01
AI is off by default
Admins enable the assistant per workspace. Environment keys are fallback-only; the repo points operators toward Admin -> Agent control and Settings -> AI & Agents.
Workspace toggleAdmin controlFallback env only
TaskNebula AI governance
AI assistance with
local control.
TaskNebula's strongest AI story is not magic generation. It is opt-in BYOK AI, native fallback, model profiles, audit logs, AGENTOWNERS policy, MCP write governance, and local coding-agent handoff.
AI off by default
30-language app shell
Docker Compose install
AGENTOWNERS governance
Real product screenshot
Self-hosted project management, AI drafting, governance, and update awareness in one workspace.
v0.7.9
Latest release
30
Locales
34
GitHub stars
MIT
License
Product evidence
The AI surface is visible in the app, not just in README claims.
The public page now shows real TaskNebula AI drafting and issue-detail screenshots before describing providers and policy. That is better SEO and better trust.
Draft with AI
One prompt becomes a single issue or editable checklist before bulk creation.
Issue detail
Labels, versions, components, comments, sub-issues, and per-issue assist in one view.
Realtime board
Kanban, sprint work, drag-and-drop planning, presence, and project status loops.
Dashboard
Standup, catch-up, analytics, deadlines, pinned work, and recent activity.
Governance model
TaskNebula is an AI-agent control plane for project work.
The repo frames AI as an opt-in operational layer with provider keys, model profiles, AGENTOWNERS policy, MCP routes, local runner settings, audit logs, and approval gates.
02
BYOK provider chain
OpenAI and Anthropic credentials can be configured at platform or workspace scope, encrypted with AES-256-GCM, redacted in previews, and audit-logged on rotation.
OpenAIAnthropicAES-256-GCM
03
Native planner fallback
If no LLM credential is configured, the deterministic heuristic planner keeps AI-adjacent drafting paths usable without external API calls.
Offline optionNo API callsDeterministic
04
Agent-as-assignee
The roadmap positions TaskNebula as a control plane where coding agents can be assigned work and routed through local policy.
Linear Agent ProtocolIssue sessionsApproval
05
AGENTOWNERS policy
MCP write tools attach configured actor markers before issue creation, updates, assignment, status transitions, comments, and subtasks.
CodexClaude CodeCursor
06
Local runner beta
Optional environment settings can dispatch issues to Codex or Claude Code installed in the runtime or custom image, with sandbox, cwd, model, and timeout controls.
Codex CLIClaude CodeSandbox
Human-in-the-loop
Draft, review, apply, audit.
That sequence is the right public story. The assistant should accelerate backlog and issue work while leaving admin policy, approval, and audit controls legible.
01
Draft-with-AI
Backlog prompts become editable single-ticket or checklist drafts. Users choose what to create before bulk insert.
BacklogStructured draftsBulk create
02
Per-issue assist
Issue detail AI assist can summarize with comments, rewrite descriptions, suggest next steps, or propose labels with one-click apply.
SummaryRewriteLabels
03
Model profiles
Provider, model, temperature, max tokens, and reasoning effort are saved as reusable profiles with revision history.
ProfilesRevision historyTuning
04
Failure notifications
Bad key, rate limit, or unavailable model becomes an in-app notification with an action hint and deep link to settings.
Action hintDeep linkAudit trail
Agent action lifecycle
Operating flow1
User or agent proposes work
prompt or MCP tool
2
TaskNebula resolves actor
TASKNEBULA_AGENT_ACTOR
3
AGENTOWNERS policy evaluates
local policy
4
Approval gate or dry-run
human review
5
Issue/comment/status write is audited
audit log
AGENTOWNERS and local runner details were checked in the public changelog and .env.example. The page deliberately describes optional beta runner controls as optional, not default behavior.
What this means
The SEO target is broader than AI project management.
This page now targets the long-tail phrases people actually search when evaluating agentic project-management software: BYOK, MCP, coding agents, approval queue, audit log, and self-hosted AI.
For teams
Run AI drafting without handing the whole work graph to a hosted vendor.
For admins
Separate platform defaults from workspace overrides and keep key rotation visible.
For agents
Route tool writes through identity, policy, approval, and audit instead of raw API access.
Run it yourself
Pull the image, keep the data, wire your own AI keys.
TaskNebula is positioned as the self-hostable project-management control plane for teams that want inspectable operations, not another black-box SaaS.