Finished Work

What’s new

v0.5.0 connected the work to your data and put it inside governed boundaries. Aimless v0.6.0 finishes it. Research turns into branded reports and slide decks you can hand off. Everything the AI produces lands in a vault. Every resource you create is private until you choose to share it. And long autonomous runs now survive retries instead of starting over.

Reports you can hand off

The report swarm now produces real deliverables. Ask for a report and it generates a server-side branded PDF — and a slide deck — styled to your organization’s brand kit, not a wall of plain text. The underlying research swarm is date-aware, renders proper paragraphs instead of fragments, and ends with actionable open questions rather than a flat summary. Reports that can’t ground a number drop NaN deltas instead of printing them, and when persistence fails the in-chat card says so instead of spinning forever. When a request is ambiguous, the report agent asks about branding before it commits.

A File Storage vault for everything the AI produces

A new always-on File Storage vault gives every organization one place for the artifacts agents generate. Reports are filed there automatically — into a dedicated Reports folder — alongside anything else the pipeline produces. Each file has its own privacy setting with explicit sharing, so a generated document is yours by default and shared only when you say so. AI-pipeline sub-connector artifacts are kept out of the Vaults grid, so what you see is the work, not the plumbing.

Private by default

Access control now starts from the safest place. Newly created resources are private by default, visible only to you until you grant access. Personal OAuth and connector credentials are never silently org-shared — they default to private. Per-file privacy in the File Storage vault follows the same rule. You opt in to sharing; you never have to discover after the fact that something was open.

One organization hub

Everything about running an organization now lives in a single admin hub: members, resources, access, budgets, and spend, plus a General tab for identity and your brand kit. The Access and Resources tabs list every shareable resource type — agents, sources, dashboards, reports, and more — so there are no hidden corners. Invitations send real emails with a working acceptance flow, brand-kit edits are restricted to owners and admins server-side, and the main workspace switcher now shows each workspace’s visibility with inline edit options.

Resilient autonomous runs

Long-running work no longer starts from zero when something hiccups. Within a turn, a stalled delegation resumes from where it stopped, guided by a failure classifier that decides whether to retry, escalate, or stop. Discovery results persist across turns so an agent doesn’t re-learn what it already found. Swarm phases checkpoint and resume on a Trigger retry instead of replaying the whole pipeline, and the builder’s step budget now scales to a request’s real complexity rather than a fixed ceiling. Widget bindings that were disabled by a transient upstream outage auto-recover once the source is healthy again, and a connector that hits a credential failure triggers a re-auth flow instead of failing quietly.

Edit dashboards in place

When you ask to change an existing dashboard, the AI edits it in place instead of rebuilding it from scratch. KPI cards collapse bound breakdowns correctly and humanize their labels, the builder matches SQL query shapes to the right widget type, and fabricated binding IDs are rejected so a widget can’t claim data it isn’t actually connected to. Binding reuse is scoped to the sources in the current build, so one dashboard can’t pull in another vault’s data by accident.

A clearer, live reasoning view

The Glass Box timeline keeps getting more legible. A live checklist renders inline on the spine and crosses items off as they complete, generated visuals appear in the data panel, and runs of identical tool calls coalesce into a single counted row instead of flooding the view. The header chip shows the model a request was actually routed to, not a static tier. Artifacts interleave into the timeline by when they were created, and your Overview now reflects real activity — a true per-organization heatmap, live Today counts, and a state-first inbox (Active now / Needs you / Recent) with durable, expandable run traces.

A full public API

Aimless now exposes a public OAuth-secured REST API and an MCP server, with developer documentation, so external tools and assistants can build on top of it directly. Agents can also pull skills from GitHub SKILL.md sources, extending what they can do without a redeploy.

Model routing that converges

Multi-model routing across OpenRouter now works end to end, the routing bandit converges faster on the best model for each kind of request, and a model-agnostic tool-reliability floor keeps tool-using agents dependable regardless of which underlying model serves them.

Infrastructure improvements

• The stable system-prompt prefix is now cached (Anthropic prompt caching via the proxy), and the command bar skips a redundant client-side intent-classification round trip and de-duplicates connected-source probes — both cut latency before the first token.
• HSTS is now sent only in production, so local development over plain HTTP is no longer force-upgraded to HTTPS.
• Per-agent skill assignment, a live exploration weight, durable cost statistics, and federated skill sources make routing and attribution more accurate.
• Marketing content approvals parse correctly and a contentless run is never gated, and the marketing site adopts intelligence-layer positioning with an animated hero.
• A capability-taxonomy categorization overlay (flag-gated, default-off) lays the groundwork for organizing resources by what they can actually do.