Free trial
Every plan is free for your first month, and a free account is available. Cancel anytime.
Learn moreAn engineering harness that governs coding agents and any model
Palgu is an engineering harness for AI coding agents. It routes tasks to models by risk, enforces team standards, validates and independently reviews changes, and records an audit trail, via a terminal CLI or the Palgu Studio desktop app.
Palgu positions itself as an engineering harness rather than another coding agent. Instead of writing code itself, it sits around agents such as Claude Code, Codex, Cursor, aider or its own agent, and manages how a task moves from a vague request to a tested, reviewed and recorded change. Its pitch is that agent orchestrators only run more agents, while Palgu applies one set of routing, budgets, policy, review and audit across all of them. In practice a developer signs in once and works from the terminal with the palgu command, or from Palgu Studio, a desktop app for macOS and Windows that groups tasks, a Validation Center, a Council Chamber, knowledge and routing. A single request triggers a sequence the site walks through with a refund-feature example: mapping the blast radius of a change, pulling architecture decision records, grading risk, requiring tests before implementation, choosing a model by task size and risk, validating functionally and visually, convening an independent multi-seat review by models that did not write the code, and logging the outcome in an audit ledger. The palgu -p flag runs a single task and exits, which suits CI and scripts, and palgu review exits with an error when the review council requests changes. The audience is engineering teams and platform teams that want agent-written code to follow their own standards. Deployment options include SaaS or a self-hosted gateway, bring-your-own keys, local or sovereign models, SSO through OIDC and SAML, audit export and data residency. The vendor says its servers never run customer code and that the audit log stores prompt hashes rather than prompt text. Among alternatives, Palgu competes less with single coding assistants than with orchestration and governance layers around them. It is aimed at teams who value policy control, model switching and proof of validation over a lightweight individual setup. The vendor also publishes benchmarks in which six models passed the same acceptance tests inside the harness, though those figures are self-reported.
Every plan is free for your first month, and a free account is available. Cancel anytime.
Learn moreFree account, and every plan is free for the first month, then from $9 per seat. Cancel anytime. Credits can be bought for managed models, or users can bring their own keys.
Learn moreDocumentation is available on the site, along with a platform-team checklist for deployment. The homepage does not describe other support channels.
Learn moreRoutes Claude Code and Codex via palgu enable; Cursor and aider connect to its endpoint. Supports frontier, open and local models (Ollama), BYOK, SSO (OIDC, SAML), CI runners, and SaaS or self-hosted gateway.
Learn moreLive code graph and architecture context, risk grading, tests-first policy, model routing by task and risk, functional and visual validation, multi-seat independent review council, audit ledger, budgets, allowlists, and CLI plus Studio desktop app.