Minimal filesystem
The session sees its own Git worktree and shared project memory. Other host files are not mounted.
Fignix product 02
Your coding agents can run longer.
Your controls should run deeper.
FignixOS runs each coding-agent session in its own container by default, with isolated workspaces, controlled network egress, enforced spend limits, and an evidence trail you can review.
Local-first desktop and headless runners · Docker, OrbStack, and Colima
One containerEvery agent session gets its own sandbox
Default denyNetwork access starts closed
Real pauseExecution freezes at the budget boundary
Local firstRepositories and credentials stay under your control
Container-per-session isolation
The agent gets a workspace.
Not your machine.
When FignixOS detects a compatible container runtime, sandboxed execution is the default. The runner creates a dedicated container for the session and exposes only the files and network paths required for that task. Advanced users can explicitly opt out, but isolation is the normal operating mode.
No direct internet
The container joins an internal network with no internet route. Approved model traffic crosses a controlled egress proxy.
Credentials stay outside
Claude Code credentials come from the runner environment, not the spawn API or stored session state.
A real runtime freeze
Crossing a budget pauses the container through cgroups. Review it, raise the limit, and resume the same process and workspace.
SESSION ISOLATION SANDBOXED
HOST
01Session worktreeread + write mount
02Project memorytask-scoped mount
DEDICATED CONTAINERClaude Codeprocess namespaceworkspace: /workspacebudget watcher attached
INTERNAL NETWORKNo direct internet route
→EGRESS PROXYApproved destinations only
BUDGET CROSSEDContainerPauseinspectresume same session
Governance below the board
Orchestration shows the work.
FignixOS governs the execution.
A kanban board is useful, but it is not a safety boundary. FignixOS connects the visible workflow to the sandbox, budget ledger, access policy, and activity record underneath it.
Sandboxed execution
Run each agent session in its own container by default, with only its worktree and project memory mounted into the sandbox.
Enforced budgets
Set token or dollar limits at task creation. When a worker crosses the boundary, execution actually pauses for review.
Scoped data access
Expose only the repositories and data sources a task needs, with credentials mediated outside the agent sandbox.
Auditable work
Keep task history, terminal output, changes, usage, decisions, and review state connected to the session that produced them.
One controlled lifecycle
From task brief to reviewed change.
Every task carries its own execution contract. The agent, workspace, permissions, and budget travel together, so the board cannot drift away from what the runner is actually doing.
Backlog
Define the task, model, workspace, and spend ceiling.
Sandbox
Launch the selected coding agent inside an isolated worker session.
Govern
Enforce access, network, compute, and budget boundaries while it works.
Review
Inspect evidence, resume paused work, and move verified changes toward merge.
Planning and release gates
Agents can draft the plan.
A person releases the work.
Turn an early requirement into an implementation-ready plan without losing the human decision point. FignixOS records acceptance criteria, exposes agent-authored steps, preserves human edits, and requires an explicit release before execution begins.
Session evidence
See the change, the budget, and the execution record together.
A session is more than a terminal. FignixOS connects the code produced by an agent with its machine, isolation state, budget consumption, human release, and timestamped activity. Reviewers can understand what happened without reconstructing the story from separate tools.
Organization oversight
Treat humans and agents as one operating capacity.
Roster, availability, concurrency, utilization, spend, and delivery belong in the same picture. FignixOS gives engineering leaders a team view for assignment decisions and an organization view for capacity and budget decisions.
Team operations
Every human and agent on one roster.
See fixed and ephemeral agents beside the people who release and review their work.
Executive visibility
Spend and capacity without guesswork.
Track delivery, budget headroom, worker availability, and utilization across every team.
Fleet visibility
One operating view, even when the agents differ.
Run local models through OpenCode or subscription-backed work through Claude Code. Each backend keeps its own strengths while FignixOS normalizes task state, usage, isolation, and review into one fleet view.
- Live board state derived from real execution
- Per-worker terminals, changes, and usage
- Persistent sessions across app restarts
Human control
A pause button backed by the runtime.
When a worker reaches its limit, FignixOS pauses the underlying process or container. A human can inspect the state, change the budget, and resume the same session without pretending that a status label stopped the work.
- Token and dollar ceilings
- Visible budget burn-down
- Resume without restarting the task
FignixOS early access
Let agent fleets move quickly,
inside boundaries you can trust.
FignixOS is under active development. Talk with us if you want to test the desktop experience, headless runners, or governance model with real engineering work.