KAGENTXINTEL TERMINAL

JCode

Jcode is an open-source AI programming agent written in Rust, designed for the terminal, offering parallel agents, persistent sessions, and multiple execution entry points.

Who it is for: Users needing to run interactive or non-interactive programming tasks via the terminal (inferred from CODEX-research-jcode-005 functionality), users requiring parallel execution of multiple agents (inferred from CODEX-research-jcode-003 and CODEX-research-jcode-010 functionality), and technical users needing access via subscription OAuth, direct API provider, or local models (inferred from CODEX-research-jcode-006 functionality)

Positioning

Jcode is an open-source AI programming agent written in Rust, designed for the terminal, offering parallel agents, persistent sessions, and multiple execution entry points.

  • Product stage: Public iteration stage; multiple versions were released between 2026-08-10 and 2026-08-17.

Core features

  • Terminal Task and Session Management: Provides TUI, non-interactive execution, session recovery, persistent background server, and client connection entry points.
  • Parallel Agents and Swarm Collaboration: Supports running dozens of agents in parallel, managed by the server across the same repository; file changes notify other agents.
  • Background Task and Progress Management: Supports listing, viewing output, checking, canceling, and waiting for background tasks, with progress cards displayed in TUI.
  • Provider and Local Model Integration: Supports OAuth login, direct API provider, and login entry points for Ollama and LM Studio.

Target users

  • Users needing to run interactive or non-interactive programming tasks via the terminal (inferred from CODEX-research-jcode-005 functionality): Jcode is an open-source AI programming agent written in Rust, designed for the terminal, offering parallel agents, persistent sessions, and multiple execution entry points.
  • Users needing to run multiple agents in parallel (inferred from CODEX-research-jcode-003 and CODEX-research-jcode-010 functionality): Jcode is an open-source AI programming agent written in Rust, designed for the terminal, offering parallel agents, persistent sessions, and multiple execution entry points.
  • Technical users needing access via subscription OAuth, direct API provider, or local models (inferred from CODEX-research-jcode-006 functionality): Jcode is an open-source AI programming agent written in Rust, designed for the terminal, offering parallel agents, persistent sessions, and multiple execution entry points.

Pricing

  • Open-source software is free; there is also a Hosted inference subscription and usage-based billing. · The software itself is $0 with no functional barriers; Hosted inference is $10 per month, including $20 monthly inference credits, billed at 90% of provider API prices after credits are exhausted; monthly cap defaults to $100 and can be lowered to $0 or increased.: Lack of evidence on active users, paying users, revenue, gross margin, inference cost, conversion rate, retention, churn rate, and funding.

Verdict

Current evidence supports Jcode as an open-source AI programming agent for the terminal, with multiple entry points, parallel agents, persistent sessions, background tasks, and multi-provider integration capabilities. User samples, operational data, retention, funding, organizational market capabilities, and sustainable moats still need verification.

  • Strength: Terminal, non-interactive, session recovery, and background server/client entry points have direct evidence; parallel agents, repository collaboration, and background task management have direct evidence; the software is free and open source, and the subscription and usage-based billing rules for Hosted inference are disclosed.
  • Limitation: Public Issues contain multiple reports related to state, terminal, diff display, and swarm model selection; lack of attributable external user feedback samples makes it impossible to judge the representativeness of feedback; lack of evidence on active users, revenue, retention, funding, and enterprise scalability capabilities.

How it delivers value

  • Selecting Entry and Model: Users start tasks through TUI, non-interactive commands, OAuth, direct API, or local model entry points. → Users start tasks through TUI, non-interactive commands, OAuth, direct API, or local model entry points.
  • Running Tasks: Users run single tasks, persistent sessions, or parallel agents. → Users run single tasks, persistent sessions, or parallel agents.
  • Observation and Recovery: Users view background outputs, progress, and recover sessions after connection interruptions. → Users view background outputs, progress, and recover sessions after connection interruptions.
  • Collaboration and Iteration: Multiple agents in the same repository continue collaboration based on file changes. → Multiple agents in the same repository continue collaboration based on file changes.

Competition

In the officially disclosed DeepSWE v1.1 harness comparison, Jcode and Codex CLI both scored 75/113; this only supports the test result and is insufficient to form an overall competitive conclusion.

  • vs Codex CLI: Wins: Did not outperform in the officially disclosed DeepSWE v1.1 harness under the same conditions, both scoring 75/113. Loses: This result is a tie, and cannot be used to prove that Jcode is superior in overall capability.

Risks

  • Lack of evidence on active users, paying users, revenue, gross margin, inference cost, conversion rate, retention, churn rate, and funding.: Lack of evidence on active users, paying users, revenue, gross margin, inference cost, conversion rate, retention, churn rate, and funding.
  • Lack of evidence on enterprise permissions, audit, data isolation, compliance, and SLA.: Lack of evidence on enterprise permissions, audit, data isolation, compliance, and SLA.
  • Lack of attributable user samples from Reddit, X, Zhihu, WeChat Official Accounts, Xiaohongshu, and independent reviews.: Lack of attributable user samples from Reddit, X, Zhihu, WeChat Official Accounts, Xiaohongshu, and independent reviews.
  • Public Issues show reports related to headless state, terminal state, diff display, and swarm model selection.: Public Issues show reports related to headless state, terminal state, diff display, and swarm model selection.

Company

  • Stage: Public iteration stage; multiple versions were released between 2026-08-10 and 2026-08-17.
  • Funding: Not disclosed

User voices

  • Negative: GitHub Issues report a problem where headless jcode run completes successfully but is marked as crashed.
  • Negative: GitHub Issues report a problem where terminal exit clears OPOST/ONLCR, causing stair-step output.
  • Negative: GitHub Issues report a problem where inline diff preview lacks file path and merges multi-file patches.
  • Negative: GitHub Issues report a problem where swarm agent uses hard-coded or random models rather than inheriting the current session model.
  • Requested: Currently, there are no attributable user samples in Reddit, X, Zhihu, WeChat Official Accounts, Xiaohongshu, or independent reviews, so it cannot be generalized about user preferences, needs, or overall satisfaction based on these.
  • Requested: GitHub Issues reports indicate there are public issues, but this is insufficient to infer issue occurrence rates or all user experiences.

Sources

Updated 2026-08-18

Visit JCode