasks, large refactors, and working with sizeable codebases.

Pick it in your workspace

You can select GLM 5.2 from the model picker in any project or automation. It joins the existing lineup of frontier and self-hosted models, all routable through the same infrastructure. Switch between models per task without changing your workflow.

The GitHub connector now lets agents work with repositories the way developers do — clone, read files, and push changes directly through the connector. Your agents can interact with your code without manual intervention or workarounds.

What changed

Previously the GitHub connector was limited to read-only API operations. Now agents can clone repositories into the sandbox, browse the file tree, read source files, and push commits back — all through the same credential you already set up. Works with public and private repositories.

What you can do

Hook an agent into a repo and let it read code, propose fixes, or generate documentation — then push the result as a branch. The connector handles auth, so the agent never sees your token.

Chat now shows agent messages as they happen — text and tool calls interleaved in real time, rather than waiting for every tool to finish before showing a word. You see the agent think, act, and respond in a single unfolding turn.

Live, not batched

Before this change, an agent turn that ran several tools would block on the last one before surfacing any text. Now the streaming output weaves together prose and tool invocations as they complete. You can read the agent's reasoning while it's still working — no more staring at a spinner wondering what's happening.

Why it matters

When an agent runs a long command, you see incremental results as they arrive. If it takes a wrong turn, you catch it earlier. And the conversation feels faster because you're consuming useful information while the agent is still producing it.

Automations now support one-time runs and a scrollable calendar — schedule a job for a specific time and let it disable itself when it is done, without cluttering your recurring jobs.

Fire once, then stop

A third trigger sits alongside recurring schedules and manual runs: once. Set an absolute date and time, and the automation fires exactly then. After the run completes, the scheduler disables it automatically — no cleanup, no forgotten cron jobs. Perfect for one-off tasks like a release-day notification or a scheduled data export.

A calendar that keeps up

The calendar view now scrolls freely through months, so you can look ahead or back without hunting through a paginated list. Past runs stay coloured by outcome, upcoming runs show what is queued, and the whole timeline feels like a calendar instead of a table.

The workspace now loads faster and feels more responsive. Pages render their shell immediately, list items open their detail views instantly, and transient errors recover on their own instead of stranding you on a dead-end screen.

Shell first, content after

When you open a workspace page, the navigation and layout appear instantly. Content fills in with lightweight skeletons instead of a blank loading spinner. No flashing, no layout jumps — the page feels responsive from the first frame.

Instant detail views

Clicking a project or automation in a list opens its detail page immediately, seeded from the data you already have. No loading spinner, no fetch waterfall — the header and status are visible while the remaining detail loads in the background.

Error recovery that works

If a workspace load fails because of a transient network issue, the page now retries on its own. A stale overnight token refreshes before it can cause a failure, and when something does go wrong you see a descriptive error with a retry button instead of a misleading dead end.

Google Search Console is now available as a connector. Connect it to bring your site's search performance data directly into your workspace — rankings, click-through rates, and indexing status, all queryable by your agents.

What you can do

Once connected, agents can pull search analytics, inspect which queries drive traffic, check page indexing status, and surface trends — all through the same read-only connector safety model as every other integration.

Setup

Add a Google Search Console credential from the connector catalog. The OAuth flow requests read-only webmaster scope so your data stays safe.

Sub-agents now run as full-fledged sessions with two distinct profiles, real-time streaming, and safe connector access. Delegation that used to feel like tossing a message into the void now comes back with a live feed and a paper trail.

Explorer and generalist profiles

Sub-agents come in two flavours. Explorer agents can read files, inspect connectors, and gather context — strictly read-only, so they can roam freely without risk. Generalist agents get a full writable sandbox: they can run commands, create files, and produce working artifacts. Pick the right agent for the job.

Real sessions, not ghosts

Every sub-agent appears in your session list with its own status, streaming output, and usage. Watch them work in real time, roll them into billing, and keep a permanent record of what they did. The parent agent pauses until the sub-agent finishes, then picks up with the results — no polling, no lost context.

Connectors for explorers

Explorer sub-agents can now use installed connectors to gather information from external services. Read-only commands keep your services safe while the agent does its research.

The d5s changelog is live — and a d5s agent writes it. Every day, an automation reads what shipped, drafts public release notes in the company voice, and opens them for the team to review. Once approved, the site rebuilds. The changelog is itself a working example of d5s agents doing real internal work.

A single scrollable feed

The changelog lives at /changelog as a single, scrollable timeline. Each entry gets a centered column with the full release content rendered inline, a header with the date, and links back to the merged changes. No cards, no separate detail pages — just the story of what changed, in order.

Dogfooding from day one

This entry you're reading was drafted by the same automation pattern that will keep the changelog current. A daily scheduled run reads what shipped, decides what's changelog-worthy, writes the prose, and opens it for human review. It's one of the first automations running in production on d5s — and it's editing its own origin story.

Chat just got a lot faster. Responses now stream token by token instead of arriving all at once, and the scroll keeps you oriented so you can watch answers fill in without losing your place.

Token streaming

Time-to-first-word drops from "the whole reply" to sub-second latency. Incremental tokens stream to the chat panel as the model produces them, with the frontend accumulating the response so it reads naturally without flicker. Streaming is now generally available — the feature flag is gone.

Smarter scrolling

New messages pin to the top of the viewport and reserve a screen of space below. The streaming answer fills that space calmly instead of chasing the bottom of the page. The scroll-up is animated (smooth, not a jump cut), and the message stays visible throughout the entire turn — no more disappearing above the fold mid-response.

Automations no longer have to fire and disappear. You can now put an agent on a recurring schedule and see the whole rhythm of your workspace in one place.

Recurring runs

Set an automation to run on the cadence that fits the work — hourly, daily, weekly, or a custom schedule. Each run picks up its instructions and executes in a fresh sandbox, so every run starts clean.

A calendar for your agents

The new Calendar and Agenda views lay every run out on a timeline. Past runs appear with their actual outcomes — succeeded, failed, or blocked — so you can see what happened at a glance. Upcoming runs show what's queued next. It turns a list of automations into something you can actually plan around.

Turn tasks on and off

A simple enable toggle lets you pause an automation without deleting it — handy when you want to mute a job for a release week and switch it back on afterward.

The agent workspace is built for teams that want frontier capability without frontier lock-in. That principle now runs through the whole product.

Model choice, end to end

Route work between frontier and self-hosted models without rebuilding your stack. The model is an implementation detail of a run — swap it without touching the workflows, tools, or memory around it.

Your cloud, your control

Every run stays on infrastructure you control, with the workspace — sandboxes, memory, tools, and governance — built around it rather than bolted on.