Evaporate complexity
Specialized agents
orchestrate pipelines for design, implementation, testing,
documentation and deployment. Agents learn from history and optimize
costs automatically.
100% self-hosted.
Developers jump between tools constantly. Vapora unifies everything in one intelligent system where context flows.
Decisions lost in threads, code scattered, docs unmaintained. RLM (Recursive Language Models) with hybrid search (BM25 + semantic) and chunking makes knowledge discoverable even in 100k+ token documents.
Orchestrating code review, testing, documentation and deployment manually creates bottlenecks. Multi-agent workflows solve this.
Handoffs between developers and operations lack visibility and context. Vapora maintains unified deployment readiness.
82 tests verify agent orchestration, learning profiles, and task assignment. stable_id = role ensures learning expertise survives restarts and hot-reloads. SIGHUP + POST /reload for zero-downtime config updates. Agents track expertise per task type with 7-day recency bias (3× weight).
53 tests verify multi-provider routing (Claude, OpenAI, Gemini, Ollama), per-role budget limits, cost tracking, and automatic fallback chains. Swarm coordination with load-balanced assignment using success_rate / (1 + load) formula.
Process 100k+ token documents without context limits. Hybrid search combines BM25 (keywords) + semantic embeddings via RRF fusion. Intelligent chunking (Fixed/Semantic/Code) with SurrealDB persistence. Perfect for large codebases and documentation.
Distributed agent coordination with task dispatch, status tracking, and result collection. Real SurrealDB persistence (no in-memory HashMap). NATS messaging for async completion. Exponential backoff retry with circuit breaker. 12 integration tests verify real behavior.
Temporal execution history with causal relationships. Hybrid search: HNSW (SurrealDB 3 native ANN) + BM25 full-text fused via Reciprocal Rank Fusion (k=60) — exact keyword matches and semantic proximity combined. Learning curves from daily windowed aggregations. 28 tests verify graph persistence, hybrid retrieval, and execution tracking.
Reliable message delivery for agent coordination. JetStream streams for workflow events, task completion, and status updates. Graceful fallback when NATS unavailable. Background subscribers with DashMap for async result delivery.
Multi-model database with graph capabilities. Multi-tenant scopes for workspace isolation. Native graph relations for Knowledge Graph. All queries use parameterized bindings for security. SCHEMAFULL tables with explicit indexes.
40+ REST endpoints (projects, tasks, agents, workflows, swarm). WebSocket real-time updates. MCP gateway for external tool integration and plugin system. Multi-tenant SurrealDB scopes. Prometheus metrics at /metrics. 161 tests verify API correctness.
161 backend tests + K8s manifests with Kustomize overlays. Health checks, Prometheus metrics (/metrics endpoint), StatefulSets with anti-affinity. Local Docker Compose for development. Zero vendor lock-in.
Cron-triggered workflow execution with IANA timezone support via chrono-tz. Distributed fire-lock using SurrealDB conditional UPDATE prevents double-fires across multi-instance deployments — no external lock service required. 48 tests.
Real-time alerts to Slack, Discord, and Telegram — no vendor SDKs. ${VAR} secret resolution is built into ChannelRegistry construction; tokens never reach the HTTP layer unresolved. Fire-and-forget hooks on task completion, proposal approval/rejection, and workflow lifecycle events.
Domain-optimized agent bundles — system prompt, preferred LLM model, task types, and MCP tools pre-configured per role. Three built-ins (code-reviewer, doc-generator, pr-monitor) loaded at startup via CapabilityRegistry. TOML overrides let you swap model or prompt without code changes. In-process executor dispatch via DashMap channels — no NATS required for standalone mode. 22 tests.
vapora-worktree isolates agent workspaces from the main repository. WorkspaceBackend trait supports jj (Jujutsu — stable change IDs, native workspaces) and git (worktrees). Auto-detected at runtime via detect_vcs. jj's stable change IDs survive rebase and amendment — agents always reference the same change regardless of commit history rewrites. Integration tests require real jj/git binaries.
vapora-ontology exposes VaporaOntology — loads .ontology/*.ncl (core, state, gate), queries invariant axioms, checks gate membranes, and returns open FSM dimensions. vapora-reflection runs NCL-defined execution modes (CreateAgentTask, DeployService) via stratum-reflection-core. Both support live reload without process restart. Architecture is machine-readable: agents can query architectural constraints before acting.
Built with Rust 🦀 | Open Source | Self-Hosted
Explore on GitHub →