Anna
Experimental source preview · CI was in progress at the initial source push (2026-09-22). Current main CI · reviewed-candidate CI. Local checks and independent review passed; use the linked current CI record for the live outcome.
Measured scope: 24 synthetic heldout cases, 22 model inputs and 2 hard prechecks. C is one
deepseek-v4-projudgment with thinking enabled/high; D isjev-1.13.0. Both match 22/22 labels (12 recommendations + 10 abstentions). Jev p50 is 85.23% lower; at this run's off-peak/cache-miss prices estimated cost is 94.71% lower (peak reference 97.35%; not invoices). Reported total tokens are 57.98% higher. This does not establish production accuracy, Worker completion or whole-workflow acceleration.
An AI companion for personal tasks, business workflows, and project collaboration.
Anna is a personal open-source project exploring how an AI agent can carry work from a conversation through to a reviewable result. The local-first desktop app brings tasks, connected business systems, and collaboration together in Home, Cowork, and Crew.
The aim is to keep the work understandable: what Anna is doing, which tools she can use, what needs your decision, and where the result came from.
Jev Crew · Experimental Source Preview · macOS arm64 · MIT License · CI
中文 · Explore Anna · Quick start · Current status · Codex pet · Development diary
Ask for an assignee suggestion, then decide
Crew can suggest a person or Worker for one unassigned task. Open the existing member picker, request a suggestion, inspect its source, and explicitly accept it or choose manually. A unique exact-role match uses a rule; Jev handles other eligible semantic choices and can abstain. Suggesting someone does not assign or run the task. This source preview adds explicit suggestions; see the changes, evidence, and limits. The RC2 source release remains a separate release record.
Measured Jev comparison
On 24 synthetic heldout cases, 22 were eligible for model judgment; two hard-precheck cases made no C/D call. Both models matched the frozen labels on 22/22: 12 correct recommendations and 10 correct abstentions, with zero wrong labels, API or format errors. Existing role rules matched 12/22 on the same subset.
| Measure (22 eligible cases) | DeepSeek, one judgment (C) | Jev production adapter (D) |
|---|---|---|
| Requested → returned model | deepseek-v4-pro → same |
jev-1.13.0 → same |
| Settings / provider requests | Thinking enabled, high; max 4096 / 22 | Typed Choice / 22 |
| Observed p50 / p95 | 2007.078 / 2692.501 ms | 296.349 / 421.889 ms |
| Reported input / output tokens | 4,887 / 2,873 | 11,233 / 1,026 |
| Estimated cost, peak/cache-miss reference | $0.017827920 | $0.000471786 |
Jev had 85.23% lower p50, 84.33% lower p95 and 97.35% lower reference estimated cost, with equal label agreement. Reported total tokens increased 57.98%; different tokenizers prevent an equivalent-compute interpretation. The cost figure uses DeepSeek peak/cache-miss prices for budget accounting, not a bill. At this run's off-peak rates with the same cache-miss assumption, estimated cost is 94.71% lower; cache hits may reduce DeepSeek cost further.
This is one small sequential comparison of model judgments, not production accuracy, browser end-to-end latency or measured speedup of the previous complete Crew/Host/OMP path (B). The 8 development cases are reported separately. See full results, denominators and method.
What you can explore
| Workspace | When to use it | What you can do |
|---|---|---|
| Home | Work through a personal task or create a reusable resource. | Bring files into a conversation, follow the task plan, inspect tool activity, and review documents or Prompt artifacts. Skill and Python Tool creation are also part of the interface. |
| Cowork | Work with information from a connected business system. | View business dashboards, ask the Hiker assistant about available data, and access existing reimbursement workflows. Connector permissions determine the available operations. |
| Crew | Coordinate a project with people and specialist Workers. | Organize tasks and dependencies on a project graph, add context in channels, assign work, and review or return versioned artifacts for rework. |
Home includes execution controls, history, files, and Trace inspection. Crew keeps project context, task discussions, artifacts, and review decisions connected, so a result can be followed back to the work that produced it.
Product walkthrough
Create, Cowork, and Crew in one loop. The walkthrough illustrates the interface; the Hiker dashboard uses synthetic data. Live model and connector validation is documented separately below.
A deliverable, its source task, the project channel, and the review decision stay connected in one workspace.
Meet Anna, your Codex companion
Anna's iris flower, ivory blouse, and purple skirt now come as a little desktop companion. Bring her into Codex to keep you company while you work.
A still capture and animation from the shared pet artwork. Includes 9 animation states and 16 look directions; requires desktop support for custom v2 pets.
Current status
Anna is in active development. The current source is intended for developers and contributors exploring the project.
| Area | Status |
|---|---|
| Jev Crew candidate | Optional single-task suggestions with explicit adoption. A real browser → Jev → assignment → SQLite readback is verified on synthetic data. The release record separates measured C/D judgments, accepted assignment behavior and remaining Worker/platform limits. |
| RC2 source | Fixes release validation around workdir identity/scope tests and CI evidence dependencies. Ordinary conversations continue through the shared Node Harness Host and Oh-my-Pi loop. See the RC2 record; the RC1 record preserves the earlier behavior and failures. |
| Earlier live validation | Home document generation, Prompt creation, Stop, and next-turn context; Crew Worker delivery, review/rework, and contextual Anna; Hiker dashboard reads and an Agent capability query. See the August 31–September 1 validation record for scope and remaining gates. |
| External business operations | The Hiker service used for that validation exposed read tools. Authorized write and read-back acceptance remain blocked on the service exposing the required capability. |
| Desktop distribution | Validation currently targets macOS arm64. The local application build is unsigned and unnotarized; Windows/Linux release acceptance remains open. |
| Application releases | v0.2.0 Developer Preview predates the current Harness execution path. The Codex pet has its own asset release. |
The historical RC2 validation used deterministic external transports for local identity, persistence, Gateway, and OMP execution. Its full live Provider/MCP gate remained blocked; the Jev judgment measurements above cover the new bounded suggestion scope. Full workbench scheduling, ask/answer, recovery, Memory, Sandbox, and Windows/Linux acceptance remain pending. Production readiness and benchmark results remain outside the current release claims. CI checks, interface demos, and live external-service runs provide different evidence; the current acceptance goals track those boundaries.
Quick start
Requirements: macOS arm64, Node.js ≥22.19.0, Python 3.12, and uv.
git clone https://github.com/Foxtailsss-Andy/Anna-Agent.git
cd Anna-Agent
npm ci
uv sync --locked --extra dev
ANNA_OMP_BUN_ARCHIVE_URL=https://github.com/oven-sh/bun/releases/download/bun-v1.3.14/bun-darwin-aarch64.zip npm run harness:omp:prepare
npm run desktop:run
To run Agent tasks, configure your model provider locally; business features also require their connector configuration. The current documented model setup uses DeepSeek through an OpenAI-compatible transport.
Follow DEVELOPMENT.md for conf