Agenta

Agenta is an open-source workspace designed to help users build, test, and share AI agents through chat. It provides tools for human-in-the-loop approvals, version control for prompts and skills, run tracing, and team collaboration. Users can deploy agents either on Agenta Cloud or self-hosted on their own infrastructure via Docker Compose or Helm charts.

• Build custom agents through chat with shared context and integrations
• Automate agent execution with schedules and event triggers
• Human-in-the-loop approvals allowing users to approve, edit, or reject consequential actions before they ship
• Version control for prompts, skills, and tools with the ability to roll back
• Run tracing to see each step, its cost, and its outcome
• Continuous improvement by turning feedback into test cases and evaluations
• Support for multiple harnesses, models, providers, and runtimes using open standards like MCP tools
• Pre-built working templates for engineering, customer support, sales, company knowledge, and operations

Pros

• Flexible deployment options allow running agents self-hosted via Docker or Helm, or through Agenta Cloud
• Open-source foundation with an MIT license prevents vendor lock-in and lets teams keep data on their own premises
• Comprehensive tracing and evaluation tools provide visibility into agent steps, costs, and failure points
• Built-in human-in-the-loop controls enable users to review, edit, or reject actions before execution

Cons

• The Hobby plan is limited to 2 team members and 5,000 agent runs per month
• The Pro plan includes 10,000 agent runs per month with additional runs costing $5 per 10,000 runs
• Trace data retention is capped at 1 week on the Hobby plan, 1 month on Pro, and 3 months on Business
• Advanced compliance features such as SOC 2 Type II reports and SSO require higher-tier plans like Business or Enterprise

Agenta fits engineering, support, sales, knowledge, and operations teams looking to build, test, and run AI agents in production with team collaboration and governance. It is well-suited for developers and organizations wanting to deploy agents either via managed cloud infrastructure or self-hosted on their own premises. It is a poor fit for teams that do not use AI agents or those requiring enterprise governance and audit logs without moving to paid tiers like Business or Enterprise.

Agenta follows a freemium model with options for both cloud hosting and self-hosting. The Hobby plan is $0 forever and includes 2 team members and 5,000 agent runs per month. The Pro plan costs $29 per month for production teams, includes 10,000 agent runs per month, and adds $5 per additional 10,000 runs with unlimited team members. The Business plan costs $299 per month, includes 10,000 agent runs per month with the same overage rate, and adds team roles, RBAC, SSO, and a SOC 2 Type II report. The Enterprise plan is custom-priced and offers advanced controls and dedicated support. The open-source self-hosted edition is free forever, while a commercial Enterprise self-hosted tier is available with custom pricing.

Unlike platforms that lock users into specific models or closed ecosystems, Agenta uses open standards like AGENTS.md, skills, and MCP tools to let teams swap out harnesses, models, or runtimes without changing code. It combines chat-based agent creation with version control, human-in-the-loop workflows, and deep run tracing in a single workspace. Furthermore, its MIT-licensed open-source code allows teams to run the exact same software on their own infrastructure via Docker or Helm.

Agenta works with harness options including Claude Code, pi.dev, Codex, OpenCode, and Hermes. It supports providers such as Anthropic, OpenAI, Ollama, xAI, OpenRouter, and custom gateways. Runtimes include local machines, Agenta Cloud, Daytona, E2B, and Vercel. Specific tools and skills mentioned include GitHub, Linear, review-diff, classify, research-company, pull-metrics, style-check, answer-cited, draft-outreach, risk-scan, and build-dashboard.