Repolode
Submit
emcie-co

emcie-co/parlant

Python framework for building customer-facing AI agents that follow business rules, guidelines and procedures consistently by narrowing the LLM context to what is relevant each turn.

Open-source alternative to:AdaDecagonSierra

View on GitHubWebsite
Stars
18,306
Forks
1,583
Open issues
40
Last commit
16 hours ago

What is it?

Parlant is an interaction control harness for customer-facing AI agents. Instead of placing every instruction in one large system prompt, developers define behavioral guidelines, observations, tools, glossary terms and multi-step journeys in code. On each conversational turn, the engine matches only the rules and tools relevant to that turn and assembles a focused context for the LLM.

The framework lets you express dependencies and exclusions between guidelines, so conflicting rules resolve predictably. Journeys model multi-turn procedures such as booking or troubleshooting, and canned responses can restrict replies to pre-approved wording at critical moments. Tools run only when their observation matches. Decisions are traced with OpenTelemetry for auditing.

Parlant works with several LLM providers, including OpenAI and Anthropic, and can use other models through LiteLLM. It can also wrap existing workflows from frameworks such as LangGraph or LlamaIndex as tools. An official React chat widget is available for the frontend.

Key features

  • 01Behavioral guidelines defined as condition-action pairs, evaluated per conversation turn.
  • 02Dependency and exclusion relationships between guidelines to control which rules apply.
  • 03Journeys for multi-turn procedures such as booking or onboarding.
  • 04Canned responses that restrict output to pre-approved templates at strict moments.
  • 05Tools that activate only when their observation condition matches.
  • 06OpenTelemetry tracing of every guideline match and decision.

Use cases

  • Customer support agents where tone, accuracy and policy compliance must hold across many rules.
  • Sales or onboarding assistants that follow a fixed multi-step procedure.
  • Regulated domains such as finance or insurance that need auditable agent decisions.

Quick start

  1. 1

    Install the package from PyPI

    pip install parlant
  2. 2

    Start a server and create an agent in Python (see the 5-minute quickstart for the full walkthrough)

Things to know

  • Requires Python 3.10 or newer.
  • Relies on an LLM backend, so you will need access to a model provider to run agents.

Topics

#ai-agents#llm#customer-service#conversational-ai#python#prompt-engineering#observability#ai-alignment#customer-success#gemini

FAQ

What is parlant?
Python framework for building customer-facing AI agents that follow business rules, guidelines and procedures consistently by narrowing the LLM context to what is relevant each turn. Parlant is an interaction control harness for customer-facing AI agents. Instead of placing every instruction in one large system prompt, developers define behavioral guidelines, observations, tools, glossary terms and multi-step journeys in code. On each conversational turn, the engine matches only the rules and tools relevant to that turn and assembles a focused context for the LLM.
Is parlant free to use? What license does it have?
parlant is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is parlant still maintained?
Actively maintained. The last commit to parlant was on Oct 20, 2026.
What can I use parlant for?
Customer support agents where tone, accuracy and policy compliance must hold across many rules. Sales or onboarding assistants that follow a fixed multi-step procedure. Regulated domains such as finance or insurance that need auditable agent decisions.
How do I install parlant?
Install the package from PyPI: pip install parlant Start a server and create an agent in Python (see the 5-minute quickstart for the full walkthrough)
Where can I get parlant?
The source code and installation instructions are on GitHub: https://github.com/emcie-co/parlant

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects