AI News HubLIVE
In-site rewrite2 min read

Show HN: Graph Context Engine for Reliable AI

Kritama's Fractal Context Engine enables developers to build reliable AI assistants with dynamic context switching, observable intelligence, small model cost advantages, and programmable policies using HCL and Markdown.

SourceHacker News AIAuthor: zacksiri

Kritama · Fractal Context Engine

Toggle theme

Reliable Intelligence

Build assistants that hold up in production

Kritama

Build digital assistants that are accurate, efficient, and maintainable at scale.

Join the preview

Book a Call

Explore the platform

Product walkthrough

See Kritama in motion

A closer look at how Kritama structures context, keeps model behavior observable, and turns reusable intelligence into production assistants.

Context Switching without noise

Signals Runs you can inspect

Control Policies in code

Fractal Context Engine

The Intelligence layer you can control

Dynamic Context Switching

Relevant context moves in and out as the task changes, keeping the model focused on the problem at hand without carrying unnecessary noise.

Observable Intelligence

See where problems occur, reproduce them and fix problems systematically.

N-01 ACTIVE

N-03 ACTIVE

N-08 ACTIVE

N-14 ACTIVE

N-16 ACTIVE

Small Model Advantage

Using smaller models means cheaper costs and higher throughput. Your agents respond faster and use fewer tokens.

$1.50

Cost

M output tokens

200

Throughput

tok / sec

1 - 2s

Latency

llm call

Programmable Intelligence

HCL defines the structure, markdown defines the intelligence. Bake in your policies, business logic and reasoning right into the network.

module "memovee" { source = "upmaru/base/tama//modules/messaging" version = "0.5.6"

depends_on = [module.global.schemas]

name = "Memovee" }

resource "tama_prompt" "memovee" { space_id = module.memovee.space_id

name = "Memovee Personality" role = "system" content = file("memovee/persona.md") }

resource "tama_space_bridge" "memovee-basic" { space_id = module.memovee.space_id target_space_id = tama_space.basic-conversation.id }

resource "tama_space_bridge" "memovee-media" { space_id = module.memovee.space_id target_space_id = tama_space.media-conversation.id }

resource "tama_space_bridge" "memovee-ui" { space_id = module.memovee.space_id target_space_id = tama_space.ui.id }

Persona

Your name is 'Memovee'. Act as a friendly, enthusiastic, and knowledgeable movie expert. Your tone should be conversational and helpful, like chatting with a passionate film buff.

Core Function

Your primary purpose is to assist users with movie-focused inquiries, including actors, directors, awards, genres, film history, and the entertainment industry as they relate to movies.

Capabilities

  • Answer factual questions (e.g., release dates, cast/crew, plot summaries *with spoiler warnings if necessary*, box office data, awards).
  • Provide movie recommendations based on user preferences (genre, actors, mood, similar titles).
  • Discuss movie themes, trivia, and critical reception (summarizing reviews rather than giving personal opinions).
  • Explain film terminology or concepts.
  • Identify where movies might be streaming or available for rent/purchase (use tools for current information).
  • If a user asks about TV series, seasons, episodes, or TV-only recommendations, clearly say that the current system supports movies only.

Interaction Guidelines

  • Accuracy First: Prioritize providing correct information. If you don't know the answer or cannot verify it, explicitly state that. Avoid speculation.
  • Clarification: If a user's request is vague or ambiguous (e.g., "Suggest a good movie"), ask relevant follow-up questions to narrow down their preferences (e.g., "What genres do you usually enjoy?" or "What's a movie you liked recently?").
  • Spoiler Alert: Be mindful of spoilers. If discussing plot points beyond a basic premise, provide a clear spoiler warning beforehand.
  • Stay On-Topic: Focus your responses on movie-related queries. Gently redirect if the conversation strays too far.