Where we are, what’s next, and where we’re headed. Transparent by design — because a governance platform should practice what it preaches.
We build in public. Every feature on this roadmap exists as a tracked issue with a full specification. Besides extensive internal testing, we use three independent external code auditor systems to verify our codebase before every release.
The roadmap below reflects reality. V1.0.2 and V1.0.3 were planned but skipped — we put their scope directly into V1.1.0 and ship earlier. V1.1.0 also ships with significantly re-scoped content vs. what was announced: some features (Brainpool, Translation Agent, Documentation Site, Community Skill System, MCP Server Mode, Agent Training Pipeline, and user-facing Chat Memory) are moved to V1.2. In exchange, V1.1.0 delivers critical infrastructure that was not announced: LLM Tool Calling, Webhook Inbound, Blue/Green Updates, Prometheus Observability, Migration System, and a deep security-hardening pass that consumed more engineering time than the announced V1.1 feature set combined.
We prefer a longer, honest roadmap over a shorter, optimistic one.
Every agent action is checked against your rules before it executes. A 5-step pipeline with 10 foundational invariants ensures nothing slips through.
Set spending limits per agent, per division, or globally. Real-time cost monitoring across all providers — no more surprise API bills. Hard enforcement in V1.0.1.
A full web dashboard to monitor agents, review audit trails, and manage governance rules. Installable as an app on any device — no separate software needed.
Talk to your agents where you already are: Email, Telegram, or Discord. Reply to an email, send a Telegram message, or use a Discord channel — your agents respond on whatever channel you choose. Works even in China and restricted networks.
Use Claude, GPT, Gemini, Llama, Mistral, DeepSeek, or local models via Ollama. Switch providers per agent without changing your governance setup.
Three commands to install. A built-in Guide Agent walks you through provider setup, team configuration, and your first task step by step. Available in 26 languages. No YAML editing, no config files — just answer the questions.
Coming from OpenClaw or MoltBot? A built-in importer brings your existing agent configurations into SIDJUA with a single command — no manual rewriting needed.
Every action, every decision, every cost — logged and searchable. When your compliance officer asks what happened yesterday, you’ll have the answer in seconds.
Runs on your own hardware. No cloud dependency, no data leaving your network. Works in fully disconnected environments with local LLMs.
Agents persist as governed background daemons — not session-bound, not reactive. Proactive task detection, configurable schedules, and daemon-level budget limits (cost/hour, tasks/hour).
Two system agents cross-monitor all agents and each other. Automatic restart with exponential backoff, circuit breaker, and systemd failsafe escalation to human notification.
Normal computer shutdown? SIDJUA detects the OS signal and saves everything automatically — zero data loss. Hard crash or power failure? Silent checkpoints every 60 seconds protect your work. At most one minute of in-flight work needs to restart. SIDJUA auto-detects whether it runs on a desktop or a 24/7 server and adjusts accordingly.
Recurring tasks via standard cron expressions with per-schedule budget limits, max-runs-per-day caps, and automated deadline escalation.
AGPL-3.0 for open source use. Commercial license for enterprises that need proprietary deployment. Both editions share the same core — no feature gating.
Secrets API defaults changed from operator-level to fail-closed. Hostname-derived encryption key replaced with random master key.
Scheduler budget checks upgraded from fail-open to fail-closed. Agents that exceed their budget are now blocked, not just warned.
Division isolation wired end-to-end. IPC authentication between agent processes. Hardened bootstrap sequence.
The HR Agent now asks the right questions when creating agents: Which team leader supervises this worker? Which division does a new manager belong to? Who reports to them? Proper org structure from day one.
Assign specific skills and tools to your agents through the HR Agent. Tell HR your video editor needs DaVinci Resolve expertise, and it updates the agent’s capabilities and skill profile accordingly.
Agents created through the HR Agent are now fully registered in the database, visible in the team dashboard, and operational immediately — no manual commands needed.
API keys injected into the management console at serve time, eliminating the bootstrap endpoint entirely. Cleaner architecture, fewer attack surfaces.
Llama tool call parsing, agent list auto-refresh, audit CSV export, provider mismatch warnings, home navigation, and tool call display — all polished for a smoother experience.
Full Gate-4 audit action taken across 135 files: 2 116 insertions, 441 deletions. Plus dual-audit review — 24 independently verified security findings addressed. The largest single security pass in SIDJUA’s history.
Every agent action now passes seven pre-execution gates. A tool-call parameter filter redacts sensitive data before it reaches the LLM. Input sanitizer blocks prompt-injection patterns. Redaction approval persistence lands in V1.2.
Session-cookie ordering hardened. CSRF protection on raw fetch flows. Cryptographic UUIDs replace weak random. Save-activate connection-status stabilised. Settings backup button wired correctly. End-to-end authentication chain reviewed.
Cross-division messaging restricted at adapter and gateway layers. IPC authentication between agent processes with 0600 socket permissions. Module capability checks enforced at load time.
Path traversal guards and real-path normalisation. Symlinks with non-existent targets rejected. Bootstrap key migration, CSRF protection on secrets routes, and secret audit pipeline hardened. CLI governance gate closes historical bypass paths.
Full reasoning loop with provider-side tool use. The LLM itself decides which tool to call, with memory verification and context budget tracking built in. Works across Claude, GPT, Gemini, Llama, Mistral, DeepSeek, and local Ollama — tool-use is abstracted, not hardcoded.
Any external system can trigger agent actions via webhook — governed, rate-limited, and token-authenticated. Per-webhook token store, REST routes, and CLI management included. Outbound webhook governance follows in V1.2.
Agent freeze/resume lifecycle, drain middleware, and live update and maintenance banners in the GUI. Updates happen without killing in-flight agent work. Full three-layer rolling upgrade (migration-unit, fixture-replay, container-rehearsal) lands in V1.1.1.
A new Update panel in Settings / About with configurable auto-check interval. See available updates directly in the console — no terminal required. The updater container cleanly shuts itself down after a verified successful update.
Prometheus-compatible /metrics endpoint and a bundled Grafana dashboard ship in V1.1.0. This is the foundation — full OpenTelemetry integration (traces and logs, not just metrics) is coming in future versions.
Schema upgrades are now versioned SQLite migrations with integrity checks and automatic backup-and-restore on failure. If a migration fails, your data rolls back without intervention.
Forty-four supported locale languages total — eighteen added during this rebuild. The entire translation architecture was rebuilt — key-gap-free, machine-verified.
Eight long-standing defects in the provider configuration UI fixed. 44 of 44 backend tests and 25 of 25 GUI tests pass. Provider switching, model family grouping, and API-key handling all work reliably across every combination.
Parser, field mapper, and executor — end-to-end. Importing an OpenClaw agent configuration into SIDJUA is now a single command. MoltBot import follows.
Qdrant REST adapter for vector-store-backed tool retrieval, plus a feature-flag-gated memory consolidation pipeline. This is the backend foundation — user-facing Chat Memory and Knowledge Store with persistent semantic search arrives in V1.2.
SIDJUA agents can now consume any Model Context Protocol (MCP) compatible external tool. YAML schema validation and full MCP client lifecycle. MCP Server Mode — exposing SIDJUA agents AS MCP tools — follows in V1.2.
A lightweight authoring API for agent modules. Third-party module development is now possible today. The fully governed community skill registry lands in V1.2.
Real-time inference output streaming across providers. Users see tokens appear as they are generated — no more waiting for full completion.
A configurable load-scenario harness for benchmarking agent runtime performance in your own deployment.
REST tool catalog with composite wiring. The HR Agent gains a dual-strategy extension, combining REST and native tools when creating or updating agents.
Documentation generator scripts, CI pipeline, HTML templates, and a PDF report generator. The foundation for V1.2’s versioned, searchable Professional Documentation Site.
tickets.sidjua.com is live with a five-tier defence stack (nginx rate-limiting, CrowdSec, bot filtering, allowlist, and Zammad-native guards). Community can open support tickets directly. Seventy-five percent of the polish items are in; remaining UI-theming and email-template work continues.
Six architectural sub-systems for the V1.2 Secrets / Consent / Tool-Call Governance epic are fully specified: Security Abstraction Layer, Tool Contract v2, Central Policy Decision Point, Consent Grant Service, Policy Migration, and Enterprise Backend Adapters. Implementation begins immediately after V1.1.0 Golden.
Organization-page agent counter, chat redaction approval persistence, multi-tab provider-config staleness, clipboard-copy over non-HTTPS, and locale-dropdown internal entries. All five documented V1.1.0 limitations resolved.
Three-layer update flow: versioned migration-unit, fixture-replay pre-flight, and container rehearsal against a rolling dev channel. The complete no-downtime, no-data-loss upgrade path that builds on V1.1.0’s Blue/Green foundation.
Expose your governed SIDJUA agents as Model Context Protocol tools. Any MCP-compatible client — Claude Desktop, VS Code, or custom apps — can call your agents with full governance applied. The counterpart to V1.1.0’s MCP Client integration.
A six-piece architectural upgrade: Security Abstraction Layer, Tool Contract v2, Central Policy Decision Point, Consent Grant Service, Policy Migration, and Enterprise Backend Adapters. Every agent action, every secret, every tool call becomes inspectable, revocable, and audit-traceable from a single control plane.
Run a professional-grade code and architecture audit on your own agent deployment with a single command. No consultants, no extra tooling. Moved from V1.1 scope.
Agents share knowledge and learn from each other’s experiences. A shared intelligence layer that makes your entire agent team smarter over time, without duplicating effort. Moved from V1.1 scope.
Agents remember past conversations and build persistent knowledge using vector embeddings. Powered by the Qdrant backend that landed in V1.1.0. User-facing semantic search across your agent’s entire history.
Agents learn their craft. After assigning skills, trigger a training run — the agent reads documentation, manuals, and best practices for its tools and builds a personal knowledge base. Re-train anytime skills change. Moved from V1.1 scope.
An optional, free agent that automatically translates documentation and content into 26+ languages. Community-extensible — add any language without code changes. Moved from V1.1 scope.
All documentation unified into a searchable, versioned website. Guides, API reference, architecture docs — everything in one place, in industry-standard format. Moved from V1.1 scope; the V1.1.0 documentation-generator is the foundation.
Extend your agents with reusable skills — built by you or shared by the community. A governed skill registry ensures every extension follows your security rules. Builds on V1.1.0’s Module SDK Light. Moved from V1.1 scope.
Optional upgrade from the built-in SQLite database to PostgreSQL for teams running hundreds of agents. Zero-downtime migration with a single command.
Stable, versioned REST API so your integrations don’t break when we ship new features. Build on SIDJUA with confidence.
Agents can create temporary sub-agents for complex tasks and merge results back. Like delegating work to a team, but with full governance over every spawned agent.
Orchestrator failover so your agent infrastructure keeps running even when hardware fails. Designed for production environments that can’t afford downtime.
Traces, logs, and metrics in a single OpenTelemetry pipeline. Builds on the Prometheus foundation shipped in V1.1.0 — exports into Grafana, Datadog, or any OTel-compatible backend.
Connect personal agents, corporate agents, and third-party agents on a single device. Each owner keeps full control over their governance rules and data.
A dedicated controller agent that monitors all spending in real-time, detects rogue behavior, and blocks unauthorized financial actions before they happen.
Destructive actions — like deleting content or revoking access — require explicit human approval. Every deletion is backed up automatically before execution.
A patented system that tracks the emotional and behavioral state of AI agents. Detect when agents are stressed, frustrated, or behaving unusually — before problems escalate.
Hardware-level agent isolation using gVisor or Firecracker micro-VMs. Each agent runs in its own secure environment, fully separated from the host system.
Run SIDJUA as a shared platform for multiple teams or departments, each with their own governance rules, budgets, and audit trails — fully isolated from each other.
Let multiple AI models collaborate on the same task and combine their strengths. Like a panel of experts instead of a single opinion.
See how your agents’ knowledge evolves over time and catch when they drift away from their intended behavior — before it affects your operations.
C-level views of your entire AI agent landscape: costs, compliance status, team performance, and risk indicators — all in one place, updated in real time.
Connect SIDJUA instances across organizations for secure cross-company agent collaboration. Your agents talk to their agents — governed on both sides.
Join our Discord community. Tell us what you need. The roadmap is a plan, not a promise — your feedback decides what moves up.