What happened
Jaquatech released AgentDesk v1.0.0 on November 7. The release describes it as a local AI development team that runs on the user's computer. A planner breaks work into tasks; developer agents build each task in its own Git branch; a reviewer checks the work; and the user presses Merge. The release frames this as a Scrum process, with a backlog, refinement, sprints, PIs, release notes, security review and release tasks.
The release says the application listens on 127.0.0.1 only, encrypts API keys on disk, sends no telemetry and requires no account. It can use Anthropic, OpenAI-compatible endpoints, Ollama or the Claude Code CLI. Users can select agents for planning, building and reviewing, and add ready-made agents from an agent gallery. The release does not name particular models or specify which OpenAI-compatible services have been tested.
AgentDesk 1.0 includes daily and per-sprint spending limits, on by default at $10 per day and $20 per sprint. Setting a cap to 0 removes that limit. Its usage and billing view breaks costs down by day, agent, model and project, with CSV export. The release also lists an isometric office view showing whether agents are working, reviewing or on break, plus team rooms and a company logo. Sprint and release tasks link to the running tasks view and back.
For sign-in protection, five incorrect passwords pause sign-in for five minutes. Jaquatech warns users upgrading to copy the data/ folder first; the database only moves forward. The release directs users to the README's “Upgrading and rolling back” section for upgrade and rollback guidance. It does not state a license, hardware requirements or installation steps in the release entry itself.
What the archive adds
New Runtime's earlier coverage described local coding agents as needing shared environments rather than separate terminals, and pointed to an emerging multi-agent workspace category. AgentDesk is a new release in that thread: it packages planning, branch-isolated implementation and review behind a local interface. New Runtime's earlier coverage addressed a different environment, focused on parallel mobile app building and testing rather than this planner/developer/reviewer workflow.
Trend signal
This supports the pattern that agent systems are reorganizing work around specifying, routing and reviewing output rather than simply maximizing production speed. The explicit planner, task branches, reviewer and user-controlled merge are concrete workflow boundaries, although the release does not demonstrate that they improve delivery outcomes. AI-native organizations move toward review and orchestration.
Why now
The release's timing is not explained beyond the v1.0.0 publication date. Its significance is practical rather than tied to a stated market event: the release makes a local, multi-backend agent workflow available as a versioned tool, with cost controls enabled by default.
Why it matters
A team experimenting with coding agents can inspect a workflow that separates planning, implementation and review, while keeping each developer agent's work in its own Git branch. That structure offers a natural place to review changes before they enter the main line, and the human-controlled Merge step makes the approval boundary explicit.
The local-only listener, encrypted keys and absence of telemetry or account requirements may also matter to builders who want to evaluate an orchestration tool without adopting a hosted service. Default caps of $10 per day and $20 per sprint make it easier to run an initial trial without leaving spending unconstrained, while per-agent and per-model reporting can help identify where costs accrue. These are product features, not proof that the workflow is secure, cheaper or more productive in practice; teams should test the review quality and operational fit against their own repository and policies.
What happens next
- Jaquatech will document a broader set of supported provider configurations or model-specific setup guidance (6 months, 55%). Watch for: release notes or README updates naming tested providers and configuration requirements.
Limits and uncertainty
The release describes capabilities but does not provide benchmarks, sample task results, user counts or comparisons with other coding-agent environments. It does not quantify review accuracy, task completion rates, time saved or spending under typical workloads. Jaquatech's statements about local operation, encrypted keys and no telemetry are not independently verified here. The entry also does not identify the license, supported operating systems, resource requirements or model versions, so teams would need to check the project documentation and test the application before relying on it.
Skill to build now
Take one small, reviewable coding task from your own backlog and map it into the release's sequence: planner-created tasks, a separate branch for each implementation task, a review step and an explicit human merge. Set a daily and sprint cap before running the trial, then use the per-agent and per-model cost breakdown with CSV export to see where spend went. Compare the resulting changes and review burden with your normal workflow; the release makes no claim that this process will be faster or more reliable.
Reader takeaway
AgentDesk 1.0 turns multi-agent coding into a local, branch-based workflow with an explicit review-and-merge gate and spending limits on by default. It is worth evaluating as a tool, but the release establishes what it offers, not whether it makes coding teams faster, safer or cheaper.