execution
Pluggable Execution Backends
A clean contract layer lets the orchestration service dispatch agent work to local processes, Docker containers, or Kubernetes pods — chosen per-project via env var, YAML, or CLI flag.
SprintFoundry orchestrates 10 built-in agents across pluggable execution backends (local, Docker, Kubernetes) to turn backlog tickets into tested, reviewed pull requests — with Mission Control v4 dashboard, snapshot recovery, and OTel-ready telemetry.
Works with your existing stack
49 commits shipped since March 6. K8s-native execution, Mission Control v4, snapshot recovery, and more.
A clean contract layer lets the orchestration service dispatch agent work to local processes, Docker containers, or Kubernetes pods — chosen per-project via env var, YAML, or CLI flag.
PVC-backed workspaces, service account secret projection, namespace isolation, configurable egress network policies, and pod-level sandbox recovery.
Complete React + Vite + Tailwind rewrite. Run dashboard with filtering, run detail with agent activity and file diffs, agent monitoring view, real-time SSE streaming.
Export and restore entire run state. Resume interrupted runs from the last checkpoint. PVC snapshot plans for K8s workspaces keep agent work durable.
Skip the full SDLC orchestration pipeline for targeted tasks. Run any agent directly from the CLI with a prompt — the default mode for single-agent runs.
Track sandbox creation, attachment, teardown, and recovery events through the OTel pipeline. Per-backend provisioning timing surfaced in metrics.
A clean separation between your planning tools and your codebase. No new workflows to learn — SprintFoundry fits into what you already use.
Give SprintFoundry a Linear ticket, GitHub issue, or Jira task. Or just describe what you want built as a prompt.
The orchestrator classifies the work, applies mandatory platform/project rules, and generates a structured execution plan (with model/runtime selection per step).
Specialized agents run sequentially or in parallel. Each reads prior artifacts, does its job, and writes outputs for the next agent. The filesystem is the message bus.
A tested, reviewed pull request is created with code, tests, docs, and a full audit trail. You review and merge — or configure auto-merge for qualifying changes.
Each agent has runtime-specific instructions, explicit handoff artifacts, and service-enforced role ordering.
A pluggable execution contract lets you run agents anywhere — from local processes for development to Kubernetes pods for production. Switch backends per-project with a single config line.
CLI subprocess or in-process SDK. Zero infrastructure. Ideal for development and single-agent direct runs.
Each agent runs in an isolated container with mounted workspace volume. Provides process and filesystem isolation without K8s overhead.
Production-grade agent orchestration. PVC-backed workspaces, service account secret projection, namespace isolation, egress network policies, and pod-level sandbox recovery and reattach.
Monitor UI gives run-level context while OpenTelemetry pipelines expose service and runtime telemetry for production operations.
Grafana operations dashboard (trimmed capture).
Service enforces constraints; the orchestrator decides execution.
Use your own Anthropic/OpenAI credentials. Keys stay service-side, not in agent artifacts.
BYOKOrchestrator plans execution; validator enforces role ordering and mandatory rules.
Deterministic gatesRun agents as Kubernetes pods with PVC workspaces, namespace isolation, egress policies, and service account secret projection.
Production-gradePer-agent/per-task token-cost caps and timeout windows are enforced during orchestration.
Cost governanceInject required agents, human gates, and model/budget overrides from metadata, labels, priority, or changed files.
Policy as YAMLPause runs for reviewer decisions with `sprintfoundry review`, then continue with feedback trail.
Controlled autonomyQA/security/code-review failures re-route to targeted agents with bounded rework cycles.
Iteration built-inLive board UI plus APIs for runs, events, logs, and workspace files (`/api/runs`, `/api/events`, `/api/log`, `/api/files`).
ObservabilityEvery step writes durable artifacts plus runtime debug metadata for traceable audits.
Traceable outputExport and restore whole-run state. Resume interrupted runs from checkpoints. PVC snapshots for K8s workspace durability.
Fault-tolerant`sprintfoundry project create` scaffolds config, integrations, agents, branch strategy, and runtime profile.
Fast onboardingSkip full SDLC orchestration. Run any agent directly from CLI for targeted tasks — the default for single-agent runs.
Flexible workflowInstall via Homebrew/npm, then scaffold config with project create and run doctor.
Set repo, ticket source, agents, runtime/model overrides, and BYOK credentials.
Run on GitHub/Linear/Jira/prompt. Use dry-run for planning-only mode.
Track runs in monitor and resolve review gates with sprintfoundry review.
# Install (Homebrew recommended) brew tap Sagart-cactus/tap brew install sprintfoundry # Or install with npm npm install -g sprintfoundry # Create project config interactively sprintfoundry project create # Run preflight checks, then validate config sprintfoundry doctor --project my-saas-app sprintfoundry validate --project my-saas-app # Run on a Linear ticket sprintfoundry run --project my-saas-app \ --source linear --ticket LIN-423 # Run on a GitHub issue sprintfoundry run --project my-saas-app \ --source github --ticket 42 # Or just describe it sprintfoundry run --project my-saas-app \ --source prompt --prompt "Add CSV export to the reports page" # Optional: plan-only dry run sprintfoundry run --project my-saas-app \ --source linear --ticket LIN-423 --dry-run # Run with Docker execution backend SPRINTFOUNDRY_EXECUTION_BACKEND=docker sprintfoundry run \ --project my-saas-app --source linear --ticket LIN-423 # Run with Kubernetes execution backend SPRINTFOUNDRY_EXECUTION_BACKEND=k8s sprintfoundry run \ --project my-saas-app --source linear --ticket LIN-423 # Export a run snapshot for recovery sprintfoundry snapshot --export run-423 # Run a single agent (skip full SDLC) sprintfoundry run --project my-saas-app \ --source prompt --prompt "Review auth logic" --agent security # Add a custom agent definition sprintfoundry agent create # Monitor runs and logs sprintfoundry monitor --port 4310 # Resolve a human review gate sprintfoundry review --workspace /tmp/sprintfoundry/my-saas-app/run-123 \ --review-id review-123 --decision approved
Stop context-switching. Stop writing boilerplate. Stop waiting for bandwidth.
Let your AI agents handle the sprint while you focus on what matters.
Open source · Self-hosted · BYOK · K8s-native · No vendor lock-in