Research

Subconscious is an AI lab solving the context window to make open models far more capable.

Inference and post-training

We research two core systems for long-horizon agents.

Inference and post-training, designed with long-horizon agents as the primary workload.

Inference

OrangeLine

Our inference runtime for long-horizon agents. It manages context on the GPU while a trace runs: compressing the KV cache, reusing prefixes and suffixes, and freeing memory mid-run, so agents reason further and faster on the same hardware.

Post-training

RedLine

Our post-training system for long-horizon agents. We RL-train open models on long, tool-heavy trajectories so they learn to manage their own context and use the runtime well, rather than fighting the context window.

Subconscious visualized

Longer runs and more concurrency on the same GPUs.

The runtime manages memory and context during long agent runs. The gains land exactly where agentic workloads hurt the most.

Handle millions of tokens with context management at runtime.

With the highly efficient Subconscious Cache, save 10x on cost at scale.

Run 2.3x as many workloads on the same compute footprint.

3.5x faster token throughput down deep reasoning chains.

SGLang

Context windowTokens processedmax

Limited to the model context window with compaction. Long tasks run with unnecessary context or compact necessary information.

Subconscious

Context windowTokens processedmax

Processes millions of tokens per run. The system compresses its context at runtime, so accuracy stays high at any scale.

OrangeLine

An inference system designed for agents

General-purpose inference systems waste compute on agent workloads. We built OrangeLine, a runtime that treats agent traces as the primary workload for better performance and reliability.

Featured · June 2026

Subconscious Cache

Our system for KV cache pruning with preserved suffix states. When a span of context is pruned, the information it contributed survives in the retained suffix states instead of being thrown away and re-encoded.

50%
Lower latency per task
2.3×
Supported concurrent workloads
3.5×
Faster token generation
Read the post

Auto Compaction

Our runtime enables models to prune their own message list at inference time. Compaction happens automatically inside the familiar chat completions and messages APIs, and the Subconscious Cache preserves memory across those prunes.

Read the post

OrangeLine in Production

This is more than research, it's live in production today. Customers utilize our inference runtime via API and on their own GPU clusters.

Read the post

Standard runtime · prefix cache only

System + tools
Pruned
Recent steps
Output

Everything after the pruned span is re-encoded from scratch.

Subconscious OrangeLine · prefix + suffix cache

System + tools
Pruned
Recent steps
Output

The suffix is reused too. Some of the pruned token information is preserved in the suffix.

Cache hitRe-encodedPruned by the agent

Each strip is one agent context after a mid-conversation prune. A standard prefix cache only reuses tokens before the pruned span, so everything after it is re-encoded. OrangeLine also reuses the suffix, keeping most of the context cached.

Post-training designed for long-horizon agents

RedLine

OrangeLine alone is powerful. RedLine is how open models learn to use it even better.

  1. 01

    Co-designed with the runtime

    RedLine uses OrangeLine under the hood and trains the models to exploit context compression and suffix reuse.

  2. 02

    Highly efficient post-training

    With context compression during training, RedLine requires up to 75% less compute to train on huge reasoning trajectories. That cost saving flows straight through to our customers.

  3. 03

    SFT, OPD, and RL on agentic workloads

    We use it to run SFT, OPD, and RL on open-weight models on long-horizon, tool-heavy tasks. Coding agents, browser control, workflow automation, research tasks, and more.

  4. 04

    For our TIM models, or yours

    RedLine is how we create our TIM family of models, but also how we post-train models for customers on their unique data and tooling.

Open collaboration

Contact us to collaborate

Our research is informed by our connections to the academic community and our commitment to publishing our findings. Today we collaborate with labs at MIT, Princeton, Harvard, Carnegie Mellon, Tel Aviv, and Berkeley.

For research collaborations, academic partnerships, or questions about our work, reach out to hongyin@subconscious.dev.