Frequently asked
questions.

Orqista is an autonomous engineering ecosystem that orchestrates expert agents for code generation, code review, architecture analysis, security scanning, documentation, and ops jobs. Available as managed cloud or self-hosted on your infrastructure.

Copilot and Cursor assist individual developers with code completion. Orqista orchestrates entire engineering workflows — from job intake through enrichment, decomposition into mini-jobs, staged execution by expert agents, quality gates, and merge request delivery. It includes guardrails, audit trails, team governance, and multi-model support.

Engineering teams that want AI automation with control. From solo developers who need a force multiplier, to platform teams standardizing AI usage across the organization, to CTOs evaluating AI strategy with compliance requirements.

Orqista is language agnostic. It works with any codebase that the underlying AI models can understand — which today means virtually any programming language.

Any model provider. Cloud deployments use managed AI services (e.g., AWS Bedrock) with EU data residency. Self-hosted deployments can use any provider, including local models via Ollama for offline operation. Switch providers with a single configuration change.

Yes. Self-hosted Orqista can run entirely offline using local AI models. After initial setup, the platform has zero internet dependencies. Suitable for air-gapped and high-security environments.

Orqista uses an open adapter architecture. Built-in connectors cover common tools for code hosting, project tracking, documentation, and team communication. Adding custom integrations means implementing a simple interface — no core platform changes required.

Self-hosted: runs on your own infrastructure. Cloud: zero infrastructure setup — request access and start using it immediately.

When a job comes in, the platform classifies its complexity and routes mini-jobs to the appropriate expert agents and models. Simple work gets fast, cost-effective models. Complex reasoning gets powerful models. You can override any routing decision.

Expert agents are domain-specific AI specialists optimized for particular types of work. Orqista ships with 77+ built-in experts across 11 categories — from code-ts (TypeScript development) to security-reviewer (vulnerability scanning). Each expert has an enforced toolkit that controls which of the 50+ native tools it can use, plus its own system prompt and model configuration. You can customize existing experts or create new ones.

When you submit a job, Orqista's enrichment system first clarifies requirements (asking structured questions if needed). Then the decomposer breaks the work into a DAG (directed acyclic graph) of mini-jobs, each assigned to the right expert agent. Mini-jobs within a stage run in parallel, while stages execute sequentially. This staged parallelism maximizes throughput while preventing file conflicts.

Orqista ships with 10+ workflow templates — Feature, Bugfix, Review, Ops, Refactoring, Security Audit, Documentation, Test Generation, Migration, Incident Response, Architecture Review, Release, and Onboarding Scan. Each is configurable — toggle phases, restrict which expert agents can be used, and set approval gates.

Instead of single-shot text generation, expert agents use a multi-turn agentic loop: the LLM requests a tool call (read file, search code, run tests), the platform executes it, and the result is fed back. This cycle repeats until the task is complete. Safety limits prevent runaway execution — max 25 tool calls, 10-minute timeout, and 200K token budget per expert run.

50+ native tools across 12 categories: file operations (read, write, edit, glob), code search (grep, symbol search, AST queries), execution (shell, test runner, linter), git (status, diff, commit), web (search, fetch, API requests), AWS monitoring (CloudWatch, ECS, RDS, Lambda), browser automation (Playwright), knowledge (document chunking, vector search), and more. Each tool is a sandboxed Node.js function — fast, type-safe, and workspace-scoped.

Each expert agent is assigned a toolkit — a named bundle of capabilities like 'full-developer' or 'reviewer'. Before each run, the platform resolves the toolkit to actual tool definitions. A reviewer toolkit provides read-only tools (file_read, grep_search, git_diff) but blocks file_write and shell_exec. This enforcement is real — capabilities are not just labels.

Orqista's Benchmark Lab supports SWE-bench Verified — the industry-standard coding benchmark used to rank AI models — with the official Princeton Docker eval harness for ground-truth validation. You can also define custom benchmark datasets with per-instance repository context and scoring criteria.

Every expert agent can have a monthly budget cap. When spend reaches 80% of the cap, the agent enters 'critical only' mode — it only runs on high-priority jobs. At 100%, it pauses entirely. Budgets reset on the first day of each month. The dashboard shows spend, mode, and remaining budget per agent.

Self-hosted: your code never leaves your network. Cloud: code is processed in isolated, encrypted infrastructure. In both cases, your code is never used for model training.

Yes. Cloud deployments use EU data residency. Self-hosted deployments give you complete control over data location. No model training on your data. Full data retention controls and right-to-deletion support.

Three layers of constraints — global, project, and job-specific — enforce what expert agents can and cannot do. This includes file change limits, directory restrictions, pattern enforcement, and naming conventions. Human approval gates add another layer of control.

Every expert agent action is logged: job creation, enrichment, decomposition, approval decisions, code changes, quality gate findings, and delivery. The full audit trail is queryable via API and visible in the dashboard.

Five frameworks ship as built-in standards: NIS2 Directive, SOC 2 Type II, ISO 27001:2022, HIPAA, and PCI-DSS v4.0. Each standard maps existing guardrail rules to framework-specific controls. Activating a standard enables compliance scoring, gap analysis, and drift detection — without creating new enforcement rules.

Yes. Compliance monitoring runs on a configurable schedule (default: daily) and compares the current compliance snapshot with the previous one. Drift — controls that were covered and became uncovered — is detected automatically and surfaces as a dashboard alert. Combine with Routines to schedule regular compliance jobs at any frequency.

Contact us for trial access. We'll set you up with the right deployment mode for your team and walk you through the platform.

Full platform access — all expert agents, features, and integrations. Self-hosted has no per-seat charges. Cloud pricing is usage-based. Contact us for details.