# Qredence Documentation > Qredence documentation for AI agent systems, including Fleet RLM, Fleet Pi, GEPA Omni, Fleet Reasoner, Qredence Skills, and plugins. ## Docs - [Introduction to fleet-rlm](https://docs.qredence.ai/fleet-rlm/introduction.md): fleet-rlm is a FastAPI SSE backend for recursive language-model turns on dspy.RLM, with a maintained terminal client (pi-tui) and Daytona-backed workspace-scoped volumes. - [fleet-rlm quickstart](https://docs.qredence.ai/fleet-rlm/quickstart.md): Install fleet-rlm with uv, pick a Daytona profile, initialize the database, and stream a Turn through pi-tui or the SSE API. - [Install fleet-rlm](https://docs.qredence.ai/fleet-rlm/installation.md): Install fleet-rlm from source with uv, provision the Daytona snapshot, initialize Postgres to the Alembic head, and pin the pi-tui workspace. - [fleet-rlm core concepts](https://docs.qredence.ai/fleet-rlm/concepts/overview.md): How fleet-rlm coordinates one native dspy.RLM per Turn over Daytona Sandboxes, streams typed Runtime Events over SSE, and persists committed Turn history. - [fleet-rlm architecture](https://docs.qredence.ai/fleet-rlm/concepts/architecture.md): How fleet-rlm layers a FastAPI SSE transport, a turn coordinator, one native dspy.RLM per run, and a Daytona interpreter into a single certified pipeline. - [Agent model](https://docs.qredence.ai/fleet-rlm/concepts/agent-model.md): How fleet-rlm runs one fresh native dspy.RLM per Turn, composes bounded Signatures, and selects the cheapest-sufficient step on the Root delegation ladder. - [Recursive RLM delegation](https://docs.qredence.ai/fleet-rlm/concepts/recursive-rlm.md): How fleet-rlm runs a Turn as one fresh native dspy.RLM, walks the cheapest-sufficient ladder, and isolates a single native child level under bounded budgets. - [Daytona runtime](https://docs.qredence.ai/fleet-rlm/concepts/daytona-runtime.md): How fleet-rlm uses one process-owned AsyncDaytona client, per-Turn Interpreter Leases, and scoped Volume mounts to run every Turn on Daytona. - [Sessions and persistence](https://docs.qredence.ai/fleet-rlm/concepts/sessions-persistence.md): Three persistence layers in Fleet — Alembic-managed Postgres for committed history, the Daytona Volume Scope for durable workspace state, and volatile interpreter context. - [Observability](https://docs.qredence.ai/fleet-rlm/concepts/observability.md): How fleet-rlm projects Runtime Events to bounded MLflow spans, streams typed SSE chunks, records cancellation tombstones, and exposes health and doctor probes. - [DSPy in fleet-rlm](https://docs.qredence.ai/fleet-rlm/guides/dspy-integration.md): How fleet-rlm composes native dspy.RLM, Signatures, instruction fragments, and DSPy sandbox-serializable inputs for one Turn against Daytona. - [Deploy the fleet-rlm backend](https://docs.qredence.ai/fleet-rlm/guides/deployment.md): Run fleet-rlm in production as a single FastAPI SSE process against Postgres and Daytona, with policy-driven profiles and no built-in caller authentication. - [fleet-rlm troubleshooting](https://docs.qredence.ai/fleet-rlm/guides/troubleshooting.md): Fixes for common fleet-rlm issues: policy selection, Daytona credentials, database preflight failures, SSE stream errors, and MLflow tracing. - [fleet-rlm CLI reference: fleet, fleet-rlm](https://docs.qredence.ai/fleet-rlm/reference/cli.md): Reference for the fleet and fleet-rlm commands: supervised cli, backend-only serve-api, Daytona doctor, and snapshot management. - [fleet-rlm HTTP and SSE API reference](https://docs.qredence.ai/fleet-rlm/reference/http-api.md): Reference for the fleet-rlm FastAPI surface: Sessions, Turns, Attachments, Artifacts, Files, Volume, Skills, Runs, and the Turn SSE stream contract. - [fleet-rlm Python entry points](https://docs.qredence.ai/fleet-rlm/reference/python-api.md): Reference for the maintained fleet-rlm Python entry points: FastAPI factory, CLI, standalone backend command, and Daytona snapshot builder. - [fleet-rlm configuration reference](https://docs.qredence.ai/fleet-rlm/reference/configuration.md): Reference for fleet-rlm runtime policy: config/fleet.toml profiles, FLEET_* environment inputs, MLflow tracing, and recursive RLM bounds. - [Introduction to Fleet Pi](https://docs.qredence.ai/fleet-pi/introduction.md): Fleet Pi is a local-first browser workspace for Pi-powered coding agents with reviewable plans, memory, skills, and repo-scoped tools committed to Git. - [Fleet Pi quickstart](https://docs.qredence.ai/fleet-pi/quickstart.md): Install Fleet Pi with pnpm, configure a Google Gemini API key, and launch the local Pi-backed chat workspace in under five minutes. - [Fleet Pi configuration reference](https://docs.qredence.ai/fleet-pi/configuration.md): Reference for every Fleet Pi environment variable — LLM provider keys, Pi runtime, logging, auth, and the Neon Postgres chat session mirror. - [Chat modes](https://docs.qredence.ai/fleet-pi/chat-modes.md): How Fleet Pi's Agent, Plan, and Harness chat modes change the tool allowlist passed to the Pi runtime, including read-only Plan mode and per-mode tool tables. - [Generative UI with inline OpenUI Lang components](https://docs.qredence.ai/fleet-pi/generative-ui.md): Stream generative UI inline in Fleet Pi chat using OpenUI Lang — cards, badges, charts, tables, progress bars, and safe conversational buttons. - [Codex worktree setup for Fleet Pi](https://docs.qredence.ai/fleet-pi/codex.md): Bootstrap Codex worktree threads against Fleet Pi using the shared .codex local environment, the workspace-bootstrap.zsh script, and advanced multi-agent flow. - [Agent workspace](https://docs.qredence.ai/fleet-pi/agent-workspace.md): Tour of agent-workspace/ — Fleet Pi's repo-local home for durable agent memory, execution plans, skills, evals, artifacts, and installed Pi resources. - [Adaptive workspace contract](https://docs.qredence.ai/fleet-pi/adaptive-workspace.md): Fleet Pi's canonical workspace contract — the manifest, section families, kinds, durable file boundary, and non-canonical projection layer for indexes. - [Fleet Pi architecture](https://docs.qredence.ai/fleet-pi/architecture.md): Runtime topology of Fleet Pi — React 19 browser client, TanStack Start Nitro backend, repo-local agent workspace, the Pi runtime, and the configured model provider. - [Fleet Pi project structure](https://docs.qredence.ai/fleet-pi/project-structure.md): Tour of the Fleet Pi monorepo — apps/web TanStack Start routes, agent-workspace layout, Pi runtime modules, key dependencies, and the runtime data flow. - [Fleet Pi runtime SDK integration seams](https://docs.qredence.ai/fleet-pi/runtime-sdk-integration.md): Implementation reference for Fleet Pi's Pi runtime seams — session construction, NDJSON streaming, queueing, and the SessionManager surfaces for adaptive work. - [Fleet Pi operational runbooks](https://docs.qredence.ai/fleet-pi/runbooks.md): Incident response and troubleshooting runbooks for Fleet Pi — LLM provider circuit-breaker recovery, chat-stream triage, and Neon Postgres mirror operations. - [HTTP API reference](https://docs.qredence.ai/fleet-pi/api-reference.md): Reference for the Fleet Pi local web app HTTP API — the /api/chat NDJSON event stream, request and response schemas, and supporting workspace endpoints. - [Introduction to GEPA Omni](https://docs.qredence.ai/gepa-omni/introduction.md): GEPA Omni packages the GEPA Anything optimization stack as an Agent Plugins 1.0 plugin, running four engines behind one evaluator contract to improve any scorable text artifact. - [GEPA Omni quickstart](https://docs.qredence.ai/gepa-omni/quickstart.md): Install GEPA Omni into Codex, configure the OpenAI-compatible endpoint, run preflight, and launch your first Omni optimization. - [Omni workflow](https://docs.qredence.ai/gepa-omni/omni-workflow.md): How the Omni two-phase workflow orchestrates GEPA, AutoResearch, and Meta-Harness, partitions its budget, and hands the winning Phase 1 candidate to a fresh continuation. - [Engines and backends](https://docs.qredence.ai/gepa-omni/engines.md): The four GEPA Omni engines — GEPA, AutoResearch, Meta-Harness, and Best-of-N — plus how the shared OpenAI-compatible Chat Completions runtime powers all of them. - [Writing evaluators](https://docs.qredence.ai/gepa-omni/writing-evaluators.md): Turn a scoring function into feedback-rich Actionable Side Information: judge-based rubrics, batching, stochastic averaging, multi-objective Pareto, and reward-hacking-resistant gates. - [API reference](https://docs.qredence.ai/gepa-omni/api-reference.md): Two intentionally distinct API layers: the published gepa==0.1.4 optimize_anything engine, and the plugin wrapper run_optimization / run_omni for the native AutoResearch, Meta-Harness, Best-of-N, and Omni workflows. - [Gotchas and pitfalls](https://docs.qredence.ai/gepa-omni/gotchas.md): Read before any real run: reward hacking, selection bias, stochastic defaults, budget sizing, stop conditions, nested PyPI config, saturated signals, and evaluator exception handling. - [Introduction to Qredence Skills](https://docs.qredence.ai/skills/introduction.md): Qredence Skills is a curated catalogue of Figma design and product skills for AI agents, installed with skills.sh and invoked by name or task description. - [Qredence Skills quickstart](https://docs.qredence.ai/skills/quickstart.md): Install the Qredence Figma agent skills into your agent with skills.sh, list what shipped, and invoke a skill by name or task description. - [Skills catalogue](https://docs.qredence.ai/skills/catalogue.md): The 62 Figma agent skills in Qredence Skills, grouped by job: accessibility, design system, components and code, layout, prototyping, research, content, and delivery. - [Authoring skills](https://docs.qredence.ai/skills/authoring.md): Anatomy of a SKILL.md, the required frontmatter and workflow sections, and the validation and release workflow for adding a new skill to the Qredence catalogue. - [Introduction to Fleet Reasoner](https://docs.qredence.ai/fleet-reasoner/introduction.md): Fleet Reasoner is Qlaw's reasoning engine re-implemented on DSPy 3.3.0 — typed lenses over a pydantic graph, one compilable ReasoningEngine module, and a ReActV2 chat agent with lenses-as-tools. - [Fleet Reasoner quickstart](https://docs.qredence.ai/fleet-reasoner/quickstart.md): Install Fleet Reasoner with uv, configure the OpenAI-compatible endpoint, run the FastAPI + SSE server, and open the tldraw web frontend against your first Qlaw graph. - [Architecture](https://docs.qredence.ai/fleet-reasoner/architecture.md): Qlaw on DSPy 3.3.0 in three tiers: typed lenses over a pydantic graph, one compilable ReasoningEngine module, and a ReActV2 chat agent with lenses as tools. - [Lenses](https://docs.qredence.ai/fleet-reasoner/lenses.md): Tier 1 of Fleet Reasoner: one optimizable dspy.Module per agent, Refine-validated with deterministic reward functions where the app's constraints matter. - [Engine and router](https://docs.qredence.ai/fleet-reasoner/engine.md): Tier 2 of Fleet Reasoner: the whole reasoning cycle composed as one dspy.Module — Selection, Routing, Invocation, Refine-validated Expansion, Critic gate, and recursion. - [Chat and tools](https://docs.qredence.ai/fleet-reasoner/chat.md): Tier 3 of Fleet Reasoner: the multi-turn Qlaw chat agent — dspy.ReActV2 with lenses-as-tools, stateful sessions, and graph mutation that still goes through the pure apply() boundary. - [Optimization and evaluation](https://docs.qredence.ai/fleet-reasoner/optimization.md): Compile Fleet Reasoner's lenses and engine with MIPROv2, BootstrapFewShot, or bounded GEPA; evaluate with dspy.Evaluate against the six per-lens metrics and full-session engine trainsets. - [API and streaming](https://docs.qredence.ai/fleet-reasoner/api.md): Fleet Reasoner's FastAPI service — /seed, /engine, /chat/stream, and /config — plus the SSE frame shapes, session model, and normalized DSPy 3.3.0 error mapping. - [Gotchas](https://docs.qredence.ai/fleet-reasoner/gotchas.md): DSPy 3.3.0 pitfalls that will cost you time: retired assertions, model-string form, optimizer API shape, module conventions, evaluation contract, tools and retrieval, streaming, and caching. - [Fleet Prime Agent](https://docs.qredence.ai/fleet-prime-agent/introduction.md): A self-improving coding agent with persistent sessions, real subagents, and a streaming web UI. Run it from the CLI, a TUI, a background daemon, or the Qredence web app. - [Quickstart](https://docs.qredence.ai/fleet-prime-agent/quickstart.md): Install Fleet Prime Agent, sign in a model provider, and run your first agent turn. - [Install](https://docs.qredence.ai/fleet-prime-agent/install.md): System requirements and install steps for the Fleet Prime Agent workspace, Python kernel, and Qredence web UI. - [Architecture](https://docs.qredence.ai/fleet-prime-agent/concepts/architecture.md): How the agent runtime, AI layer, coding agent, and web surface fit together. - [Sessions](https://docs.qredence.ai/fleet-prime-agent/concepts/sessions.md): How Fleet Prime Agent keeps state between turns: persistent IPython kernels, transcripts on disk, branching, and compaction. - [Subagents (RLM)](https://docs.qredence.ai/fleet-prime-agent/concepts/rlm-subagents.md): How rlm(...) spawns real child agents from inside an IPython cell, and how the runtime schedules them. - [Refinement](https://docs.qredence.ai/fleet-prime-agent/concepts/refinement.md): How /refine improves the agent from its own trajectory without rewriting the base prompt. - [Streaming protocol](https://docs.qredence.ai/fleet-prime-agent/concepts/streaming-protocol.md): The event stream Fleet Prime Agent emits: AgentSessionEvent internally, ChatStreamEvent NDJSON to the web. - [Tools, skills, and extensions](https://docs.qredence.ai/fleet-prime-agent/concepts/tools-and-extensions.md): The three ways to give Fleet Prime Agent new capabilities: built-in tools, skills you write, and extension packages. - [Streaming chat](https://docs.qredence.ai/fleet-prime-agent/guides/streaming-chat.md): How agent turns stream to your client, and how to consume the stream in the web UI, CLI, or your own code. - [Tool cards](https://docs.qredence.ai/fleet-prime-agent/guides/tool-cards.md): Every tool call renders as a card. Here's what each type looks like and how to read them. - [Providers and models](https://docs.qredence.ai/fleet-prime-agent/guides/providers-and-models.md): Pick a provider, sign in, and choose a model. Fleet Prime Agent ships with 19 providers and a registry that knows each model's cost, context, and thinking levels. - [Sessions and branching](https://docs.qredence.ai/fleet-prime-agent/guides/sessions-and-branching.md): Save, resume, fork, and search your Fleet Prime Agent sessions. - [Slash commands](https://docs.qredence.ai/fleet-prime-agent/guides/slash-commands.md): Built-in commands: /refine, /login, /tokens, /status, /doctor, /update, /shutdown. - [Skills](https://docs.qredence.ai/fleet-prime-agent/guides/skills.md): Turn a recurring workflow into a personal or project skill the agent loads automatically. - [Extensions](https://docs.qredence.ai/fleet-prime-agent/guides/extensions.md): Ship your own package that adds tools, slash commands, shortcuts, or UI to Fleet Prime Agent. - [MCP integration](https://docs.qredence.ai/fleet-prime-agent/guides/mcp.md): Register Model Context Protocol servers so their tools show up alongside Fleet Prime Agent's built-ins. - [OAuth sign-in](https://docs.qredence.ai/fleet-prime-agent/guides/oauth.md): Sign in to providers over OAuth, and understand how tokens are refreshed during long-running sessions. - [Refinement (/refine)](https://docs.qredence.ai/fleet-prime-agent/guides/refinement.md): Run /refine to improve the agent from its trajectory, and roll back if a refinement makes things worse. - [Monitoring and diagnostics](https://docs.qredence.ai/fleet-prime-agent/guides/monitoring.md): Check the daemon, the kernel, and running sessions. Fleet Prime Agent is local-first — no dashboards, just commands and logs. - [CLI](https://docs.qredence.ai/fleet-prime-agent/interfaces/cli.md): Run Fleet Prime Agent from the command line. Modes, flags, and examples. - [TUI](https://docs.qredence.ai/fleet-prime-agent/interfaces/tui.md): The interactive terminal client for Fleet Prime Agent. - [Daemon](https://docs.qredence.ai/fleet-prime-agent/interfaces/daemon.md): Run Fleet Prime Agent as a background process that CLI, TUI, and web all attach to. - [Web app (Qredence UI)](https://docs.qredence.ai/fleet-prime-agent/interfaces/web.md): The Qredence web UI for Fleet Prime Agent: chat, tool cards, session list, model picker. - [Configuration](https://docs.qredence.ai/fleet-prime-agent/reference/configuration.md): Config directory layout, environment variables, and auth storage. - [Security model](https://docs.qredence.ai/fleet-prime-agent/reference/security.md): Trust boundaries, credential storage, and sandbox options for Fleet Prime Agent. - [Daemon protocol](https://docs.qredence.ai/fleet-prime-agent/reference/daemon-protocol.md): Wire protocol between the Fleet Prime Agent daemon and its clients. Versions, capabilities, and compatibility rules. - [Web API](https://docs.qredence.ai/fleet-prime-agent/reference/web-api.md): HTTP routes exposed by the Qredence web app and the events they emit. - [Glossary](https://docs.qredence.ai/fleet-prime-agent/reference/glossary.md): Fleet Prime Agent terminology. - [Background](https://docs.qredence.ai/fleet-prime-agent/background.md): Project history, design rationale, and major rewrites. See Lore for timeline and Design decisions for architectural choices. Key events:... - [Design decisions](https://docs.qredence.ai/fleet-prime-agent/background--design-decisions.md): Key decisions: AgentConnection seam (packages/agent/src/types.ts); InteractiveModeLocalSessionHost deletion (docs/adr/0001-kill-interactivemode-localsessionhost.md); web workspace... - [Cleanup opportunities](https://docs.qredence.ai/fleet-prime-agent/cleanup-opportunities.md): Source TODO markers: 4 in packages/coding-agent/src/core/ (persistent kernel reconsideration, abort plumb, best-effort exit). Test FIXMEs: packages/ai/test/tokens.test.ts,... - [By the numbers](https://docs.qredence.ai/fleet-prime-agent/by-the-numbers.md): Data collected: 2026-08-17. - [Fun facts](https://docs.qredence.ai/fleet-prime-agent/fun-facts.md): Fleet Prime Agent — Fun facts. - [Lore](https://docs.qredence.ai/fleet-prime-agent/lore.md): Prime Agent is a hard fork of PrimeIntellect-ai/prime-agent (lineage pi-mono by Mario Zechner). It delivers persistent agent sessions, a Qredence web workspace (web/app, design,... - [Maintainers](https://docs.qredence.ai/fleet-prime-agent/maintainers.md): Upstream lineage: Mario Zechner (pi-mono), Prime Intellect team (PrimeIntellect-ai/prime-agent). Recent work (last ~50 commits): Zachary BENSALEM (72 commits total). Web workspace... - [How to contribute](https://docs.qredence.ai/fleet-prime-agent/how-to-contribute.md): Read AGENTS.md, CONTRIBUTING.md, and CODE_OF_CONDUCT.md. Fork the repo, work on a feature branch, and open a PR into main. Add pkg:* labels (pkg:agent, pkg:ai, pkg:coding-agent,... - [Development workflow](https://docs.qredence.ai/fleet-prime-agent/how-to-contribute--development-workflow.md): Branch from main, work locally. Install: npm ci at repo root, pnpm install --dir web for web. Never run pnpm install at root (rewrites node_modules to pnpm layout). After code changes:... - [Testing](https://docs.qredence.ai/fleet-prime-agent/how-to-contribute--testing.md): Run npm run check for format/lint/type-check (not tests). Focused tests via vitest CLI from package root: `cd packages/coding-agent && npx tsx ../../node_modules/vitest/dist/cli.js --run... - [Debugging](https://docs.qredence.ai/fleet-prime-agent/how-to-contribute--debugging.md): Inspect daemon state via prime-agent status. Capture TUI logs (PI_TUI_WRITE_LOG creates tui--.log). Use tmux for controlled interactive tests (see AGENTS.md tmux... - [Patterns and conventions](https://docs.qredence.ai/fleet-prime-agent/how-to-contribute--patterns-and-conventions.md): Fleet Prime Agent — Patterns and conventions. - [Tooling](https://docs.qredence.ai/fleet-prime-agent/how-to-contribute--tooling.md): Fleet Prime Agent — Tooling. - [Packages](https://docs.qredence.ai/fleet-prime-agent/packages.md): Four npm workspace packages (packages/ai, agent, coding-agent, tui) plus prime-agent-runtime (Python). Four pnpm workspace packages (web/app, design, protocol, server). - [Agent package (`packages/agent`)](https://docs.qredence.ai/fleet-prime-agent/packages--agent.md): Core agent session runtime (packages/agent/src/agent.ts, agent-loop.ts, types.ts). Defines AgentEvent, AgentMessage, AgentTool, AgentContext, AgentState. The agent loop... - [AI package (`packages/ai`)](https://docs.qredence.ai/fleet-prime-agent/packages--ai.md): LLM provider abstraction (packages/ai/src/types.ts), model registry (models.generated.ts, 578 KB), stream parsing (packages/ai/src/utils/event-stream.ts), provider implementations... - [AI models (`packages/ai/src/models.generated.ts`)](https://docs.qredence.ai/fleet-prime-agent/packages--ai--models.md): Generated registry (578 KB) mapping KnownProvider to model definitions (Model interface in packages/ai/src/types.ts). Each model defines: id, name, api, provider, baseUrl,... - [AI providers (`packages/ai/src/providers/`)](https://docs.qredence.ai/fleet-prime-agent/packages--ai--providers.md): 19 implementations: openai-completions.ts, anthropic.ts, openai-responses.ts, openai-codex-responses.ts, amazon-bedrock.ts, google.ts, google-vertex.ts, google-shared.ts,... - [Coding agent package (`packages/coding-agent`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent.md): CLI SDK (packages/coding-agent/src/core/sdk/), session manager (packages/coding-agent/src/core/session-manager.ts — current session version 3, CURRENT_SESSION_VERSION), compaction... - [Session runtime (`packages/coding-agent/src/core/session-manager.ts`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--session-runtime.md): Manages session persistence (SessionHeader, SessionMessageEntry, NewSessionOptions), transcript recording, branch summaries, compaction entries. Key constants:... - [Coding agent tools (`packages/coding-agent/src/core/tools/`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--tools.md): Built-in tools: ipython (persistent Python kernel via IpythonKernelProvisioner), bash/shell, edit/file operations, grep, find. Extension runtime... - [CLI (`packages/coding-agent/src/cli/args.ts`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--cli.md): Argument parsing for text, json, rpc, daemon, acp modes. Key flags: --provider, --model, --api-key, --thinking (levels: off/minimal/low/medium/high/xhigh/max), --autonomous,... - [Daemon (`packages/coding-agent/src/modes/daemon/`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--daemon.md): Daemon mode (DaemonAgentConnection, DaemonClient) separates agent runtime from terminal/web clients. Protocol versions (DAEMON_PROTOCOL_VERSION, DAEMON_SCHEMA_REVISION) govern wire... - [Interactive mode (`packages/coding-agent/src/modes/`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--interactive-mode.md): Interactive terminal agent (packages/tui is the TUI layer). The AgentConnection seam connects interactive clients to daemon-backed sessions. InteractiveModeLocalSessionHost was deleted... - [Skills (`packages/coding-agent/src/core/skills/`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--skills.md): Executable packages: MarkdownSkill, PythonSkill. Loaded via loadSkills, loadSkillsFromDir. Skills register as extensions (ExtensionRuntime, ExtensionAPI) and add... - [Extensions (`packages/coding-agent/src/core/extensions/`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--extensions.md): Extension runtime (ExtensionRuntime, ExtensionAPI) allows importable packages to extend agent behavior. Extensions register tools, commands, and shortcuts. The adapter (web/server) consumes... - [Refinement (`packages/coding-agent/src/core/refinement/`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--refinement.md): Self-improving harness (/refine) reviews trajectory and applies evidence-backed updates to supplemental harness state (packages/coding-agent/src/core/refinement/). Never rewrites the immutable... - [RLM runtime (`packages/coding-agent/src/core/rlm-runtime/`)](https://docs.qredence.ai/fleet-prime-agent/packages--coding-agent--rlm-runtime.md): Recursive language model (rlm(...)) spawns real child agents (SubagentRuntimeHost, RlmSubagentRuntime). Subagents run in parallel or background and return results programmatically. The RLM... - [Prime Agent runtime (`prime-agent-runtime/`)](https://docs.qredence.ai/fleet-prime-agent/packages--prime-agent-runtime.md): Python IPython kernel shim (requires Python >= 3.10, ipykernel, prime-agent-runtime installed). Provides IPythonKernelProvisioner, lazy-provisioned on first use. The rlm bridge uses Python... - [TUI (`packages/tui`)](https://docs.qredence.ai/fleet-prime-agent/packages--tui.md): Terminal interactive agent UI (packages/tui/src/). Differential rendering (surgical redraw, scrollback preservation, container change detection) introduced 2025-08-11. Key detection via... - [Web app (`web/app`)](https://docs.qredence.ai/fleet-prime-agent/packages--web-app.md): React 19 TanStack Start frontend (web/app/src/routes/__root.tsx, index.tsx, routes/api/*). Talks HTTP (NDJSON + SSE) only — never imports @earendil-works/* directly. Renders agent turns... - [Web design (`web/design`)](https://docs.qredence.ai/fleet-prime-agent/packages--web-design.md): Shared components and BEUI renderer (web/design/src/components/). Design tokens (token.ts), chart components (chart.tsx — includes sanitizeChartToken / sanitizeChartColor), input groups,... - [Web protocol (`web/protocol`)](https://docs.qredence.ai/fleet-prime-agent/packages--web-protocol.md): Zod wire contracts: chat-protocol.ts (ChatStreamEvent: start, delta, tool, thinking, plan, state, queue, compaction, retry, done, error); chat-protocol.zod.ts (zod... - [Web server (`web/server`)](https://docs.qredence.ai/fleet-prime-agent/packages--web-server.md): HTTP adapter (prime-bridge.ts, event-mapper.ts, packages/web-server/src/). Only web package that imports @earendil-works/* (via pnpm link:). Consumes createAgentSession,... - [Primitives](https://docs.qredence.ai/fleet-prime-agent/primitives.md): Base abstractions shared across agent, AI, and protocol layers. - [Agent message (`AgentMessage`)](https://docs.qredence.ai/fleet-prime-agent/primitives--agent-message.md): Union of LLM messages (Message: UserMessage, AssistantMessage, ToolResultMessage) and custom app messages (CustomAgentMessages). AgentMessage extends via declaration merging... - [Chat stream event (`ChatStreamEvent`)](https://docs.qredence.ai/fleet-prime-agent/primitives--chat-stream-event.md): Event protocol for streaming agent turns (packages/agent/src/types.ts). Types: start (partial AssistantMessage), text_start/delta/end, thinking_start/delta/end,... - [Deployment](https://docs.qredence.ai/fleet-prime-agent/deployment.md): Installation via install.sh (repo-hosted public installer, 2026-08-15). Requirements: Node >= 22.8.0, npm >= 11.10, pnpm >= 11, Python >= 3.10, Git. The web workspace (pnpm --dir web) is... - [Installer (`install.sh`)](https://docs.qredence.ai/fleet-prime-agent/deployment--installer.md): Repo-hosted public installer (install.sh, 2026-08-15). Sets up node, python runtime (prime-agent-runtime), links packages, and installs web dependencies (pnpm install --dir web). The... - [Release pipeline](https://docs.qredence.ai/fleet-prime-agent/deployment--release-pipeline.md): Lockstep versioning: npm run release:patch (fixes/additions) or release:minor (API breaks) updates all package versions, finalizes changelogs, commits, tags, and publishes. 7-day minimum... - [Introduction to Qredence Plugins](https://docs.qredence.ai/qredence-plugins/introduction.md): Qredence Plugins is a catalogue of dual-target plugins for Claude Code and OpenAI Codex — repo audits, harness optimization, research wikis, and orchestration. - [Qredence Plugins quickstart](https://docs.qredence.ai/qredence-plugins/quickstart.md): Set up the Qredence Plugins repository with uv and gh, then install and run your first plugin from the marketplace inside Claude Code or OpenAI Codex. - [Use Qredence plugins in Claude Code](https://docs.qredence.ai/qredence-plugins/claude.md): Add the Qredence Plugins marketplace to Claude Code, install a plugin from GitHub, and invoke its skills and agents from inside a Claude Code session. - [Use Qredence plugins in OpenAI Codex](https://docs.qredence.ai/qredence-plugins/codex.md): Add the Qredence Plugins Codex marketplace, install a plugin from .agents/plugins/marketplace.json, and run its skills inside an OpenAI Codex session. - [Author a Qredence plugin](https://docs.qredence.ai/qredence-plugins/authoring.md): Folder layout, Claude and Codex manifests, shared skills and agents, and the marketplace checklist for shipping a dual-target Qredence plugin bundle. - [harness-engineering plugin](https://docs.qredence.ai/qredence-plugins/plugins/harness-engineering.md): Audit repository legibility, scaffold durable repo docs, and add validation lanes and drift controls so coding agents work with stronger guardrails. - [meta-harness plugin](https://docs.qredence.ai/qredence-plugins/plugins/meta-harness.md): Scaffold a workspace, validate candidate harness.py files, and run outer-loop search over prompt, retrieval, parsing, and memory strategies for LLM tasks. - [rlm-wiki plugin](https://docs.qredence.ai/qredence-plugins/plugins/rlm-wiki.md): Wrap Fleet-RLM with a Daytona-backed markdown wiki to ingest URLs, files, PDFs, and transcripts, then query and maintain a durable agent knowledge base. - [symphony plugin](https://docs.qredence.ai/qredence-plugins/plugins/symphony.md): Reference Symphony service that polls Linear issues, creates per-issue workspaces, and runs Codex app-server sessions governed by a repository WORKFLOW.md file. - [development plugin](https://docs.qredence.ai/qredence-plugins/plugins/development.md): Skill-library plugin bundling cross-examine for plan stress-tests, data-viz-renderer for HTML and SVG charts, and skill-evaluator for scoring local skills. - [legal plugin](https://docs.qredence.ai/qredence-plugins/plugins/legal.md): Consumer-perspective contract review plugin with the tos-clause-scanner skill — audit Terms of Service, user agreements, and privacy policies for risky clauses. - [autoresearch-dspy plugin](https://docs.qredence.ai/qredence-plugins/plugins/autoresearch-dspy.md): Ratchet-style autonomous experiment loops built on DSPy 3.1.3, dspy.RLM, and confidence-aware classification scoring via the GEPA ConfidenceAdapter. - [Product changelog and release notes](https://docs.qredence.ai/changelog/overview.md): Weekly release notes for Fleet-RLM, Fleet Pi, and Qredence Plugins, covering new features, runtime updates, bug fixes, and security patches. - [README](https://docs.qredence.ai/components/README.md) - [MIGRATION GUIDE](https://docs.qredence.ai/MIGRATION_GUIDE.md) - [DOCUMENTATION SUMMARY](https://docs.qredence.ai/Reasoning-Kernel/DOCUMENTATION_SUMMARY.md) ## OpenAPI Specs - [openapi](https://docs.qredence.ai/openapi.json) ## Optional - [GitHub Organization](https://github.com/qredence) - [Discord Community](https://discord.gg/ebgy7gtZHK) - [Status Page](https://status.qredence.com)