- Category
- AI Agents
- Rank
- No. 3993Tools index
Previous survey · No. 3757 ·
- Pricing
- Open Source
- Type
- AGENT
- Use case
- Coding
- GitHub
- 4 stars
- Latest release
- v0.8.1-beta
- Date
About
Loa is a local-first autonomous coding agent for project-wide engineering tasks and long-running work. Its control layer uses explicit state transitions, structured tool use, persistent task state, memory retrieval, evaluation and recovery. For complex work it builds a mutable directed acyclic graph that serves as both the execution plan and the context topology, so each step's dependencies determine which completed results are passed in as prerequisites.
At a glance
Loa is a coding agent that plans complex work as a graph of dependent steps and can revise the plan as it goes. It is worth evaluating when you want to inspect how a long task is broken down and can run inference on infrastructure you control.
Good fit
Developers with a local or self-hosted model endpoint and a test repository for evaluating autonomous changes.
Consider something else when
Pass if you need a lightweight assistant for short edits or rely on a metered endpoint with a tight call budget.
Start here
Give it a small two-step refactor in a disposable repository. Inspect how the second step uses the first result and record the inference calls before trying a larger job.
Cost and requirements
Needs an OpenAI-compatible inference endpoint. Docker isolation is optional, and planned runs and indexing can make many model calls.
Limits to know
The optional container mounts the target project read-write. Shell tools have the permissions of their environment, and configured remote endpoints receive model inputs.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Loa structures agent work as a mutable dependency graph rather than a single open-ended loop, giving long-running coding tasks explicit state, recovery and context topology instead of ad hoc prompt chaining.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.