Atlas is built to run more than one coding session at the same time without the sessions overwriting each other's work or state. Each session runs as its own runtime instance keyed to the directory it works in, and any state that sessions share is protected by locks so two sessions can never half-write the same record. That means you can keep several agents or terminals going at once, including across different git worktrees of the same repository, and each one stays consistent. Atlas runs as a single self-contained binary.
The problem with concurrent AI sessions
Many terminal agents assume one active session per project. Open a second session and they can contend for the same files and state, so one session's action overwrites another's. Atlas is designed for concurrency from the start.
When two sessions share a single project directory, an action in one, such as writing history, saving config, or removing a working copy, can silently overwrite or invalidate what the other is doing. The failure shows up as lost edits, corrupted session state, or a crash. Atlas treats concurrent sessions as the normal case rather than an edge case, so parallel work is safe by construction.
Per-directory runtime isolation
Atlas keys each live session to the working directory it operates in and gives that directory its own runtime instance. Multiple instances coexist in one process, each with its own language servers, snapshots, and permission boundary.
Because each directory, including each git worktree checkout, gets its own isolated instance, two sessions running in different places do not share the live runtime state that would otherwise collide. Concurrent loads of the same instance are deduplicated rather than raced, so starting several sessions at once is deterministic instead of a scramble.
Locking that prevents state bleed
For the durable state that sessions genuinely share, Atlas serializes access with locks. In-process reentrant read and write locks guard each storage record, and a cross-process file lock coordinates work across separate Atlas processes.
The cross-process lock is a careful directory-based lock with heartbeats, stale-owner recovery, and process metadata, used for config writes, plugin installs, and other shared operations. The combination means concurrent sessions in different worktrees still write into shared repo state safely: the locks make that contention correct rather than destructive. Each instance also keeps its own permission boundary, so isolation holds for what an agent is allowed to do, not just for the files it writes.
Frequently asked questions
- Can I run multiple Atlas sessions at the same time?
- Yes. Atlas is built for concurrency: it runs multiple sessions in one process, each isolated by working directory, and guards shared state with locks.
- Will one session overwrite another session's work?
- No. Each session runs as its own per-directory instance, and shared writes are serialized by in-process and cross-process locks, so sessions do not clobber each other.
- Does this work across git worktrees of the same repo?
- Yes. Sessions in different worktrees get separate runtime instances, and any state they share in the repo is lock-protected, so concurrent worktree sessions stay consistent.
- Is the concurrency safe on Windows?
- Yes. Atlas runs natively on Windows, and its locking and per-directory isolation apply the same way as on macOS and Linux.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas with Mistral Medium 3.1 (2508): Setup, Cost, and Fit in 2026
Run Atlas on Mistral Medium 3.1 (2508): a 262,144 token window at $0.40 / 1M input tokens and $2.00 / 1M output tokens. Setup steps, costs, and honest tradeoffs.
Atlas with Command R: The $0.15 per Mtok Background Model for 2026
Command R runs Atlas background work at $0.15 per Mtok input and $0.6 per Mtok output, with a 128,000 token context and a 4,000 token output cap on patches.
Atlas with Phi-3 Medium 14B (Ollama): 128K Context at 7.9GB in 2026
Phi-3 Medium 14B (Ollama) gives Atlas a 128K tokens (131,072) window from a 7.9GB download, Free (self-hosted). Pull phi3:14b, not :medium-4k. Setup and tradeoffs.
Atlas vs JetBrains AI Assistant: Terminal AI Coding Agents in 2026
Atlas and JetBrains AI Assistant comparison for 2026. Atlas offers a terminal-native TUI and BYO model keys, while JetBrains AI Assistant integrates into IDEs with credit-based cloud models.
Atlas with MiniMax-M2.1 in 2026: A Free Upgrade Over M2
MiniMax-M2.1 runs Atlas at $0.30 per Mtok input and $1.20 per Mtok output with a 204,800 token context and 131,072 max output. Setup, tradeoffs, and when to move on.
Atlas with Qwen Plus: 1M Context for $0.40 per Mtok in 2026
Run Atlas on Qwen Plus in 2026. Alibaba's mid tier gives 1M tokens (1,000,000) of context with reasoning at $0.40 per Mtok input, $1.20 per Mtok output.
Atlas with Qwen2.5-Coder 7B (local via Ollama): the Laptop Setup in 2026
Qwen2.5-Coder 7B runs Atlas on a laptop with no discrete GPU: about 5GB at 4-bit, a 32,768 token context, and free self-hosted. Setup, limits, and when to upgrade.
Atlas with Amazon Nova Lite in 2026: 300K Context on Your AWS Bill
Amazon Nova Lite drives Atlas at $0.06 per Mtok input and $0.24 per Mtok output with a 300K token context, billed through IAM with no new vendor API key to manage.