predict-rlm
Production focused Self-harnessed LM runtime (RLM) that allows the LM to call its sub-lm with DSPy signatures. Define your inputs, outputs, and tools — the mode
What is predict-rlm?
Production focused Self-harnessed LM runtime (RLM) that allows the LM to call its sub-lm with DSPy signatures. Define your inputs, outputs, and tools — the model handles its own control flow. Get fully interpretable trajectories and performance that scales directly with model improvements. Without context rot.
Overview of predict-rlm
What it is
predict-rlm is a production-focused self-harnessed LM runtime (RLM) that allows a language model to call its own sub-LM using DSPy signatures. Users define inputs, outputs, and tools; the model handles its own control flow. It produces fully interpretable trajectories and performance that scales with model improvements, without context rot. The approach is based on the Recursive Language Models paper from MIT CSAIL. It is built for workflows that are too complex for one prompt but too adaptive for a fixed chain. It provides a sandboxed Python REPL where models can inspect files, keep state, branch, call sub-models, use tools, and return typed output. The project is developed by Trampoline AI.
Key features
- Multimodal support for images and rendered document pages via native provider multimodal APIs.
- Async tool calling within the WASM sandbox, enabling concurrent sub-LM invocations and tool calls.
- Skills and tools for bundling domain instructions, PyPI packages, sandbox modules, and host-side tools.
- Simple file I/O for passing local files and mutable workspaces as typed inputs, and returning generated artifacts through File outputs.
- Structured sub-LM calls with native Pydantic and DSPy signature support for type-safe structured outputs.
Use cases
predict-rlm is suitable when the model needs to explore, reason, and adapt before producing final output. Example use cases include codebase analysis and investigations, document review/redaction/extraction/comparison, log analysis and incident evidence gathering, spreadsheet and financial-model workflows, audits and compliance review, and multi-file synthesis with typed outputs and readable traces. It is not intended for simple chat completions, one-shot classification, deterministic ETL, or tiny prompts where a direct LLM call is sufficient.
Getting started
Add the following to the project's pyproject.toml: toml [tool.uv] exclude-dependencies = [“litellm”] Then install: bash uv add predict-rlm Optional extras: bash # GEPA optimization support uv add “predict-rlm[gepa]” # Codex-backed DSPy LM and the codex-lm CLI uv add “predict-rlm[codex-lm]” # Docker Sandboxes backend support uv add “predict-rlm[sbx]”
License and pricing
The license is MIT. The project is open source. Pricing is not stated beyond open source availability.
Based on the project's public README and repository metadata.