Scale
Custom held-memory capacity, reranking, and priority. Priced to your codebase after we size it with you.
Contact salesEstelle holds your repository in memory and cites every answer to the code it came from. The plan price buys how much stays held; model tokens go to your own provider.
One repo, grounded and gated.
10M
memory-tokens held
one repo
packageone size: 10M
—
$0no card needed
The self-correcting suite.
20M
memory-tokens held
5 package sizes, reranking on
Reranking · Self-correcting suite · Custom suites
$29per month
Teamper seat
One shared brain for the team.
Reranking · Self-correcting suite · Custom suites
Need a bigger pool? Memory Packs add +50M pooled for $15/mo each, and they stack.
Scale
Custom held-memory capacity, reranking, and priority. Priced to your codebase after we size it with you.
Contact salesCapacity and price above come directly from the live billing catalog. The feature descriptions are product policy; the monthly operation allowances and their limit behavior are stated below.
| operation | past the included allowance |
|---|---|
| reranks | then retrieval keeps first-pass order |
| repairs | then $0.02 per call with prepaid funds; otherwise it blocks |
| improvement runs | then $0.01 per call with prepaid funds; otherwise it blocks |
| repo sweeps | then $0.05 per call with prepaid funds; otherwise it blocks |
| paid research calls | then the paid research rung stops |
| monitor events | then $0.10 per 1,000 events |
| plan | reranks | repairs | improvement runs | repo sweeps | paid research calls | monitor events |
|---|---|---|---|---|---|---|
| Free | 1,000 | 3 | 5 | 20 | 25 | 1,000 |
| Ultra | 60,000 | 600 | 1,500 | 2,000 | 2,500 | 500,000 |
| Team | 120,000 | 3,000 | 6,000 | 10,000 | 7,500 | 2,000,000 |
“Included” means up to this many in one monthly period. It is not unlimited, and a zero or missing allowance is never read here as unlimited. BYOK model charges are separate and go directly to your provider.
Free keeps the parts you should not have to pay to trust: the deterministic gate, whole-repo memory, code-graph navigation and the session diary. The gate runs no model, so it costs us nothing to give away, and it is the one thing you should be able to check before you spend anything. Ultra adds capacity, grounded review on your pull requests, cross-encoder reranking and the self-correcting suite. Team shares Ultra across everyone.
| Capability | Free | Ultraeverything | Team |
|---|---|---|---|
| Foundation | |||
| Deterministic grounding gate on every proposed change. Free on every plan: it runs no model, so it costs us nothing, and it is the part you should not have to take on faithGate | |||
| Whole-repo memory, cited to file and lineMemory | |||
| Unlimited recall and queries, never meteredMemory | |||
| Builds in your real files, on your own keysCode | |||
| Install the plugin for always-on grounding, or connect the remote MCP URL | |||
| OpenAI-compatible API, one base URL and one model name | |||
| Code graph: go-to-definition, references, blast radius, import cyclesMemory | |||
| Always-on hooks: grounds every edit without the agent opting inGate | |||
| The %estelle notebook magic for Jupyter, Colab, Kaggle and Deepnote | |||
| Retrieval and review | |||
| Hybrid retrieval with rerankingMemory | |||
| Grounded review that lands on the pull requestReview | |||
| Completeness gate: names the symbols a change failed to touchGate | |||
| Session diary: what was decided, and where that decision livesMemory | |||
| Context survival: checkpoint, resume and govern past the windowGuardian | |||
| The self-correcting suites | |||
| Traces a production failure to the file and lineMonitor | |||
| Drafts the fix, proves it in a sandbox, opens a cited PRCode | |||
| Holds answer quality across a long sessionGuardian | |||
| Routes work across every provider key you ownOrchestra | |||
| Continuous pentest against a sandbox cloneSecurity | |||
| How it runs | |||
| Propose-only by default: every fix arrives as a reviewable PRCode | |||
| Opt-in auto-merge, off by default, only once the sandbox proves it greenCode | |||
| Protected surfaces stay human-gated: auth, billing, schema, deploy | |||
| Global kill switch and a signed audit line for every autonomous action | |||
| Cost routing across your keys: cheap work never escalates itselfOrchestra | |||
| In production, and across a team | |||
| Grounds the agents already running in your production, not just your editorAgent | |||
| Checks tool calls and claims against the repo as they happenAgent | |||
| Slack, GitHub and Linear as connected surfaces | |||
| Shared team memory with role-based access | |||
| Pooled capacity that stacks with Memory Packs | |||
Call the suite on Free and Estelle answers with a clean upgrade prompt, never a crash: reranking and the auto-repair and auto-find suite both stay an Ultra unlock. Auto-repair and proactive auto-find run today with a human on the autonomy dial, filing verified, tested fixes through your own tests. Fully autonomous self-healing in production is on the roadmap.
| add-on | what it adds | price |
|---|---|---|
| Memory Pack | 50M of persistent held capacity. Packs stack; on Team they grow the shared pool. | $15 / mo |
| Memory overflow | Held memory past your cap, charged as you go against prepaid funds. | $0.35 / 1M |
| Seats | Team is billed per seat over one shared pool. A seat buys a person access to the memory; it does not multiply the pool. | $99 / mo |
Set it to your numbers.
Yes. 10M of held memory, no card, and you bring your own provider key so there is nothing for us to bill through. It is a real tier, not a trial with an end date.
Prices and limits on this page are read live from the billing catalog, and the calculator is priced by the same service that bills. Model spend goes to your own provider at their rates. Metered usage is counted today and not yet billed. These are Estelle’s first published prices and none has changed since launch; if one changes, this line will say when, and which plans keep the rate they signed up on.