Atlas Commitment Tracking reconciles what you told a client you would do against what is actually in the repository. It takes the commitments extracted from your calls and checks them against your code and commit history, then sorts them into 4 states: open, delivered, at-risk, or ambiguous. At-risk means a promised date passed with no matching work. The engine ships in Atlas today behind the paid tier, and Atlas Pro is not yet on sale.
Promised on a call, lost by Friday
The commitments you make on a call live in 2 places, a transcript and your memory, and neither is near the repository.
Commitment Tracking treats each commitment as a claim and the codebase as reality. The gap is the point: a promise with matching code is delivered, and a promise with no matching work and a passed date is at risk and worth your attention now.
Evidence from code and commits
Atlas checks each commitment through a 2-tier deterministic cascade before it spends a single token.
Tier 1 looks for an explicit target actually present in the codebase index. Tier 2 does a keyword match against the index and against recent commit messages and paths. Both are deterministic, so the routine pass costs nothing to run. Every verdict carries the evidence behind it rather than arriving as a bare judgement.
Semantic matching is opt-in, not automatic
A promise fulfilled in different words than it was made in needs 1 extra step, and that step is deliberately manual.
The deeper LLM judgement that catches paraphrased delivery runs only when you ask for it with --deep. The background pass stays deterministic and token-free, so a low-confidence match is flagged rather than assumed. You decide what counts as delivered, and a commitment you confirm as delivered never silently downgrades on a later check.
Availability
Commitment Tracking is 1 of the 8 record engines in Atlas, and it sits behind the paid tier.
The engine is built and running in the product, gated to Atlas Pro. Pro has no checkout yet, so it is not something you can buy today. It also depends on Intake for its input, since the commitments come from call transcripts. Shipped, the automated work log, is the one record engine available free right now.
Frequently asked questions
- Is Atlas commitment tracking available now?
- The engine is built and running in Atlas, but it sits behind the paid tier and Atlas Pro is not yet on sale, so you cannot purchase access today.
- Where do the commitments come from?
- From the commitments Atlas Intake extracts from your call transcripts. Email as a second source is designed but not built.
- How does it know something was delivered?
- A two-tier deterministic cascade: an explicit-target check against the codebase index, then a keyword match against the index and recent commit messages and paths. Each verdict carries its evidence.
- What does at risk mean?
- A commitment whose promised date has passed with no matching code or commits, so it needs your attention now rather than at the next client call.
- Does it use an LLM on every check?
- No. The routine pass is deterministic and spends no tokens. The LLM judgement that catches paraphrased delivery runs only when you pass --deep.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas with GPT-5.3 Chat: The small_model Slot Pick for 2026
GPT-5.3 Chat is a non reasoning 128K model at $1.75 per Mtok input, $14 per Mtok output. Why it belongs in the Atlas small_model slot, not the main build loop.
Atlas for Svelte in 2026
Adopt Atlas, the terminal-native AI coding agent, for Svelte and SvelteKit projects in 2026. Enhance development with intelligent code search, secure local embeddings, and guided code generation.
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 Llama 3.1 8B (local via Ollama): The 4.9GB Baseline for 2026
Llama 3.1 8B (local via Ollama) in Atlas for 2026: a 4.9GB pull that fits 8GB of VRAM, Free (self-hosted), and honest limits on a general-purpose 8B model.
Atlas with Gemini Flash Latest: The Rolling Alias Explained (2026)
gemini-flash-latest in Atlas: a rolling alias, not a pinned checkpoint. $0.3 per Mtok input, $2.5 per Mtok output, 1,048,576 token context, and no reproducibility.
Atlas with Magistral Medium: Multi-Hop Root-Cause Debugging in 2026
Magistral Medium reasons across a 128,000 token window at $2.00 / 1M input tokens and $5.00 / 1M output tokens. Atlas setup, the 16,384 token output cap, tradeoffs.
Atlas with Qwen3-Coder 480B-A35B Instruct: The Open Frontier Coder in 2026
Qwen3-Coder 480B-A35B Instruct in Atlas: 480B total parameters, 35B active per token, 262,144 tokens of context, $1.50 per Mtok in and $7.50 per Mtok out.
Atlas with DeepSeek V3.2 (open weights): Sparse Attention at $0.38 Output, 2026
Run Atlas on DeepSeek V3.2 (open weights) in 2026. DeepSeek Sparse Attention gives 160K tokens (DeepInfra) at $0.26 per Mtok in and $0.38 per Mtok out.