Claude Code
Codex
Docker
Git
miseGive every agent its own space.
Each agent works in an isolated container and private clone. Path claims and checked submissions protect shared work as commits come together.
A ZEUSFYI PRODUCT / EARLY ACCESS
Coding agents.
Working as one.
Build faster with a coordinated team of Claude Code and Codex agents. Give every agent a role, an isolated container, and the context to move work forward.
A TEAM, IN SYNCONE PROJECT. MANY HANDS.
DockingbirdWORKSPACE
Ship the authentication refresh
ISOLATED WORKSPACES · SHARED CONTEXT
Coordinator splits the work, routes questions, and keeps the team moving.
TEAM CONTEXTCoordinator assigned two implementation tasks
API builder owns src/auth/. Interface builder owns src/ui/.
API builder shared the response contract
A direct handoff gives the interface builder the context it needs.
Interface builder submitted its commit
Changed paths are checked against its active claims.
Review & test agents inspect the work
Results return to the coordinator with the relevant artifacts.
FEATURE DELIVERY
One pattern. A team of specialists.
A reusable swarm brings together the right role instructions, tools, and environment for this project.
- Agents
- Coordinator, builders, review, test
- Instructions
- AGENTS.md + role prompts
- Toolchain
- mise.toml + mise.lock
- Hooks
- Trusted repository settings
- Runtime
- One container per agent
3 working 2 waitingfeat/auth
Interactive illustration · sample workflow and data
YOUR TOOLS. BETTER TOGETHER.
Claude CodeCodexDockerGitmise
01 / COORDINATIONParallel work.
Parallel work.
Shared direction.
Let agents build, review, and test together while you focus on the decisions that need you.
Keep handoffs moving.
Agents send questions, reports, and artifacts through a shared message bus. Context reaches the right session without you relaying it between terminals.
Stay close to the work.
See what is working, waiting, or asking for your input. Open a session when you need the detail, then let the team get back to it.
02 / CONFIGURATIONYour way of working.
Your way of working.
Ready for every agent.
Turn the setup that works into a reusable pattern. Choose models and effort, compose role prompts, and carry your project’s tools and hooks into each container.
Use a single configured agent or compose a swarm. The pattern belongs to your workflow.
# A specialist in your feature swarm
harness = "claude"
workspace = "repo"
mode = "bus"
tools = "full"
instructions = "prompts/builder.md"
repo_settings = true
[env]
APP_ENV = "development"AGENTS.mdrole promptsmise.lockhooks
Pair builders with separate review and test agents. Each receives its role instructions and project context.
Agent files and system instructions, model settings, environment variables, toolchains, and trusted repository hooks. Configured for the job.
Coordination at a glanceSAMPLE DATA
Working06
Waiting02
Needs you01
Context usageBy agent
coordinator
64%
builder-api
38%
reviewer
47%
Model cost estimate*$4.82
Compactions*02
API retries*01
*Illustrative Claude Code figures. Availability varies by agent; cost estimates are not billing data.
03 / VISIBILITYKnow what’s moving.
Know what’s moving.
See what needs you.
Track agent activity, context usage, turns, and reported model usage. Bring supported metrics into Prometheus and Grafana to understand how your coordination is working.
- Spot waiting agents and pending questions.
- Inspect context pressure and reported retries.
- Compare activity across agents and hosts.
04 / FLEET SCHEDULINGBigger workloads.
Bigger workloads.
A wider runway.
Put your local machines and remote runners to work. Queue swarms by priority, place work against CPU and memory headroom, and see why a job is waiting.
Local workstationAvailable for work
38% CPUCloud runnerRunning feature delivery
62% CPUDevelopment laptopDraining · existing work continues
DrainWORKLOADSTATE
Authentication refreshFeature swarm · cloud runner
RunningDependency reviewWaiting for memory headroom
This sample workload needs more memory than the eligible host currently has available. Dockingbird keeps it queued and reports the hold reason.
Test suite cleanupPaused by operator
PausedA FEW DETAILS
Before you take flight.
Do I keep using Claude Code and Codex?
Yes. Dockingbird coordinates the coding agents you already use. Each runs in its own container with its own workspace, while Dockingbird manages communication and shared work.
Can I bring my own instructions and tools?
Yes. Reusable configurations can select the harness and model settings, supply role instructions and environment variables, and load trusted repository settings. Your project’s agent files, hooks, and mise toolchain can shape how agents work. Hook support depends on the harness.
Can the agents run on different machines?
Dockingbird can coordinate local and remote hosts. Its fleet scheduler queues work, considers CPU and memory availability, and provides controls for priority, holds, pause, resume, and host draining.
What does joining the waitlist mean?
We’ll email you about Dockingbird early access and availability. There’s no payment or commitment to join. You can ask us to remove your email at any time.
GET EARLY ACCESSMake room for
your next big build.
A coordinated team of agents.
A little more space to think.
Be first in the flock.
Join the Dockingbird waitlist.