Talk through the work
Discuss a project, task, milestone, board, or plan, then review a proposed change before it is applied.
The execution system for humans and AI agents
Plan the work.Orchestrate the agents.Ship verified.
Inflara / acme-storefront
Simplified demo Agent Center
Configured machines, capabilities, and runs
Workflows
Full Team With Tests
16 steps
Robust Workflow
12 steps
Ship checkout coupon fix
running
Workflow run
Ship checkout coupon fix
recovery loop
Plan AClaude
Plan BCodex
Judgepicks best
ImplementCodex
Test gateverify
Shiphandoff
Live workflow log
Dispatching to agents…
Simplified, on-rails preview — no real data or agents are changed here.
A working taste of Inflara — plan the week, chat with the Assistant, and watch a dispatch get verified
Inflara Assistant
Turn a conversation into reviewable work.
Assistant is a first-class workspace for planning, administration, and supervised execution. It keeps context across turns and proposes supported actions for you to review, within your configured capabilities and permissions.
Assistant · Planner MVP
Workspace context connected
I can plan, inspect, and propose supported actions. Nothing changes until you review the proposal.
Monday is overloaded. Protect onboarding and keep the Thursday due date.
What this preview shows
Stay with the run
From a source-linked conversation, follow configured workflow progress and reply, retry, or cancel when those actions are available.
Use configured capabilities
Bring the connected tools, skills, and runner capabilities your workspace has enabled into one persistent conversation.
Keep authority explicit
Assistant works within configured permissions, surfaces proposed administration, and preserves a reviewable conversation record.
Availability depends on workspace permissions, connected capabilities, and runner configuration. Proposed changes remain reviewable.
From this morning's plan
to tonight's verified merge.
01Plan
A week that plans itself around reality.
The AI planner drafts your days from deadlines, priority, effort, and real capacity — meetings stay fixed, focus blocks get protected, and agent work gets its own lane in the flow.
- Capacity-aware daily and weekly drafts you accept, not obey
- Task flow from To do through Agent, Review, and QA
- Focus blocks that survive replanning
02Dispatch
Hand real work to real agents.
A supervisor agent turns selected tasks into validated workflows. Roles are assigned by policy — balanced, highest reliability, or lowest consumption — with verified local outcomes entering only after enough comparable evidence. Every selection explains itself.
- Codex, Claude Code, and Antigravity as first-class workers
- Auto assignment with a preview and a written rationale
- Manual pools always win when you want control
03Orchestrate
Workflows with judgment built in.
Planners compete, a judge synthesizes, implementers build, and verify loops re-test until tests actually pass. Graphs are validated before every save — cycles, dead ends, and deadlocks never reach a runner.
- Named templates from Single Worker to Full Team
- Bounded loops, joins, gates, and recovery branches
- One git worktree per run; merges are planned, not hoped for
04Verify
Evidence, not vibes.
Every step ends in a typed handoff: artifacts, decisions, risks, confidence, provenance, and changed files. Budgets are enforced by the engine, failures are classified and routed to the right owner, and each finished run distills into memory the next dispatch learns from.
- Success is only reachable through a passing test gate
- Live budget usage — nodes, agents, recovery, wall clock
- An audit trail your compliance team can actually read
New · Inflara Assistant
Your whole workspace, one conversation.
The Assistant is a durable teammate living inside Inflara. Tell it what you want in plain language and it plans the week, administers the workspace, dispatches agents, and reports back with verified outcomes.
Runs your workspace from plain language
Creates and reshapes projects, milestones, tasks, Kanban flow, and weekly plans — ask for the outcome, skip the clicking.
Dispatches and supervises real agent work
Builds workflows, launches runs, answers agent questions mid-flight, and reports back with verified results.
Works on your machine, with your tools
Executes through your own runner — your shell, skills, and MCP servers — so code and context never leave home.
Durable across sessions
One continuous collaborator, not a throwaway chat: it keeps context, remembers decisions, and picks up where you left off.
Assistant · Launch weekOn your runner
Plan my launch week and get the coupon fix shipped.
On it — planning around Thursday's launch and dispatching the fix.
Created 3 tasks in Planner MVPScheduled 6 focus blocks · Mon–ThuDispatched verify-loop workflow · Codex + Claude Code
Coupon fix verified — 12 tests passing, review branch ready for you.
Ask the Assistant to plan, build, or dispatch…
Agentic orchestration
Multi-agent workflows that only succeed through a passing test gate.
Most stacks leave reliability as advice. Inflara ships it as runtime behavior — six guarantees the engine enforces on every run.
Verification-gated success
Verify loops make “done” mean “verified” — the success branch is only reachable through the passing test gate.
Typed failure routing
Failures are classified — environment, test, merge, budget, permission — and routed to the owner who can actually fix them.
Hard budget envelopes
Node, agent, recovery, and wall-clock ceilings enforced by the engine with typed overrun codes. Never silent truncation.
Local-first privacy
Agents run on your machine. The server orchestrates and receives bounded artifacts — your repository never leaves home.
Git-native isolation
A worktree per run, review branches for successful work, merge plans across parallel branches, changed-file evidence everywhere.
Routing that explains itself
Benchmark priors plus conservative calibration from verified comparable runs. Every auto-assignment states its basis — and manual always wins.
Works with the agents you already run
CodexClaude CodeAntigravity
Full agent CLIs with their native tools and sandboxes — not raw model calls behind a wrapper. Your code stays on your machine; the server only ever sees bounded, size-capped artifacts.
Fifteen minutes from install to your first supervised run
Stop babysitting agents.
Start holding them to account.
Plan today, dispatch this afternoon, and review verified work tonight — from one calendar.