Top Sites>Iimbue.com

Not stored for this site.

# Imbue

> Imbue builds powerful computing tools controlled by individuals — empowering humans in the age of AI.

Imbue (formerly Generally Intelligent) was founded in 2021 by Kanjun Qiu and Josh Albrecht. The company's mission is to democratize software creation so that every person can shape their own computational future. Rather than selling AI that serves its creators' interests, Imbue builds tools that let individuals create, customize, and fully control their own AI agents and software.

Imbue's flagship product is Sculptor, a coding agent environment that embeds software engineering best practices — sandboxed execution, automated testing, parallel issue fixing — directly into AI-assisted development. The company also conducts original research on reasoning models, hyperparameter optimization, and code evolution, and publishes policy work on AI safety and individual rights.

Imbue raised a $200M Series B in 2023 at a valuation over $1 billion, with backing from Astera Institute, NVIDIA, Kyle Vogt, Simon Last, and others. The company operates an ~10,000 H100 cluster and trains its own foundation models.

## Main Pages

- [Home](https://imbue.com/): Overview of Imbue's mission, product, and recent work.
- [Our Work](https://imbue.com/our-work/): Products, research, and policy work.
- [Company](https://imbue.com/company/): Mission, core beliefs, team, advisors, and press coverage.
- [Careers](https://imbue.com/careers/): Open positions (pulled live from Greenhouse).
- [Blog](https://imbue.com/blog/): All posts across research, product, company, and policy.

## Product

- [Sculptor](https://imbue.com/sculptor/): Imbue's coding agent environment. Runs code in sandboxes, suggests and fixes issues in parallel, supports custom checks (LLM prompts and shell commands), and integrates with editors and linters (pylint, ruff, pyre). Currently in early research preview.
- [Sculptor: Catch and fix issues as you code](https://imbue.com/product/sculptor/): Product announcement and detailed walkthrough of how Sculptor works.
- [Sculptor Changelog](https://imbue.com/sculptor/changelog/): Running log of Sculptor updates.
- [Vet: Prevent coding agents from making mistakes](https://imbue.com/product/vet/): Open-source tool that verifies AI-generated code. Available at github.com/imbue-ai/vet.
- [How we made sandboxed coding agents 10x faster to start](https://imbue.com/product/containers/): Technical post on container infrastructure for safe agent execution.
- [Use plain old REST instead of MCP](https://imbue.com/product/latchkey/): Post on Imbue's agent communication approach.

## Research

- [Beating ARC-AGI-2 with Code Evolution](https://imbue.com/research/2026-02-27-arc-agi-2-evolution/): Imbue's code evolution method achieves 34% on ARC-AGI-2 with open-weight models and 95% with Gemini 3.1 Pro — a 2x–3x improvement over base model performance. (2026)
- [LLM-based Evolution as a Universal Optimizer (Darwinian Evolver)](https://imbue.com/research/2026-02-27-darwinian-evolver/): Description of the open-source Darwinian Evolver framework used for prompt and code optimization via fitness-based sampling and mutation. Available at [github.com/imbue-ai/darwinian_evolver](https://github.com/imbue-ai/darwinian_evolver). (2026)
- [Training a 70B model from scratch](https://imbue.com/research/70b-intro/): Overview of Imbue's 70B parameter model pre-trained on 2T tokens, which outperforms GPT-4o zero-shot on multiple-choice reasoning benchmarks. Includes release of evaluation datasets, infrastructure scripts, and CARBS. (2024)
- [Sanitized open-source evaluation datasets](https://imbue.com/research/70b-evals/): Cleaned versions of 11 NLP reasoning benchmarks plus a new CodeComprehension benchmark; 450,000 human judgments on question quality. (2024)
- [From bare metal to high-performance training: infrastructure](https://imbue.com/research/70b-infrastructure/): Scripts and best practices for standing up a large-scale H100 training cluster. (2024)
- [Open-sourcing CARBS: cost-aware hyperparameter optimization](https://imbue.com/research/70b-carbs/): How Imbue used CARBS to scale to 70B parameters with minimal instability. (2024)
- [Scaling Laws For Every Hyperparameter Via Cost-Aware HPO (CARBS)](https://imbue.com/research/carbs/): Original CARBS research: cost-aware hyperparameter search that predicts optimal configurations across model sizes using scaling laws. Available at [github.com/imbue-ai/carbs](https://github.com/imbue-ai/carbs). (2023)
- [Avalon: A Benchmark for RL Generalization](https://imbue.com/research/avalon/): Procedurally generated game environment for evaluating RL generalization. NeurIPS Spotlight. (2022)
- [On the Stepwise Nature of Self-Supervised Learning](https://imbue.com/research/ssl_stepwise/): Theoretical analysis of how self-supervised learning progresses in discrete steps. (2022)
- [The state of LLM ethical decision-making](https://imbue.com/research/llm-ethics/): Evaluation of how frontier LLMs handle ethical dilemmas. (2023)
- [Update: The state of LLM ethical decision-making](https://imbue.com/research/llm-ethics-update/): Follow-up evaluation as models improved. (2023)
- [VICReg: Tutorial and PyTorch implementation](https://imbue.com/research/2022-04-21-vicreg/): Tutorial on the VICReg self-supervised learning method. (2022)
- [A PyTorch implementation of SIMONe](https://imbue.com/research/2022-04-14-simone/): Implementation of slot attention for object discovery. (2022)

## Company

- [Empowering humans in the age of AI](https://imbue.com/company/vision/): Imbue's core manifesto by CEO Kanjun Qiu. Argues that the central challenge of AI is managing power concentration, and that giving everyone the ability to build and control their own software is the path to a democratic AI future. (2025)
- [A more radical Imbue](https://imbue.com/company/a-more-radical-imbue/): CTO Josh Albrecht's 2026 all-hands address on returning to Imbue's roots: ultra-high agency culture, open software projects, and empowering every team member to be a DRI. (2026)
- [Imbue raises $200M to build AI systems that can reason and code](https://imbue.com/company/introducing-imbue/): Series B announcement. Describes Imbue's full-stack approach (models, agents, interfaces, tools, theory) and the rename from Generally Intelligent to Imbue. (2023)
- [Imbue Raises Fresh Capital, Amazon's Alexa Fund Joins Investor Lineup](https://imbue.com/company/imbue-raises-fresh-capital/): Additional $12M from Alexa Fund and Eric Schmidt. (2023)
- [In-person culture and habits](https://imbue.com/company/2023-09-01-in-person-culture-and-habits/): Description of Imbue's working culture: rituals, habits, and how the team collaborates in person. (2023)
- [Our technical hiring process](https://imbue.com/company/2023-09-06-our-technical-hiring-process/): How Imbue evaluates and hires engineers and researchers. (2023)
- [Choices and Knives](https://imbue.com/company/choices-and-knives/): Essay on decision-making culture at Imbue. (2023)
- [Team spotlight: Bryden Fogelman](https://imbue.com/company/2023-09-06-bryden-team-spotlight/): On coding agents and finding meaningful work. (2023)
- [Team spotlight: Glenn McDonald](https://imbue.com/company/2025-01-30-glenn-team-spotlight/): On data alchemy and radically open software. (2025)

## Policy

- [Our approach to safety](https://imbue.com/policy/safety/): Imbue's framework: agents conditioned entirely by end-users' goals, natural-language reasoning, and regulatory engagement. (2024)
- [Towards guardrails, not guidelines: a policy framework for powerful AI systems](https://imbue.com/policy/common-good/): Imbue's policy philosophy — enforceable rules over voluntary commitments. (2024)
- [Digital Freedom Depends on Access Rights (Lawfare)](https://imbue.com/policy/lawfare-piece/): Op-ed arguing individuals must have the right to access their own data from corporate silos. (2024)
- [Where are the good policy ideas? Analyzing the NTIA's AI RFC](https://imbue.com/policy/ntia-rfc-analysis/): Analysis of public comments on the NTIA's AI accountability framework. (2023)
- [The need for trust: Analyzing individual NTIA RFC comments](https://imbue.com/policy/ntia-rfc-analysis-individual/): Deeper analysis of individual submissions. (2023)
- [Our submission to the NTIA's AI accountability RFC](https://imbue.com/policy/ntia-rfc/): Imbue's formal policy submission. (2023)
- [A Healthy Ecosystem for AI Agents: Introducing the AFI](https://imbue.com/policy/afi/): Imbue's proposal for an Agent Fairness Infrastructure to protect individuals from powerful AI agents. (2024)
- ['We need legal protections from AI's risks' (The Hill)](https://imbue.com/policy/legal-protections-op-ed/): Op-ed by Matt Boulos on legislative requirements. (2024)

## Open Source

- [darwinian_evolver](https://github.com/imbue-ai/darwinian_evolver): Problem-agnostic code and prompt evolution framework. Used to achieve state-of-the-art ARC-AGI-2 results with open-weight models.
- [CARBS](https://github.com/imbue-ai/carbs): Cost-Aware Random Search for hyperparameter optimization. Predicts optimal configurations via scaling laws.
- [vet](https://github.com/imbue-ai/vet): Verify Everything — open-source tool to prevent coding agents from making mistakes.
- [cluster-health](https://github.com/imbue-ai/cluster-health): Infrastructure scripts for standing up a high-performance H100 training cluster from bare metal.
- [Evaluation datasets on Hugging Face](https://huggingface.co/imbue): Cleaned NLP reasoning benchmarks, CodeComprehension benchmark, human question-quality judgments, and a fine-tuned 70B question-quality model.

## Podcast

- [Generally Intelligent Podcast](https://imbue.com/podcast/): ~38-episode series (2020–2024) featuring conversations with AI researchers including Noam Brown, Chelsea Finn, Sergey Levine, Martin Arjovsky, Jacob Steinhardt, Jonathan Frankle, Jim Fan, and others.

## Contact & Social

- Press inquiries: press@imbue.com
- Twitter/X: [@imbue_ai](https://x.com/imbue_ai)
- Newsletter: https://bit.ly/subscribe-to-imbue-emails