Open Agento
Open Source · MIT license

Autonomous AI agents that get the work done.

Open Agento is a platform for building and running specialized agents that pick up tasks from your business systems, complete them, and deliver finished results — fully under your organization’s control.

Security by design, not by prompt.

PROJ-1284 Fix flaky checkout test Done
Jira → Developer agent Claude Code sandbox
  1. ✓Task picked up from Jira, priority set
  2. ✓Repository cloned in isolated runtime
  3. ✓3 files changed, tests green
  4. ✓Pull request opened in Bitbucket
  5. ✓Result posted back to PROJ-1284
tools: bitbucket · jira · gitno secrets exposed to the model
15–20×lower cost than per-token API
24/7always on, no manual handoff
100parallel jobs per agent, verified
MITOpen Source, no vendor lock-in
What is Open Agento

Not a chatbot. A digital employee.

Chatbot

Waits for a question, generates an answer, and stops. A human still has to move the result between systems.

Open Agento agent

Picks up a task from Jira, Outlook, or Bitbucket on its own, gathers data, does the work with approved tools, and delivers the result back to the right system.

Each agent has its own role, instructions, knowledge, AI model, and individual permission scope — and the organization decides which data and operations it may use.

Architecture

How a task flows

From the work channel, through the queue and a specialized agent, to the secure tool layer that alone reaches your company data.

01

Work channels

JiraOutlookBitbucketSchedule
02

Task queue

Sender identification · agent selection · priority · duplicate protection
03

Specialized agents

Analyst · Developer · QA · PMO · Support · DevOps
04

AI engines

OpenAI Codex · Claude Code · Pi.dev · any model via OpenRouter.com, incl. local / EU
05

Secure tool layer

Executes operations on behalf of the agent without exposing a single secret to it.
06

Company data

Jira · Outlook · Bitbucket · MySQL · MSSQL · OpenSearch · files
07

Results

Comments · code · pull requests · reports · emails · documents

Operational oversight at every step — statuses · logs · retries · alerts · audit

Security

Security built into the architecture

Agents never receive passwords to databases, mail, or external systems. Operations run through Open Agento’s secure layer — within granted permissions.

01

Secret separation

Passwords and keys never reach the model. Only the tool layer knows them.

02

Permission-based access

Read-only, allow-lists of recipients, domains, repositories, and tools.

03

Isolated runtime

The agent runs in an isolated sandbox, with no direct access to databases.

04

Configurable scope

Each agent has an individual permission profile tailored to its role.

05

Full audit trail

Every job has a status, history, attempt count, and result — with retries.

Principle

Security by design, not by prompt.

Use cases

One platform, many roles

Agents complete whole tasks and deliver a concrete result to the right system.

Jira & processes

Issues, statuses, comments, assignments

Software development

Code, tests, commit, branch, pull request

Code review

Fixes, replies to reviewers, PRs

Business analysis & PMO

Specs, backlog, ADR, RAID, reports

QA & testing

Testing web and mobile apps with artifacts: screenshots, videos, reports

Data analysis

MySQL, MSSQL, OpenSearch, CSV, Excel, PDF

Support & diagnostics

Log analysis, reproduction, fix

Email handling

Reading, replies, attachments in Outlook

Monitoring & follow-up

Recurring tasks, alerts, verification

Business value

What your organization gains

Lower cost

Subscription licenses for leading agents instead of per-token API — 15–20× cheaper.

Always on

Tasks around the clock, with no manual handoff between systems.

No vendor lock-in

Open Source under the MIT license — can be maintained in-house.

Parallel work

Verified up to 100 simultaneous jobs handled by each agent.

Local / EU model

Open-weights model hosted locally or in an EU data center — or any of hundreds of models via OpenRouter.com.

Faster time to result

From request to finished work — without waiting in a team queue.

Repeatable quality

Tasks executed to the organization’s instructions, every time.

Lower risk

Permission-based access and a full audit trail for every operation.

Full oversight

Statuses, logs, retries, and alerts for every job, from day one.

Open Source

Yours. On your terms.

Open Agento is available under the MIT license. Run it on your own infrastructure, audit the code, extend it with your own modules, and maintain it in-house — with no dependence on a single vendor.

  • MIT license. Full freedom for commercial use and modification.
  • Self-hosting. Runs on your infrastructure or in an EU data center.
  • Module system. Extend with your own channels, tools, and AI engines — Codex, Claude Code, Pi.dev, or any OpenRouter model.
QUICK STARTPython 3.12+ · Docker
# install the CLI
$ uv tool install agento-core

# create a project, run the installer
$ mkdir my-agents && cd my-agents
$ agento install
✓ cron · toolbox · sandbox containers ready
Next step

Launch an agent on one of your processes.

We’ll propose a pilot on a selected process — with a clear permission scope, a measurable result, and full oversight from day one.