Good work.
Multiplied.
A goal, a team, a schedule, or a signal. Put agents to work, see every handoff, and keep the decisions that matter in your hands.
In Locus 2.7, the composer keeps Work, Plan, Grill, Goal, Task Capsules, and Solo or team options in one workflow menu. Your draft stays right where you left it.
Find the failing test.
Fix the cause.
Prove it works.
Give each part to the right mind.
Configure a dispatcher and named specialists. Choose the model and access ceiling for each role, with bounded work and a visible run record.
- Dispatcher01
Decompose the goal into a bounded job graph.
- Researcher → writer02
Investigate first. Then propose a focused change.
- Reviewer03
Check the evidence and review the resulting diff.
Give work a starting signal.
Set a rule. Change the sample. See what would run.
In Locus, schedules and event polling run locally. Approvals appear in Attention; run history stays in Activity.
Delegation should
make the work clearer.
More agents are useful when they have distinct jobs. Locus gives each role a model, an access ceiling, and a bounded part of the goal.
Different roles. Different models.
Use local Ollama for one role and a connected account for another. Authorize hosted routing before a dispatcher sends task context to that provider.
Changes you can actually inspect.
Run team work in a managed Git worktree. Review the diff before it reaches your project. Locus checks for conflicts before applying it, or you can land the result on a branch.
A ceiling on access. A budget for work.
Configure role access and limits on jobs, calls, tokens, and concurrency. Permission modes govern tool access; command guardrails block configured catastrophic patterns.
Longer than
a single turn.
Persistent goals
Continue across turns and app restarts. Keep the goal and its progress together, with pause, resume, edit, and end controls.
Task Capsules
Keep a plan with separate planning, implementation, and review models. Revisit revisions and carry the work into the next stage.
Lasting conversations
Each scheduled agent keeps a primary conversation. Return to its run history and results, with side conversations kept separate.
The workspace is free.
The model is your choice.
No agent seats. No Locus credits. Local models, hosted APIs, and your connected plan each have their own cost picture.
No provider fee.
Inference runs on your hardware. Install Ollama and choose a model that fits your Mac.
Local compute · your machineSee what the work costs.
Usage & Costs shows model calls, hosted tokens, team runs, and estimated cost. Break it down by day, workspace, model, or agent.
Set limits before the run.
Set team limits for jobs, rounds, concurrent calls, total calls, metered tokens, and estimated cost. Keep the amount of work deliberate.
Cost estimates use your entered rates. An estimate limit stops further routing after it is exceeded; it is not a guaranteed cap on your provider’s bill.