An honest, point-in-time look (June 2026) at where Last Light, a self-hosted, open-source software factory, sits among AI coding agents, against Factory.ai droid, 8090, the vendor-hosted GitHub bots, and the self-hosted open-source agents. Pricing and benchmark numbers move fast, so figures link to sources rather than being treated as durable truth.
Last Light is not in the same category as most "AI coding agent" tools, and comparing it feature-for-feature is misleading. It's a self-hosted, event-driven, single-tenant software factory: a YAML workflow engine that reacts to GitHub webhooks, Slack messages, and cron ticks by running sandboxed agents on a production line to triage issues, review PRs, build features, and report on health and security.
Almost everything it's compared to is either a developer-driven pair programmer you sit in front of (Cursor, Claude Code, Codex CLI, Pi, OpenCode, Aider, Cline), or a vendor-hosted issue→PR agent you delegate to in someone else's cloud (Devin, Copilot coding agent, Codex cloud, Jules), or its commercial-platform siblings, Factory droid and 8090.
The important exceptions, and Last Light's truest peers, are the open-source agents that also run self-hosted and autonomously: OpenHands (Resolver / headless), Goose (scheduler / recipes), and Archon (an OSS workflow-harness builder). Last Light is not alone in that quadrant. The honest distinction from those is shape, covered on each comparison page.
Each of these gets a full page: what the other tool does better, what Last Light does better, and when to pick which.
A flat feature table would flatter or unfairly penalise everything, because these tools live on two axes: who drives (you, per task ↔ it reacts to events) and who hosts (a vendor ↔ you). Last Light lives in the self-hosted, event-driven quadrant, but it shares it.
Cursor, Claude Code, Codex CLI, Windsurf
Devin, Copilot coding agent, Codex cloud, Jules, Factory droid, 8090
Pi, OpenCode, Aider, Cline · OpenHands / Goose (interactive)
★ Last Light, Archon, OpenHands (Resolver), Goose (scheduler / recipes)
So the meaningful question isn't "is anything else self-hosted and autonomous" (things are) but what shape that autonomy takes: the vendor bots are autonomous but hosted; OpenHands / Goose are autonomous and self-hostable but as a single pipeline or a general agent you compose; Archon is a self-hosted harness that orchestrates external CLIs; Last Light ships a purpose-built, multi-workflow software factory for repo maintenance. That breadth is the point, and also why it carries operational burden a per-task CLI or a SaaS bot doesn't.
Treat each cell as a generalisation, not a spec sheet. See each tool's own docs for specifics, and the sources for where these claims come from.
| Tool | Category | Hosting | Trigger model | BYO model? | Isolation | Human gates | Cost model | Best for |
|---|---|---|---|---|---|---|---|---|
| Last Light | Self-hosted software factory | Self-hosted only | Webhooks + Slack + cron + CLI | Yes (Anthropic / OpenAI / OpenRouter) | Docker / micro-VM sandbox + default-deny egress firewall | First-class (GitHub / Slack / dashboard) | Free / OSS + your model spend + your infra | Self-hosted autonomous repo maintenance with approval gates |
| Factory droid | Commercial multi-agent platform | SaaS (on-prem at Enterprise) | IDE + CLI + browser + Slack + Jira/Linear | Yes (Claude / GPT / Gemini / DeepSeek / Qwen) | Managed cloud sandboxes (“Droid Computers”) | Tiered autonomy levels | $20 / $100 / $200 per mo + Team/Enterprise | Enterprises wanting a polished multi-surface droid fleet |
| 8090 | Enterprise software factory | Vendor-hosted or delivered as a service | Business intent → work orders | Vendor-managed | Vendor-managed | Governance + audit trail by design | Enterprise contract | Regulated enterprises modernising legacy applications |
| Devin | Autonomous cloud agent | SaaS | Task / issue assignment | No (vendor model) | Vendor cloud sandbox | Review the PR | Subscription | Delegating well-scoped tasks to a “junior engineer” |
| Copilot coding agent | Autonomous GitHub agent | SaaS (GitHub) | Assign a GitHub issue | No (vendor model) | Sandboxed GitHub Actions | Review the draft PR | Copilot subscription | Teams already living in GitHub |
| OpenAI Codex (cloud) | Autonomous cloud agent | SaaS | Task / multi-surface | No (vendor model) | Vendor cloud sandbox | Review the PR | Subscription | GPT-5.x-grade autonomy |
| Google Jules | Autonomous cloud agent | SaaS | GitHub issue→PR | No (Gemini) | Google Cloud VMs | Review the PR | Free preview / subscription | Gemini-grade autonomy in GCP |
| OpenHands | OSS agent + Resolver | Self-hosted or cloud | CLI/IDE and event-driven (issue label → PR) | Yes (full BYOM) | Sandboxed runtime | In the loop or review the PR | Free / OSS + your model spend (or Cloud) | Self-hosted single-pipeline issue→PR autonomy |
| Goose | OSS general agent | Self-hosted | CLI/IDE, recipes, built-in scheduler | Yes (full BYOM) | Local / your choice | Configurable approval gates | Free / OSS + your model spend | Composing your own autonomous / scheduled agent |
| Archon | OSS harness builder | Self-hosted | Terminal + Slack + Telegram + GitHub comments + web | Yes (orchestrates Claude Code / Codex / etc.) | Per-run Git worktrees | Workflow gates | Free / OSS + your model spend | Dispatching parallel coding runs from anywhere |
| Pi / OpenCode / Aider / Cline | OSS dev-driven agents | Self-hostable | CLI / IDE, per task | Yes (full BYOM) | Local / your choice | You’re in the loop | Free / OSS + your model spend | Free, private, hackable per-task coding |
The flip side of the design choices above, stated plainly.
External claims about other tools are point-in-time (June 2026). Last Light's own specifics are drawn from the spec and the project repository.
Pricing, model menus, and benchmark scores change frequently, so verify against each vendor's own pages before relying on a specific figure.